Intel Beast Lake Specs Leak: Pat KILLED this 6C/24T Monster!

[@MooresLawIsDead]

[SPON: FlexiSpot Anniversary Sale! Use “24AUG30” for $30 OFF E7 Desks w/ Combo Discounts on C7 Chars: bit.ly/46FUDNz (US), bit.ly/3YEHamX (CA)]

[@Nobe_Oddy]

NOW PAT IS MULLING OVER SELLING ALL THEIR FABS!!!!!!!! -- The American TAXPAYERS give Intel $8.5 BILLION to BUILD OUT THEIR FAB, AND NOW PAT WANTS TO SELL THEM!!!!! PAT NEEDS TO GO, LIKE YESTERDAY!!!!!!! HE'S A MORON!!!!!

[@ADB-zf5zr]

Tom, how would you react "if" this "senior employee" had directly lied to you and this whole thing is an elaborate ruse to throw off AMD, and the Royal Core project is complete and those employees that left, had completed their work and just moved on. Obviously this is incredibly unlikely, even more unlikely than a senior employee of a major company personally leak out,that the company that they work for has been kneecapped by it's CEO "Pat" by making several poor choices, and this one is so catastrophic for the company that he feels the need to do this.! . Intel is in really really deep doo-doo now.! I now expect Pat to be gone by October, this surely has to be the last straw, especially when the shareholders hear about this and ask for confirmation from Pat, Pat will be forced out and this will happen very quickly, especially with the "other problem" Intel is facing at the moment and that Intel's share price will now tank.!

[@Elam51]

Moore’s law is Dead?😂😂😂 So Why AMD Fanboys are So Angry?!😮😂

[@dontmatter4423]

So it seems that this could've been intel's response to 3d V-cache.. maybe they don't care about gaming anymore?

[@benjaminlynch9958]

The most fascinating aspect to this story is that someone high up inside Intel is willing (enthusiastic?) to go on the record - anonymously, of course - to throw Pat under the bus. That seems like a major story in and of itself…

[@lolmao500]

Well Pat is crashing intel to the ground because he wont listen to actually engineers who know wtf they are talking about. Typical of billion dollars companies, where the idiots in suits at the top trash the veterans who made the company what it is today. Classic american capitalism... at this point corporatism sounds a lot like communism and fascism, where one asshole at the top thinks hes better and smarter than everyone else... and crashes the whole thing.

[@Just4Games2011]

Honestly, this smells of either politics, or Pat really did mess us that bad this time.

[@albal156]

I honestly don't know if I rate Pat Gelsinger anymore. He seemed better than the Intel CEOs with MBA backgrounds but he doesn't seem to have the foresight of Lisa Su and Jensen Huang to know how the industry will develop and make a bet that then pays off for them.

[@kaseyboles30]

@@albal156 He's still better than having an MBA in charge of a tech company by being someone who actually understands the tech. But yeah on cpu's Dr. L. Su is much better, just needs to get gpu and marketing teams on that level. Jensen is a bit better on guessing the market than either of them. That said I think Pat and Lisa may suffer from optimism bordering on hubris occasionally. They think they can pull of miracles and push past the limits. This means sometimes they do. And sometimes they get dying I9's and it looks like no high end gpu's from team red this next round. Problem is Intel hides from it's mistakes and AMD claims more than it has.

[@RafitoOoO]

you can only miss the boat so much before your employees start questioning your leadership.

[@spoonikle]

Scaring away Jim is like burning the golden goose alive and then dumping every solid gold egg it laid into the Mariana trench.

[@mm-yt8sf]

a bold move! this is the sort of innovative thinking intel needs more of! 😀

[@jesh879]

@@mm-yt8sf ahahaha

[@theanglerfish]

Probably even deeper because leader of that project can potentially join AMD 😐😂and i wish he will just because that project

[@SteveBennet500]

they'll send pat to the bottom of mariana's trench to look for those eggs in a plastic submersible...

[@theanglerfish]

@@SteveBennet500 well said

[@dex6316]

Pat was supposed to be this rockstar CEO that takes Intel back to having unquestioned leadership, and here we have him gutting the projects that would take Intel there. It’s truly a shame because Royal Core looked absolutely amazing.

[@danipubg4946]

@@dex6316 I am kind of an AMD fanboy, but I don't want them to have a monopoly just like intel had. I want to see competition as that drives innovation and brings prices down too. From what's being told, it's really sad to see what Pat's done. This chip could've been a monster and potentially brought intel back at the top forcing AMD to innovate as well. Such a missed opportunity. But I guess we'll have to see what intel has to offer now in place of this and hope it's something good....

[@CruiseTom-rz8qu]

Intel is 5 years behind. That's not because of Pat. It all started Paul Otlenni and then Krzanich, both took their monopoly for granted and just took their innovation for granted. Intel is in the footsteps of Kodak and Nokia. There is not much Pat could do about it.

[@Pushing_Pixels]

He was the chosen one.

[@rstewart2702]

@@CruiseTom-rz8qu​​⁠​⁠​⁠yes this is the tragic story: Otellini and Krzanich left a horrible mess for Gelsinger…

[@djcetra]

@@CruiseTom-rz8qu lol how are they 5 years behind if they're still faster? 🤔

[@kyleheaser2385]

Not to defend Pat, but he might be worried about Intel even reaching 2028. I'm wondering if this isn't a tell that Intel is weaker than anyone imagines.

[@Kiwing827]

There's rumours that Intel are thinking of selling off the Fabs...it looks like Intel is just going through what AMD did before Ryzen. Intel is going to have very rough few years but the US government wouldn't let it go under since producing chips on US shores is an important part of national security especially with China constantly eyeing on Taiwan. We have already seen what happens when the Military Industrial Complex runs out of chips (e.g Russia - battle tanks sit partially built, salvaging old chips from consumer electronics)

[@PwntifexMaximus]

This is especially concerning if it's true that all the milestones where being achieved on time. This implies that Intel needs cash RIGHT NOW and having the design-team going for another two years sinply isn't possible.

[@richard.20000]

Take his "news" with giant rock of salt. More than one year he was saying that ARC GPUs were cancelled .... which was true only for high-end desktop, otherwise not true at all (and IMO whole GPU development was the best move Intel did in 20 years, just see how nVidia rises thanks to GPU computation). Basically Intel tries to do the same thing as Apple does: powerful iGPU for laptops. 3rd gen of Royal core being cancelled ... might be true. But it can mean that it was replaced by better 3rd gen Royal core design from other team. Year 2028 for 3rd Royal Core means that it was in 1st year of design (high level architecture). Maybe there were two competing designs and one of them won, the other lost. This is probably what happened because this is the common and healthy internal competition. It's highly unlikely that Intel would kill good product in favour of worse product.

[@PwntifexMaximus]

@@richard.20000 They would if it would cost them money long-term.

[@ericbagley3613]

​@richard.20000 Are we still pretending that Tom was wrong about Arc being "effectively cancelled"? Because I remember seeing Intel roadmap PowerPoints that showed they wanted to compete in every price point of desktop gpu.

[@TheDrunkeNNN]

Now that I think about it, this is exactly what my workloads in laptop, desktop and workstation look like. You either need extreme processing power for a handfull of threads or extreme multithreading. You never really need both at the same time. BIGlitte does not reflect that. For single threading, on a modern P+E chip, most of the transistors on the chip do nothing. Sounds like this architecture was supposed to change that. Think of caches that could be allocated to one thread or four, think of branch and data prediction resources that could work on branches of four threads or look deeper for one single thread. And finally an execution unit that could process either the four likeliest branches of a single thread or four different threads. Its a hell of a project to pull off but in the end, the architecture reflects what modern CPU loads look like. Jim Keller really is a mastermind.

[@aravindpallippara1577]

Yep, you could arguably have something like this in Big Little if you decide to turn off the one of the sides as performance requirements change which would have given massive power efficiency (computer per watt) at essentially killing most of your top end performance in multi threading. If they pulled off the dynamic core merge process they didn't need to take that hit in maximum performance at all.

[@e2rqey]

Based on the description, with Royal Core you could still do a combination of high single core + multicore workloads as well. Some of the cores could be in the multi-threaded config and some could be in the single threaded config. It could dynamically shift between single threaded perf and multithreaded perf. So that you have like 4 in single thread config and 2 in multi thread config, then alter that ratio throughout the workload depending on what it needs at that moment.

[@davidcadley1513]

@@aravindpallippara1577 that was the idea behind intel big little, but the problem is that for the vast majority of customers, across all markets, you are effectively forcing them to buy two CPU’s, one of which they don’t want. It sounds like gelsingers response has been : “just throw out the performance core then! We don’t make money on them anyway…”

[@highdefinist9697]

​@@e2rqey You could take the idea even further: Rendering a single frame within a modern video game is essentially a sequential mix of some single-core and some multi-core work loads. So, what if these cores can switch between both modes multiple times within a single frame? This would allow the game to perform very well when it is limited by some computations being done on its game thread or render thread, but still scale well in situations where a workload can be spread over a lot of threads, as the CPU could quickly reallocate its computational resources within each frame as needed.

[@Gen0cidePTB]

Everyone here is failing to see that this was the idea for Bulldozer and it was a massive flop.

[@danipubg4946]

I would absolutely love to see Jim Keller on MLID one day. Maybe you can try asking him?

[@bmqww223]

maybe there was one mlid episode you can search in his youtube... he talked on risc v and his own venture too

[@LouisDuran]

@TechPotato had Jim Keller on several times. I think you need to have a PhD to talk to Jim.

[@humbleeagle1736]

Especially if he could about his canceled work.

[@v3xx3r]

He's not allowed to talk about anything we care about.

[@danipubg4946]

@@humbleeagle1736 that's exactly what I'm interested in. Soooo sad to see such a potentially beast of a chip getting cancelled 😕

[@ClayinSWVA]

One of the worse parts for us server guys is VMware wants $135 per core per year and Microsoft wants about $310 per core every 5 years to run windows servers. So more cores on a chip helps but the costs scale with the number of cores now and gets really expensive over time.

[@andersjjensen]

Yup. Which is why AMD offers an 8 core Epyc CPU with 96MB L3 cache per core. Those are almost a cheat code in the database world. AMD does charge for them, but with the license savings on MSSQL Enterprise compared to Intel's offerings for the same performance, you're looking at a free CPU in 3 years. And that's before accounting for the difference in power costs.

[@tuckerhiggins4336]

Yeah that's the impression I got from the anonymous server engineer a couple podcasts ago. Glad to know Amd listens to their customers

[@bionicseaserpent]

aaaand that's why we don't bother with Windows on server! arbitrary price increases based on hardware oh how lovely

[@CommanderRiker0]

People still use Vmware and MS Server? Nearly every job I've seen lately is people migrating off of those platforms due to cost.

[@marsovac]

@@CommanderRiker0 enterprise support licenses for Linux also cost a ton, or you need to hire Linux experts in the company which is even worse - not every company is Meta or Amazon sized to be able to do that.

[@EdDale44135]

It is so easy for someone to praise to the sky a product that doesn’t exist. Think about how RDNA 3 was supposed to be this GPU powerhouse that was the “zen moment” in video cards and had Nvidia scared it would destroy them on the high end. Only for the released product to compete with the 4070 ti super on the high end. And this whole “focused on GPU instead of AI” seems to miss the point that Nvidia and AMD ‘s AI lineup is based on their GPU products, and that ARC is good at AI features in the Alchemist release. The enthusiast market is not the main driver of PC sales. It is business market - servers and laptops. I cannot recall a time where I entered an office and saw a desktop. Everything is a laptop, or a net PC based using laptop components.

[@highdefinist9697]

Yeah, it's hard to say... I believe it still would have made sense for Intel to create this CPU, in order to have the best product at least with this one relevant niche, but we also don't know about the development costs... Perhaps, they really would have been too high, relative to what Intel can afford, and Pat has chosen the overall safer route.

[@jdargui1537]

Unfortunately Pat Gelsinger didn't let Jim Keller Cook 😢

[@ZeroUm_]

Kinda crazy to say your top product is just a connector to GPU/AI workloads. Companies that make GPU and AI hardware would eat your market fast, since you're "just in the way, not doing much".

[@willgart1]

now we have to know why Pat did that. we don't know the full story. Did the risk is too high? Did some deadline were not reached? It's a change on the focus and priority to go all-in AI?

[@MooresLawIsDead]

No deadline issues. Pat just doesn't think you need high performance cores.

[@willgart1]

@@MooresLawIsDead if he really think that, and it's the only reason, he has a big problem! but I'm not surprised. I saw an analysis long time ago (more than 10 or 15years ago) describing the long term objective of Intel. and everything is confirming these objectives: moving Intel to israel first by moving fabs also moving the patent under the israel umbrela and closing the US based fabs the plan was more complicated, I don't remember all the details. but the analysis of all the relation between the manager and Israel was clear. and I'm sure if you investigate you'll find that a lot of managers and projects will be managed by Israeli people. funny, just found a 2021 news, Intel hired 10 Israeli people as VP (apparently the same day) Jim is not Israeli... who is in charge of the AI department at Intel? and other big projects?

[@HKNotch]

@@MooresLawIsDead Did you see the reuters article about Lip Bu Tan leaving the board because he wanted to rehire Jim and fire Pat? Just saying this maybe has a linkage.

[@lightward9487]

@@HKNotch Pat Gelsinger is the most legendary CPU engineer of all time with no guarantees when he knew Jim Keller bought Intel chip fabs but Jim left Intel before Pat returned!

[@mdavid2822]

I agree, the Intel story is huge and this is only one part of it. It is too easy to dogpile on Pat with the vultures who want to break apart Intel without knowing the full story based on anonymous leaks.

[@H8ts]

I saw many interviews with Jim Keller that time and he was always talking about a CPU being able to decouple parts of single thread workloads to multiple cores. That's probably what he meant. CPU's have extremely complex preemptive phase (analyzing code, translating instructions, branch prediction etc.) and in this phase, it could also be able to automatically identify parts of code that could be executed on multiple cores, return results and continue as a single thread until it finds another part that could be parallelized again. This is pretty much a holy grail of CPU design not yet put into practice, because this is not only hard, it is hard squared.

[@alexanderbelov6892]

Out-of-Order execution already does necessary instructions analysis for (in-)dependence, also CPU core has very long ROB for OOO execution. The only limit for parallel execution is the number of scheduled ports for one thread, and number of decoded instructions per cycle. If Royal/Beast Core can decode 32/64 instruction per cycle including 4-10 successfully predicted conditional branches inside it can have benefits of scheduling instructions to different rented units of several cores. But I suspect 4-10 "successfully predicted branches" in sequence per cycle look like a miracle. Also instructions retirement in ROB can become a bottleneck.

[@aladdin8623]

So quantum computing is a joke to you?

[@dslay04]

While Intel doesn't appear to be taking this path any longer, it's hard to believe that AMD is not on both their CPUs and GPUs. Navi 5x is supposed to be have dynamic resource allocation and Zen 5 has been widening everything and in some cases doubling resources. The fact that Zen 5c, Zen 5 classic, and mobile Zen 5 classic are the same front end but different FP execution, shows they're thinking about running the same code 3 different ways potentially on the same die.

[@systemBuilder]

Truthfully, the zen5 mistake (cross-CCD latency is too high) should be easy to fix. Either they can do it with microcode (and Lisa Su recently said it would be fixed with microcode in a Taiwan meeting), or they could just forward-port the 7000-series mesh fabric to zen5 to get their desired 20% uplift in single-threaded (gaming) applications. They already got the 20% uplift in massively parallel (linux) applications. They should not have given the CCD fabric design to the Zen2 team because it was screwed up on Zen2 and that the Zen2 used the screwed up fabric as a basis because it's all they understood.

[@saricubra2867]

"AMD is not on both their CPUs and GPUs." What the heck are you smoking?, Geforce is lightyears better than Radeon. Radeon is just way behind and a joke that no one on the professional industry takes seriously.

[@saricubra2867]

@@systemBuilder "Either they can do it with microcode (and Lisa Su recently said it would be fixed with microcode in a Taiwan meeting)" Same BS as 14900K and 13900K i9s being fixed through a microcode patch and the straight false and missleading marketing doesn't help.

[@henry_tsai]

@@saricubra2867 Tell me you only read news titles without telling me you only read titles.

[@systemBuilder]

@@saricubra2867 Except you didn't know that NVidia has to license AMD patents for the inventions of HBM and Shader 6, you really are an ignorant fanboy, aren't you?

[@jimtekkit]

Seriously, are all modern CEOs trained the same way? Seems like all they ever do is kill the most promising projects, and divert those resources into the projects doomed to failure.

[@LouisDuran]

You mistake is listening to this guy. He's so full of shit.

[@FrancisBurns]

MBA, short term revenue so I can get shit ton of bonuses by firing people, closing cool projects and divisions, long term f*ck you, douchebags.

[@tensai1shounen8]

It's much easier for CEOs to deliver short-term gains and make themselves look competent to investors for those big bonuses, and throw all the potential issues and problems under the rug for the next CEO to fix. By the time issues arise, the CEOs can conveniently "take responsibility" by retiring after taking that huge bonuses.

[@LouisDuran]

@@tensai1shounen8 Pat G. is in it for the long haul. Unless the board fires him. Intel CEOs are supposed to retire at 65. Gelsinger said he won't be able to see through his plan before he turns 65. So maybe they will fire him. But Pat does believe in the company and he believes he can fix the problems. He may take it a little bit too personally if he isn't successful

[@systemBuilder]

@@tensai1shounen8 I agree, the average CEO lasts about 3 years and so they just have to cancel everything else and dump everything into "The Chosen One" project to get a success and another chance to last another 3 years... Long term success? What's that?

[@Azureskies01]

Intel: We don't need powerful CPUs Intel in the next 5-10 years: We are filing for bankruptcy.

[@SunsetNova]

He’s joined the Dark Side and become corporate and lost his engineering identity 😔

[@RH-jj7ck]

naa. us goverment will not allow it . Strategic company for US

[@SunsetNova]

@@RH-jj7ck we want a competitive innovative Intel who can give AMD competition. Relying on the US Government for handouts to keep a failing company afloat isn’t exactly reassuring.

[@Azureskies01]

@@RH-jj7ck Right just like how the us government was so worried about saving AMD (they werent). All the US cares about is having top end chips, they don't give a shit by who. With thinking like this I'm surprised you don't have a youtube channel.

[@richardyao9012]

It sounds like Intel will be the next Fairchild semiconductor, a talent organ donor for the rest of Silicon Valley.

[@trousersnake1486]

Sounds like Intel needs to pull a starbucks and fire their ceo and put a new one in.

[@Banzeken]

Yeah, f*ck Pat. We need someone less obnoxious and more open to performance-per-core innovations. The stagnation has been painful to watch.

[@esra_erimez]

Jim Keller is a living legend.

[@Dante_S550_Turbo]

was* he's in a.i. start up hell now. Basically nft 2.0

[@esra_erimez]

@@Dante_S550_Turbo Tenstorrent has a very compelling RISC-V architecture

[@Zephyr-wb4vo]

How come? I don't know any of his other projects but people rate him very highly, so i would like to know

[@MikeKrasnenkov]

@@Zephyr-wb4vo he was involved in zen and apple silicon afaik

[@benc3825]

@@MikeKrasnenkovYes, Jim was, as he put it, the crazy uncle of Zen. He knows how to make legendary teams as we can clearly see with Zen, Tesla auto pilot, and Apple’s A4/A5 SOCs, which was the first Apple made silicon

[@hellcoreproductions]

Pat doesn't know how to play in competitive markets, look at his history.

[@lightward9487]

Pat Gelsinger is the most legendary CPU engineer of all time with no guarantees when he knew Jim Keller bought Intel chip fabs but Jim left Intel before Pat returned!

[@bradleyswaney6100]

The computer world needs Intel to succeed. I am starting to worry. A one company computer market will suck big time. I am and have been an AMD guy since my K6. But we need Intel, too. Back in those days, i always said AMD would someday dominate, but damn. Come on Intel! Pulling for ya.

[@kingkrrrraaaaaaaaaaaaaaaaa4527]

Maybe if Intel implodes Nvidia can buy an X86-64 licence and make CISC based CPUs themselves.

[@codycumby1170]

@@kingkrrrraaaaaaaaaaaaaaaaa4527 At triple market rates plus Nvidia premiums? No thank you.

[@mikebutterface8583]

Let’s not forget that AMD was really in deep doo doo until Ryzen saved their bacon. The reason Intel is importance is due to US fabs. They could still sell assets and be fine to exist. They could do like they’ve fine before and buy fab space with their engineered plans. They have options.

[@WhiteG60]

@@mikebutterface8583 Problem is that their fabs are pretty behind in tech. There's a reason they started farming work out to TSMC. Intel and AMD both took money from the US gov recently to build more on shore fabs, too.

[@mikebutterface8583]

@@WhiteG60 yes, I know they’ve had a lot of issues, but with the existing infrastructure I’m sure things can turn around.

[@TheEVEInspiration]

Betting a company on an ongoing hype is not very smart. in fact it is almost a guarantee to failure.

[@dmand1111]

Intels solution to Windows crap scheduler is one massive Pcore. Glorious

[@benjaminwlang]

I would have thought they'd want a killer CPU product. They really need a win. I'm really surprised to hear this.

[@jabronilifestyle]

Even if they had a supposed win, who is gonna buy Intel? If you didn't personally get burned by 13/14th gen, you definitely have seen someone on the internet complaining about it. My 13900k didn't even last 6 months. Now I'm stuck with a 1700 platform I don't trust. I'm not investing $2000 on an intel platform again.

[@damara2268]

How's a paper product a killer? In reality beast lake would have huge problems due to latency spikes when a core is splitting and windows scheduler absolutely destroying anything that's left. The more simple the CPU is, the better, because that way all apps can properly utilize it without needing special modifications for running on that exact one architecture.

[@xponen]

​@@jabronilifestyle they can rebrand, that's what company do when reputation took a dive. So, abandon the 20-year old i/core brand, make a new one with Royal Core 1.0.

[@jabronilifestyle]

@@xponen You're not getting it - Intel burned their supporter base with two (2) back to back generations of failure. Nobody wants to give them money now. Naming their CPUs something different isn't going to fix that. The lawsuits haven't even started yet and Intel already fired 15,000 people. People really don't get just how bad it is.

[@Real_MisterSir]

@@damara2268 You have zero insight nor basis to assume there would be "huge problems due to latency spikes when a core is splitting". That's just baseless cope to further an imaginary argument. We are already seeing Windows improving in its current scheduler and core task division being heavily utilized, and the systems will only improve further from here. Why would it be any different to having big and little cores like what we have now? In theory it should even be better, because it's not physically different cores, but an internal per-core division. At the end of the day, Intel needs a notable win. They need a product that can establish the firm ground under their brand identity. And whatever the fuck they're focusing on right now aint it. They're losing in server, they're losing in productivity, and they're now also losing in consumer -which has traditionally been the largest of the three. If they are betting on something other than an actual halo project, they have to hit it real big cus their brand image is declining faster than a hooker past the age of 40.

[@esra_erimez]

15:18 Jim Keller is not coming back. I'm waiting for Tenstorrent to go public. Intel has become Boeing

[@thepunish3r735]

Caz it takes years to unfold ? .. beast is canceled caz the production using the high N.A. euv machines will used and in a smaller chip ..beast lake was 10nm and they are going smaller end of 2027 ..his “research” was an ex employee that no longer works there and wasn’t a good fit ..also he was drinking 👏

[@myyoutubecommentschannel8784]

@@thepunish3r735 caz

[@Gen0cidePTB]

​@@thepunish3r735 You're right man, if someone is drinking it's totally not a symptom of a larger issue. 😂 Jim helped write the x86_64 specification. He also invented SMT, or Hyper-Transport. Jim was on the team for Ryzen, Apple Silicon, K8, K7 and Broadcom's Gbit switches. Indirectly it was his designs that influenced the Core architecture as Intel received them in an AMD lawsuit settlement a few years into P4's lifecycle. If you know anything about CPUs, the fact that a single man can be responsible for both the K7/K8 and the Core 2 Duo alone makes him godlike. He destroyed Intel with the K7/8 and then saved them with the Core architecture after their terrible P4. So now you know why I like him, why don't you?

[@m8x425]

me too, buddy

[@rolandlucmayon7480]

hes not going anywhere, he may just knocks other companies door and share his food

[@returningwhisper]

If you killed a project that had this much promise, you shouldn’t be in a leadership position.

[@Noisy_Cricket]

I wouldn’t say that until I knew where it was in development, but yeah, you’re probably right. Will be interesting to see if AMD can integrate these ideas (broadly) into a new product in a decade or so. They might have to pay Intel for the IP, but if they can get it working internally it would be worth it.

[@benjaminlynch9958]

That’s an overly simplistic view, and one that presumes client/desktop CPU’s should be a priority. Diverting resources from a 30% gross margin category to one that would yield 80%+ (as nVidia gets) if successful is not a terrible decision and one he will get LOTS of credit for if it works. Gamers may hate it, but gamers by and large aren’t shareholders. The fact is that client CPU’s are the lowest gross margin category Intel has, and if money is as tight as it is at Intel right now, it makes a lot of sense to cut your lowest gross margin products first.

[@anivicuno9473]

​@@benjaminlynch9958 Are you trying to say that Intel Arc, had they been second time lucky, would be able to command the same margins as Nvidia's products? Because that's a retarded take. Radeon, a company with a much better established brand and track record of not making ewaste can't command such a margin, what makes you think a new entrant like intel can? Your argument breaks down as soon as we rephrase it as "It's smart to divert resources from a 30% margin class we're established in to one where the second market leader makes below 50% margin and where we specifically have already failed once with the launch of DG1"

[@dex6316]

@@benjaminlynch9958why are you assuming that rentable units wouldn’t be a monster for data center. You can have extremely high IPC for applications that need it or extremely high thread throughput. There are also some server app licenses that charge per core so having fewer cores would actually save a good amount of money.

[@bryce.ferenczi]

Dude its been 3 years, intel is a giant ship to steer. This is also a one sided story, the engineers are talking up their own work without any dissenting opinion. Engineers were talking up chiplets and foveros but became such a shit show they reduced from 4 to 2 from sapphire to emerald rapids, from 4 to 2 from meteor lake to lunar lake, and nvidia are having problems with just two on blackwell.

[@e2rqey]

I feel like that quote about high performance cores is going to age about as well as Steve Balmer saying the iphone isnt going to succeed because it doesnt have a physical keyboard

[@SunsetNova]

He’s joined the Dark Side and become corporate and lost his engineering identity 😔

[@InvidiousIgnoramus]

I wish Steve was right about that.

[@ablesam]

Intel is still going after high performance cores, but not the way Royal was going. Royal had almost terrible PPA for any given node due to its size and complexity. Of course, it was going to dominate the industry and put Intel years ahead of others in Single-Core performance. But it's multi-threading was almost lacklustre. v2 from what people were saying would have taken care of multi-threading, but not in the same way the E-core clusters would have.

[@dansanger5340]

I'm not sure he was wrong. Sounds super complicated to get right, especially the thread scheduling. There might be all sorts of pathological situations to work through. Maybe he was afraid it would become another Itanium.

[@Pressbutan]

More threads = more crowded ring bus

[@zuckdaddy1596]

@@Pressbutanthat's not an issue

[@spoonikle]

@@Pressbutan Act like you know better than Jim Keller. 🤣😂💀

[@Voidkitty_]

Honestly all i could think through this entire video was how much of a disaster scheduling would be, big little and x3d on 1/2 ccds are already causing problems for NT and Linux, modern OSes aren't equipped for complex core layouts

[@tuckerhiggins4336]

​@@zuckdaddy1596It's an issue right now lol

[@big.atom37]

I think Pat made the most sensible decision. This "core splinting" could have easily become a resource sink hole in terms of protecting against new security vulnerabilities, configuring load balancing and power distribution and ironing out all the bugs and issues that different applications would have had with the new topology. I don't think that the current Intel could afford that especially if it turned out to be a flop for some unexpected reason.

[@Cineenvenordquist]

Have you seen its market cap? No. If Pat's not spinning out the Beast Core as a competitor he's faceplanting in RICO or just pivoting to fab spinups. You get a MdB (cough) fab, you get a MWCNT/MXene fab, YOU...get a diamond semi NPU fab with quantum and optical sidelines, you get a PV+compute fab! Your fab is polymer semiconductors, but you also do buried hi-V plenum and actuators!

[@dragline.]

Intel's downfall is truly astounding.

[@SAKTHITech]

This CPU sounds like a scheduling nightmare though

[@Pushing_Pixels]

So was Ryzen at first, and so were E-Cores. It's a problem with all new multi-core arrangements, but one that can be overcome.

[@flimermithrandir]

Should not be though. If the Program knows what’s better, it should easily be able to tell the CPU what to do imo. Doesn’t sound to hard to program to me.

[@thewhiteknight9923]

​@@flimermithrandir problem comes with whether or not development teams are still working on the programmes. Windows is the only one who can effectively do this

[@EthelbertCoyote]

Pat's only real failure in my mind is I think he is intentionally following what others are doing relying on Intel's size and position instead of finding underserved markets and niches and making wins. It's not about preserving your past it's about setting in place small wins for the future. Like a squirrel stirring nuts for winter. Winter is coming.

[@SunsetNova]

He’s joined the Dark Side and become corporate and lost his engineering identity 😔

[@EthelbertCoyote]

@@SunsetNova Sadly maybe, but few people would not make that error? When a business has a money problem the most straight forward thing to think of is to solve it with business solutions and likely as an engineer those seem the fastest to him.

[@esra_erimez]

I'm truly shocked by what I'm hearing about Pat Gelsinger. I really had expected Pat to turn the company around. It seems like he may be stuck in i386 mentality.

[@thepunish3r735]

Ehh…. I like alot of the roadmap thst has been happening ,, idk about behind the scenes , but I can’t take the word from an employee that had it out with Pat and left

[@NJ-wb1cz]

​@@thepunish3r735 what do you like the most?

[@thepunish3r735]

@@NJ-wb1cz I like how they made a department for automotive., even non electric cars have 1,000 to 3,000 chips inside of them. They partner with zeekr a Chinese car company and practically is powered by intel, software , chips, mobile eye. Every car that can drive by itself will be Intel and/or mobile eye involved ( except Tesla ). Also a department for green energy / getting their chips to be the brain of solar plants, geothermal, wind , hydro … will be the first to produce off the high N.A. euv machine end of 2027, “ reason why they. Canceled beast lake” beast lake is 10nm and they’re going smaller … and very excited for the 1 trillion transistors on a chip , q4 2030

[@NJ-wb1cz]

@@thepunish3r735 I see, so you're viewing it as an investor, not as the end user of their processors. When you're buying a car you don't really care who made the chips in there, it's just some OEM for you. You're using the car not the chip. But you do care about the processor you yourself bought and selected for yourself based on your own needs, which is why people watch videos on Intel but they don't watch videos on Texas Instruments or Omron, despite everyone owning way more of their chips than Intel's or AMD's

[@mraltoid19]

So, Royal Lake 1.1 was kind of a Super Sayain Fusion most of the time, and can De-Fuse when two Sayains are needed for a Job?

[@NadeemAhmed-nv2br]

AMD tried something similar with bulldozer but failed. Pat was smart not risk the companies future for some sci-fi project which has already failed once and those same engineers are trying again, this time at intel instead of AMD

[@mraltoid19]

@@NadeemAhmed-nv2br Now that you mention it....Royal Lake sounds like an Evolved Bulldozer...hey SMT was garbage at first, maybe Intel figured this out?

[@Gen0cidePTB]

​​@@mraltoid19 Ryzen is an evolved Bulldozer. Bulldozer evolved by separating the cores. Royal Core sounds like an evolved Itanium, because it sounds great at first but then when you start to pick it apart it sounds terrible to work with. I mean, what's the cost of a "split" or "merge" in terms of power and latency? How do you constantly maintain a fast cache that must divide into two caches with the right info on each side without having to dump and reload data? If the split cores have lower performance than the joined cores, how much more complex does the scheduler have to be to reliably guess when a task would complete so cores can be equally loaded, especially when the task is happening after a transformation that hasn't started yet? Does the CPU overhead of this more complex scheduler, the power and latency overhead of this constant FPGA style rewiring and the complexity overhead of the fetch and control system for this still outperform classic cores?

[@Cineenvenordquist]

​@@NadeemAhmed-nv2brHey if all I have are (old!) laptops should I have enduring bitterness about bulldozer or not?

[@ravewulf]

While it sounds really cool, I bet the Windows scheduler would royally screw it up for a good while. It has enough trouble dealing with multiple types of cores and adding in big core that can dynamically split themselves would be a nightmare to work with in software at the low level

[@Noisy_Cricket]

@@ravewulf what would help is that Windows wouldn't have to split it if it didn't want to, or it could have some kind of "default" mode when it isn't doing anything particularly complex (like 1 big + 4 little) in an "8" core configuration. If the 1 big core had 4 "normal" cores worth of cache, then it could be REALLY fast and do away with the need to have higher clocks and hyper threading. That would make Windows' job simpler in some ways. When encountering tasks that demand parallelism, Windows could split the other core at will and improve the execution speed. It's not *that* complex to be honest.

[@aravindpallippara1577]

Windows may not be the dominant consumer OS for long if they continue their current direction. It takes one company like valve to release something like Steam Deck to laptop and desktop sphere to show people Linux on consumer hands isn't scary at all.

[@gamingmarcus]

Didn't Tom say that it just appears as 6 big cores in Windows? Sounds to me like the scheduling would happen on the hardware side. No way you should let the windows scheduler anywhere near that.

[@ravewulf]

@@gamingmarcus That's only when they are combined, otherwise you wouldn't be able to run concurrent threads on a single core. The CPU also doesn't know what needs a big core vs splitting into smaller cores without input from the OS. You have to expose it to the OS somehow. Hypertheading appears as individual cores but Windows still has to treat them differently (fill the primary thread of each core then use the secondary thread only if needed). Same goes for P-cores vs E-cores and V-Cache vs non-V-Cache cores. Then there will be latency penalties for splitting and recombining cores due to the need to flush the registers/caches. It's nowhere near as simple as "make Windows see discrete cores and treat them all the same

[@Noisy_Cricket]

@@ravewulf yeah. This is why you'd want to stay in one configuration or the other most of the time. If you had 2 big cores configurable to 8 little cores, most of the time you'd probably want 1 big and 4 little in Windows. If the user session encountered a situation that demanded alot of multi threading, then Windows would have to briefly shift everything to the 4 cores while the big core split itself. Ideally, you'd do this while the software appeared to be loading something or was mostly idle.

[@TrueThanny]

Based on what you're describing, rentable units was a type of SMT. Instead of just duplicating registers and using the same execution units, it would split off execution units as well. Still hard to say how that would be better. Sounds more like the full wide core would run at much lower clock speeds, while the split core sections could ramp speeds up to run multiple threads faster than multitasking on the single larger core would allow (with, say, four-way SMT to avoid having to do constant context switches). Sounds like we may never find out.

[@humbleeagle1736]

This is yet another nail in the coffin for Intel to file for bankruptcy.

[@zeitgeistx5239]

They can’t go bankrupt when their getting billions from Uncle Sam.

[@DanielM.-mq4rm]

Intel will be split in several companies. CPU-Intel is dead for sure, but they will survive with contract manufacturing. Western countries need an alternative for tsmc when china invades taiwan

[@rosomak8244]

@@DanielM.-mq4rm They tried contract manufacturing already several times. There is not really that much demand for that. Western countries need many things when looking at it rationally. However please explain to me how it is better to rely on the sanctions this or that creasy americans compared to the Chinese who are usually just happy to take your money and shut up about politics?

[@leoSaunders]

it's funny bc if AMD were to go bankrupt/into restructuring, x86_64 licensing would and would've gone (before Ryzen release) to Intel according to agreements (which the state could probably still null and void).

[@Akkbar21]

Dude. You have NO idea what their situation is. This is the geek narrative on yt and social media, but in reality Intel still have tons and tons of revenue and will be fine

[@thepunish3r735]

I asked my brother who works at Intel Oregon facility.. beast lake was supposed to be on 10nm and it got cut to roll out 7nm chips instead.on the new high N.A. euv machine. While Beast Lake was a significant step forward for Intel, the 7nm process represents an even more significant advancement and is the foundation for Intel's current and future products. 🤷 that’s what I was told

[@TheDarksideFNothing]

That's bizarre if they were planning to be putting out 10nm chips as new high end products in 2027...

[@thepunish3r735]

@@TheDarksideFNothing that’s what I said 🤷 … caz I know 20A and then 18a will be out next year. Caz they have a contract with Microsoft to be making 18a chips with them in 2025.. which is 1.8 nanometers… will just say just say they wanna be on a smaller node than what was planned .. maybe I heard it wrong ..but I know they want to be using the high N.A euv machine by then

[@dakalro]

​@@TheDarksideFNothingcause mlid got played, beast lake is now in its 6th year since inception ... imagine how well it did

[@TheDarksideFNothing]

@@dakalro not sure what you mean

[@dakalro]

​@@TheDarksideFNothingA viable product was at least very delayed so initial plans for Intel 7 make sense. Someone on reddit was adamant royal core was great it just had poor multicore perf in beast lake but would have been great in the new 2028 deadline (v2). which funny enough matches mlid's source timeline. But royal core started in 2018 ... 10 years for the first release is something only the very rich can afford, not current Intel.

[@gammafilter]

😂Keller joined AI chip startup Tenstorrent as CTO in December 2020 and became its CEO in January 2023. Pat f'ed up. Maybe Jim will buy intels chip fabs....

[@cairnex4473]

I think it's time to concede that Pat Gelsinger is no Lisa Su.

[@ablesam]

except Intel is more than a decade's worth of mess for just one guy. Pat's made wrong bets, but what's happening rn is internal rot exploding out. AMD just needed one really solid cpu uarch (Zen) and a solid roadmap and they were good on all fronts. remember AMD also killed Keller's K12 ARM cpu, he himself said it was stupid of AMD to do that.

[@Keyboarder11376]

I just wonder how many speculative execution vulnerabilities these "rentable" cores would've had.

[@augustodeleon4358]

Beast Lake Next would have been amazing! Each core more than double the IPC of Raptor Lake!

[@Pavlobot5]

yeah... that never would have worked. Would windows decide when to split the cores or would the cpu? could you lock it in the BIOS and would that break the scheduling? You're saying the default state is super single core but would that be the lowest power state in reality? Would the "preferred" high clock core be split first since it requires the lest voltage? It would be cool if it worked, but Pat probably asked all the same question I just did and decided he didnt want a "Bulldozer" 1 cpu but really 2 cpus product

[@haven216]

Windows never optimised for bulldozer. Intel typically gets better treatment by Microsoft for os-level support.

[@Noisy_Cricket]

You’d simply have to program the software for that. Yeah, it’d be complicated, but if Intel were to properly test for it, it could be worth the gains.

[@Lue1337]

Yes It looks like a nightmare tbh.

[@NadeemAhmed-nv2br]

​@@Noisy_Cricketyou really think optimising specifically for this hardware which is a very small fraction of the market specially when Intel is hemrajging server market share which who is to AMD being preferentially optimized just like Intel was for the past two decades. This is something you should attempt when your in the firm lead and one bad gen won't destroy you

[@Noisy_Cricket]

@@NadeemAhmed-nv2br that’s a good point lol. With the recent Zen 5 and Windows shenanigans, it seems like Microsoft is finally starting to look to AMD as the “leader” AMD develop software to suit them.

[@mercurio822]

AMD Should swoop in and get those engineers and develop rentable units

[@zodwraith5745]

It's easy to play the blame game and throw it all down at Pat's feet, but there's a _massive_ difference in time and costs in having a theoretical blueprint and silicon powered up. With all the financial difficulty of _everyone_ right now outside of Nvidia, but especially Intel, it's entirely possible Pat just couldn't take the price tag and time to ROI so he shelved it. Hindsight is always 20/20 and it's easy to look back and say Arc was a mistake but keep in mind _everyone_ was hyped about Arc all the way up to when it actually surfaced. *_You included._* To insinuate in the same breath that Beast Lake would have been an automatic win and should have had billions continue to pour into it is just brainless. Intel is making plenty of money, even today. They have a _spending_ problem that has them pouring _vastly_ more into R&D than Nvidia or AMD thanks to GPUs, CPUs, _and_ foundry nodes that are even beyond what TSMC has on deck. It's ironic that you laid all the blame on Bob Swan for playing it safe and having no desire to innovate, but now that Pat's swinging for the fences he has to go because he hasn't brought home an instant championship?

[@aravindpallippara1577]

So Pat's swinging for fences plan is to compete directly against Nvidia and AMD's GPU divisions? From a company that doesn't know how to be in second place at anything, let alone 3rd place.

[@zodwraith5745]

​@@aravindpallippara1577 They not only surpassed AMD's second attempt at raytracing and upscaling on their _first_ attempt, they did it on their first DGPU _EVER._ That's flat out embarrassing. Then when you account for how far drivers improved in just a year? They accomplished more in a year of driver updates than AMD did in a _DECADE._ Not to mention they did and are still doing it in a day where drivers are infinitely more complicated than they were back in the days where AMD's sucked. Anyone that expected Arc to be a market leader right out of the gate is an idiot. Sure the overall performance isn't there yet and that's why none of us bought them, but it's honestly impressive what they accomplished in their freshman outing. How successful do you think AMD would be if they tried to jump back into wafer production tomorrow? Their fab Global Foundries they spun off that never stopped trying is still well behind Intel and TSMC. Most of this is armchair hindsight QB bullshit to begin with because a lot of this was began when Intel was falling over it's stacks of money but was losing it's edge. But I'm sure in your wisdom you're a billionaire because you knew when to buy and sell bitcoin?

[@jesh879]

@@zodwraith5745 who hurt u

[@zodwraith5745]

@@jesh879 AMD, Intel, Nvidia, Asus, Gigabyte, MSI, Corsair, Crucial, etc. etc. The only company I _don't_ hate after 27 years of building and selling PCs is EVGA and now they're gone. Point is, there's plenty of _valid_ reasons to hate Intel. I can do it in seconds: Their chips guzzle power and if they didn't know WTF was wrong with i9s they should have said so from the beginning instead of dragging all our dicks in the dirt. If you have to stoop to attacking a single person based on speculation and objectivism that's no longer journalism. That's fanboism. Show me where Tom singled out Lisa Su for the Zen5 flop, Zen4 pricing, X3Ds blowing up, over promising on RDNA3, Zen3 not being able to hit advertised speeds, trying to force AM4 users to 500 boards only if they wanted Zen3, the joke that was the 64-6500xt, 6600xt BIOS, straight up lies in 5800xt and 5900xt marketing, the 5700 non-x not being a fucking 5700, cancellation of high end RDNA4, etc. etc..... He doesn't because he's biased towards AMD and has violently different standards for them vs anyone else. He blames any AMD fuck up on engineering, marketing, timing, limitations, investors, literally _anything_ and anyone but the person running the F*ing company. Remember how much he sucked Raja Koduri off when he was at AMD making mediocre GPUs but when HE'S the one that designed Arc magically it's all Pat's fault? I hate all these companies, but you at least need to be fair, and this channel is anything but.

[@jesh879]

@@zodwraith5745 not rly he has talked about a lot of those things. but this channel is mostly leaks and smaller current issues are not the subject matter for here. there are other hardware news outlets for that. i think he is quite fair to all companies. i have been listening long enough to know he is an intel fan. it's just now is not a good time to be one.

[@excitedbox5705]

It is actually genius. What is most power intensive is moving data across the chip. And what is slowest is waiting for data to come from cache/memory. Where an instruction is ~4 cycles, fetching it even from L1 cache can be 20-100+ cycles on the low end and increasing far beyond that. Increasing performance depends on removing those wait times. Instead of normal branch prediction, AMD is following more branches for longer in ZEN 5. So if you can split a thread every time you are not sure about a branch and follow both paths, you can avoid dumping everything and starting over whenever a prediction fails. At this point compute pipelines are 16-20 stages meaning you have a huge lag whenever you have a miss. You will need cores big enough to handles every branch for as long as the longest time it could possibly take to fetch the next instruction. Then multiply that for as many tasks as the PC needs to handle at once for the CPU to never wait for data.

[@tranquil14738]

Oh my gosh I didn’t realize that’s why it splits that is genius

[@mraei]

The idea of reverse hyperthreading has been around for 20 years

[@TheDarksideFNothing]

What products feature it?

[@Dudewitbow]

​@@TheDarksideFNothingtech generally has ideas out before hardware is possible to come into fruition. e. g the idea of VR, AR, Quantum Conputing, and Real Time Raytracing existed long before their actual implementations were made.

[@TheDarksideFNothing]

@@Dudewitbow I'm well aware. Which is why I'm wondering what OP is getting at with the comment. Having an idea is great and all but is only like 1% of the effort. Lots of people have ideas. Actually making it happen and work well is most of the effort. Without context the comment has an air of "what's the big deal this is old news." Maybe they didn't mean it that way, but if they did, I'd say "who cares about someone's pipe dream that they never brought to market." It's not every day a team fundamentally changes the paradigm for how a CPU functions. I'm really bummed to not get to see Royal Core fully realized.

[@ivankovachev8835]

Multi-Chip Modules, a.k.a infinity fabric and chiplets were invented in the 90s as well.

[@millosolo]

Flying cars have been longer

[@douglasmurphy3266]

Are we sure that the switch away from pure high performance cores isn't some open secret that we are hitting some sort of physical limit where things are not working as imagined once they are put into a smaller transistor process? Like that cache gets diminishing returns below a certain point, logic is hitting its limit, and dynamic f*ckery is the only solution they see moving forward?

[@LiveType]

No. Lots of single core scaling left. Not like 10x, but certainly room for ~3x over the next 15 years.

[@gamingmarcus]

The whole world of 3D technology is still available and in its infancy. There is lots of potential left in the good old silicon.

[@olo398]

aliens.

[@TheDarksideFNothing]

It's fairly widely accepted that we are in a sort of final descent approach to the limits of silicon as a substrate, but I don't think that's why they abandoned single core performance projects. Also, I forget what it is now, but there's a planned next substrate that should keep things moving forward.

[@Hugh_I]

It may be getting harder, but we sure aren't at the end of it yet and this decision has little to do with it IMHO. If we were at the end of it, clearly Intel wouldn't scrap their best chance of pushing it the last bit and potentially beat AMD in single core for as long as they can. This decision sounds to me like exactly what Tom explained: Intel thinking some of their biggest markets like AI datacenters, hyperscalers and HPC are segments that are ok with not having the fastest cores if they can have many instead, and more efficient ones at that. Its a gamble I would not make if I were Intel, but trying to focus on those market instead of having the best for desktop/mobile/workstation may be one desperate attempt of trying to save profitability.

[@Summanis]

Considering how Intel missed the crypto GPU boom and now the AI boom, I can only assume that if they're axing CPU investments, CPU markets are about to explode.

[@techluvin7691]

If there were no issues, why was it cancelled? Under Intel’s present architecture, an increase in performance cores isn’t possible. Too much power…..too much heat.

[@MooresLawIsDead]

Watch the video. I literally spend a bit talking about why he made the decision.

[@danielkarlsson9326]

Sounds alot like how GM handled SAAB. "Everybody loves rear wheel drive its the future." Gm VD Robert A. Lutz on why he wanted SAAB to be badge engineered Holden cars...... SAAB is the quintessential Forward Wheel drive car and has only done Forward wheeled or all wheel drive cars. To me Intel reminds me more and more of 2007 GM.

[@kingofstrike1234]

16:00 I don't see a guy, I see a lizard

[@Zorro33313]

interesting idea, but in reality having a 12 raptor lake cores level of performance as you best CPU by 2030 is mildly impressive putting it nicely. It's pointless in servers and gamers don't really need it as well, cuz gpu is always a bottleneck. It's a pipe dream design reinventing SMT for millions $ and Pat was completely right to ditch it. Remember Nuvia with the same ideas? Remember Jim working with them? Where is Nuvia now? That's the case. You need to stop praising Jim as god and look at the situation realistically. 12 raptorlake cores performance by 2030 is trash. It's laughable and we're not even talking about manufacturing problems, density of those cores, internal bottlenecks, TDP and dark silicon effect. AMD already tried to reinvent SMT and they got Bulldozer that was amazing on paper. Another paper-beast. But ho it went in reality? Paper-beast lake was likely to turn out the same way.

[@ZeroUm_]

Having seen the whole video, honestly, that dream architecture seems to be too good to be feasible. I'm not surprised it was cancelled, given the troubles they've been having, for a decade already, delivering "simpler" designs. It makes sense for these more revolutionary concepts to be done in start ups, and for Intel to staunch the bleeding, refocusing and trimming down their business, without these Hail Mary attempts.

[@millosolo]

They don't understand the level of mental masturbation at work when writing an ISCA paper.

[@saidalamry125]

This video really feels like a hit piece on Pat. I get it now. It seems a lot of people in intel hate him that they are willing to offer enough information to make hit piece like this that would make anyone supporting him stop doing so.

[@waynesworldofsci-tech]

I’m wondering if Pat is living in the past. He may no longer be on the cutting edge in his thoughts on the future.

[@SunsetNova]

He’s joined the Dark Side and become corporate and lost his engineering identity 😔

[@thequestingblade]

You're probably right. It's also possible Jim Keller is past his prime, being even older than Pat.

[@modalityepstmlgy]

Pat preparing another Bible verse to post after watching this video.

[@Strykenine]

I have always stood by the idea that Pat was the right choice for intel as CEO. This sure is disappointing to hear.

[@SunsetNova]

He’s joined the Dark Side and become corporate and lost his engineering identity 😔

[@Real_MisterSir]

Intel is confused. It hurt itself in its confusion

[@hammerth1421]

Seems like Royal Core dared to ask the philosophical question of what a core actually is. Sounds like an easy question to answer, but it kind of isn't. For the longest time, every single core has contained multiple execution units which can execute several operations in parallel. There's no good reason why you shouldn't be able to design more architecture around them which can dynamically split one core into several separate units with fewer execution units or remain combined as one fat high-performance core.

[@Razzbow]

Buddy met the most legendary CPU engineer of all time. What a cool guy.

[@legi0naire]

I just want a 8-12 core on a single die, with as much cache as possible. Good power efficiency is a big +

[@christophermullins7163]

So.. AMD?

[@isa_L]

Well you want 7800X3D, now with 20% performance increase with The Latest KB5041587 update.

[@chillnspace777]

​@@isa_Lthat much of an uplift?😮

[@isa_L]

@@chillnspace777 see for yourself, Hardware Unboxed and KitGuru already Testing this Updates 2 days ago

[@christophermullins7163]

@@chillnspace777 no. 10% is what you get. Not 20

[@tpurves]

I was with Pat this week at the DB Tech Conf, he was very clear that Intel is too far behind to compete with nvdia in high-perf cloud training. But that intel would focus on client and on-prem server (e.g. Xeon+ attached AI). He was also clear that era of intel trying to build lot of designs on a lot of nodes at same time is done. He's cutting 15k staff and cutting projects trying to pick fewer categories to compete in. He said last 2years was to catch up on innovation (in process and uarch) at the expense of gross margins and efficiency, now they are done and cutting people and projects and only focusing on what products could get them back to profitability by the time they get to 18A.

[@ablesam]

you were at that conference? cool. so, he's cleaning house and exiting from various markets, so that client (Core Ultra) and datacentre (Xeon with AMX) remains the sole focus of Intel Products. thing is, anything after Nova Lake (2026-7 CCG) and Diamond Rapids (2026-7 DCAI) is being either cancelled or undergoing huge redefinition, since Pat's cut most of the R&D for those projects and moved all the people from there to make sure everything else between 2025-26 launches on time. The Forest line is getting axed completely after Clearwater. so the last three years he's spent capital on foundry and some uarch and now has decided Royal isn't worth the effort when it was supposed to launch. Idk what his plan is, but it's short-sighted as of its revealing right now. The Oregon team was the some of the best talent Intel had, and now they are getting rid of them.

[@tpurves]

@@ablesam Ya. He said almost all of the 15k to be let go have been identified. He acknowledged that included letting people go, some of them with 20,30+ yrs experience at Intel. For better or worse, his strat was innovate hard as possible for burst of 2 years, soak what they could from existing teams, then cut aggressively just before the cash runs out and go 100% for efficiency with whatever they've got by now. Will it work. I don't know. risky plan. Maybe only plan they had available.

[@ablesam]

​@@tpurves I sympathize for those being let go, or in this case, pushed out. Some of those people especially in the design side were some of Pat's good peers during his first period at Intel. They believed he would be able to fix majority of the issues within Intel. I doubt them being let go was part of their idea of him fixing the problems. sigh. But Intel is definitely overpopulated, so streamlining the company is definitely needed. I just wish Pat knew who he was getting rid of and what the current teams can achieve in reality. Anyways, the IDC and Austin teams are being merged and the Royal ideas will be worked upon for Intel to create a new core competitive in PPAC. Will have to wait till 2029 for the results of that however. Foundry though, needs to go. I'm hoping talks of spinning it out within the next three years are in the works. Design is the only thing Intel needs right now to survive and prosper in the next 5 years. Foundry needs to go after 14A launches with customers. I can't believe this was his plan the whole time though. Ruthless, painful, but seemingly what Intel needs. I hope it works out for them.

[@TitanPlakInside]

I don't know what the heck is Pat planning to do next about royal core project. Splitting its features into several gens?? That is so dumb!!!🤦‍♂️

[@jaim1555]

Correct me if I'm wrong. Royal Core sounds similar to AMD's bulldozer architecture.

[@ironhead2008]

Yeah, it has a certain similarity

[@Noisy_Cricket]

Bulldozer was probably ahead of its time tbh. There just wasn’t the density to build a real splittable CPU at that time. Now there is. And tbh chiplets would probably make such a concept work better than a monolithic chip. If AMD could make a CPU SOC with 2 chiplets that each had 2 big CPUs per die that could split into 4 small and 1 big, or 8 small depending on the situation, that would be *crazy.* I think this might even be a better idea for AMD than for Intel.

[@dex6316]

Royal core has multiple execution units that can either combine to form one super execution unit or operate independently as 4 cores. Bulldozer was basically one core, but there was an extra set of integer execution units allowing it to operate as 2 cores for integer and one smt core for floating point. The integer execution units also couldn’t combine to massively boost integer IPC on 1T per core. Pretty different I would say as Bulldozer was basically one FP core and 2 Int cores per module.

[@jesh879]

@@dex6316 true, and there was a reason for this. AMD was betting the farm on HSA and that would have offloaded a lot of FP workloads.

[@pneuma23093]

15:17 don’t do that. Don’t give me hope😭. Also, “Off with his head” for Pat.

[@thepunish3r735]

Beast lake was canceled caz the new high na euv machines will be used to make the chips .. the chips will be smaller with more transistors where as beast lake was for 10nm chips.. I get my info from actual employees , and not a company hopper that butted heads while he was there and said all this all while drinking and wasn’t even there when the high N.A. euv machines arrived

[@aravindpallippara1577]

​@@thepunish3r735you can shrink existing designs easily for smaller nodes, it's the other way around that's stupid hard. I don't see this as a good argument. Also what happened to the CPU architecture team operating like an outsider customer to foundry team if they have to worry about NA machines and such?

[@mlw636]

@@aravindpallippara1577 No, that isn't how designs work for chips, you can't just "shrink" them, you'll have to redesign them.

[@aravindpallippara1577]

@@mlw636 Yes you do, but it's significantly less of an effort than coming with another ground up microarchitecture for _NA_ EUV machines.

[@NZRanger]

When you're financially strapped you tighten your belt... While these chips look great on paper their production cost vs sales will be very limited to start with.. potentially leading to a "Moore's law is dead" video saying "We can't believe Intel made this chip when they had no money!" because that seems consistent with the way this channel appears to work.

[@intelshill5099]

Royal Core maximizes power per area instead of thread performance or energy efficiency. With tiling being a solved problem, die space is scalable, so you don't need to maximize power per area if you want to be competitive on total power. Having P-cores and E-cores allows a single chip to maximize for both thread performance and energy efficiency at the cost of requiring more die space. The Royal Core strategy ends up being the Budget Core strategy that only really competes on manufacturing cost.

[@aravindpallippara1577]

If you want energy efficiency in Big Little architecture you should be willing to turn off one side of the chip when you want to save power. I haven't seen Intel turning off any part of their chip ever because that loses multithreading peak performance massively.

[@TheDrunkeNNN]

It does not have to be that way. Think of one thread allocating the cache and branch predictors of 4 little cores when needed. Statistics tells you that that would roughly yield a 2x IPC gain -> no competition -> lower clocks -> less power. BIGLittle in that sense is simply a flawed architecture.

[@intelshill5099]

​@@TheDrunkeNNN The more you tailor the chip to a specific purpose, the better you can make it for that purpose. The rentable units idea is definitively committing multiple purposes to the same die space.

[@intelshill5099]

@@aravindpallippara1577 Depending on the workload, keeping everything on can still be more efficient than relegating everything to the efficiency cores. Chips are most efficient near their lowest operating voltage, and if your workload is maxing out the performance available at that voltage, engaging more die space at the most efficient voltage can be more efficient than raising the voltage to increase clock rate, even if the additional die space you recruited is not specifically optimized for efficiency. However, if the processor is only loaded to a point where the efficiency cores alone were able to handle that load at their optimal voltage, then it does make sense to turn off or idle the P-cores. Some of the information about Lunar Lake has indicated that it has switched to an E-core first strategy, which is more like what you are saying. Completely shutting down the p-core die area probably doesn't give as much advantage as you might think, but the option is there. Minimizing the power consumption for inactive silicon is already a consideration even in performance oriented designs, as wasted electricity is not providing performance.

[@DJdoppIer]

Dang, the Royal Core project sounds like it could've been really interesting! Having P-Cores that can act as multiple E-cores when needed would be WAY more efficient than the current fixed P-core / E-core setup. I imagine it would've been a nightmare to program this configuration, but it almost certainly would've been a game-changer! It's so sad that, once again, someone in power but with no real knowledge of the products they make ends up killing something promising in a damage-control panic.

[@NadeemAhmed-nv2br]

Amd bulldozer was exactly that made by these same morons who are now working for intel, sure it might work, it could lead to Intel Being broken into half does in companies like AMD

[@defeqel6537]

Just thinking how caches would work from a performance perspective is giving me headaches, let alone security

[@aravindpallippara1577]

​@@defeqel6537 don't worry there are smarter people than us who can figure that out. AMD is doing perfectly fine with shared cache and mountains of it right now.

[@defeqel6537]

@@aravindpallippara1577 yeah, L3, not L2 or L1, registers, etc.

[@aravindpallippara1577]

@@defeqel6537 Well that is true, somewhat.

[@AthensEnjoyer]

@10:08 I think AMD had tried something similar in the past. With their Bulldozer(?) architecture (FX 8000,6000,4000). Two cores working as one, or tricking the software to be viewed as one? I remember that it didn't work out that well for them then. It will be very interesting to see what Intel is going to offer. 😅

[@noobgamer4709]

Yep. bulldozer seems like the infant state if it. and apps were design with intel arch where single thread is everything back then. these do have drawbacks. as a result, bulldozer was a failure which prompt them(AMD/Jim) to designed/makechanges to zen. which is why zen1 was not the full realization but instead zen2 was.

[@AthensEnjoyer]

@@noobgamer4709 Yeah, hopefully now that software can take advantage of more cores, we'll see interesting results from this attempt.

[@xuyukun123]

it's different, AMD was forcing cores to share vital parts making each core weaker, Jim Keller's vision is to have cores combine into even stronger ones

[@FutureChaosTV]

If I am not mistaken two cores with one ALU each shared a single floating point unit. That and the fact that the frontend feeding the ALU's were too weak to full load the cores lead to the outcome that Bulldozer would suck at single thread applications. But if you played a game and at the same time used something like OBS it would still have a little of performance left and as such the stream/encoding would be very smooth... That's how I understood it at the time from people using the CPU.

[@kevinerbs2778]

@@FutureChaosTV Cache trashing kills it's performance & so does the long FPU pipeline.

[@nazgul77de]

Operating Systems would need serious extensions in their kernels to support dynamic CPU hardware.

[@j340_official]

Given how long it takes to bring up a cpu core, you’d think he would’ve cancelled it from back in 2021/22. Why now?

[@LiveType]

Tom doesn't have the full story here. Pat isn't stupid. He would have asked the exact same questions and for some reason after analyzing it all, arrived at the conclusion that saving Intel from bankruptcy while focusing on manufacturing capability rollout is more important than the focus on single core performance. Plus what was described sounds like an even bigger nightmare than e cores. Whether or not that is the right call is hard to say. These things are only obvious in hindsight. But laying off $1.5 billion per year in talent is very questionable. Like that scale just doesn't make sense. Maybe Intel was really that bloated. I don't know. But it's fishy. There's more to the story than what is being told to the media. Intel also won't go under basically no matter what. The government will bail them out every time until the military switches suppliers which will take a very long time.

[@j340_official]

@@LiveType​​⁠there has to be far more to the story because if what Tom is describing is true that a cpu could split into multiple cores how would windows have handled that and also the thread scheduling ? Are there security vulnerabilities there? Power consumption concerns ? Were there alternative ways to scale performance? Avx10.2 will be supported on both p and e cores. Is there even a need for this particular design? Finally the suggestion that Intel doesn’t care anymore about CPU core performance is questionable given the evidence we have from Skymont. Why spend all that time redesigning the E core to have much higher IPC than Gracemont if cpu core performance is not a priority ? Pat also knows ARM is coming (incl possiblynvidia/mediatek), Apple is crushing it, and AMD is firing on all cylinders in the data center. What were yields like of the new royal core design? So many questions. Just because a particular design is not chosen doesn’t mean there aren’t viable alternatives. There’s more to the story.

[@ablesam]

@@j340_official In regards to your first question, Intel is running out of money to keep funding 3 core design and arch teams (Core, Atom, Royal), so they've axed Royal, and are merging Atom and Core. Why Royal got cancelled? Few things ik, but take it with a grain of salt. Royal had terrible area efficiency, and it was always focused on ST perf, MT came with v2. Secondly, it was taking too long. LNC which started back in 2019 (same time as RYC), is now out in 24. RYC was scheduled for 28. It's almost a decade of development for a single arch. It's just a pitfall of money for Intel. Also, this isn't Keller's core, he's a guy really focused on PPAC, and Royal was terrible on Area and Cost. This was the Oregon guys trying to make an ST 'beast'. Pat's moving resources to IP that really matters (E core guys and GPU IP), where they can make a difference/turnaround for the design side in the 2028-30 timeframe. Until then, he's axing teams, archs, projects, product lines left and right until he is sure Intel can execute properly before scaling up again.

[@j340_official]

@@ablesam very interesting perspective thank you

[@MrHav1k]

I'm with you Tom, this is one of Intel's dumbest decisions of all time if true and BOY have they made plenty of dumb ones.

[@Lue1337]

I think its an awesome concept, but the software side doesnt seem easy.

[@johnnyringo35]

AI will replace 90 percent of coders.....do it faster, better , and cheaper. Learn to code they said.... 😂

[@jondoe9596]

Tbh, I'm glad Pat cancelled Beast Lake. That project sounds like a scheduling nightmare for the OS. God knows AMD is already having a hard time with scheduling tasks between their CCDs, and design of Zen is fairly straight forward. Jim Keller is great and all, but this might have been too ambitious on his end. Intel just needs a 12 p-cores only gaming variant, and 8pXe variants for multithreading tasks.

[@xmj6830]

You should organize a podcast with Dr. Ian Cutress about CPU design and Intel future. That would be very interesting conversation.

[@lorenzzz8630]

This is giving me dejavu of when that one AMD executive gave you leaks for a while

[@Seanweekhaizhen]

what happened?

[@lorenzzz8630]

@@Seanweekhaizhen it was years ago but an amd executive (I forgot the name but it was an old dude) gave him leaks for like 3+ vids. It was for CPUs I think. It was revealed who the leaker was when the leaked CPU gen was released

[@DWGoogleUser]

Still decades behind the best Threadripper

[@alexanderbelov6892]

Threadripper is the same EPYC, but targeted for HEDT E-ATX. Intel has its own server line Xeons. They only need to deploy single processor config into E-ATX MoBo.

[@endersjehuty7721]

0:28 and thats how toy story 5 will be out on torrents in premiere day.

[@3-valdiondreemur564]

I don't think Intel will ever be competitive again, hearing this. AMD keeps fumbling at every opportunity, but at least they know where they're going. Intel isn't doing either of those things.

[@russellmm]

You are focused on what was canceled. You should be asking yourself *why* Pat would have done this. It does not have to do with an engineering mind (Pat) wanting to get rid of good engineers/architects. He obviously thinks that what hardware is coming down the pipeline over the next several years is enough to be competitive. Pat is not stupid (I used to work for him), regardless of what you might think.

[@djchristian82]

I wonder if Intel ever will catch the gap between performance per watt with their M-CPU’s? What did Apple do which neither AMD nor Intel can’t?

[@ablesam]

Pat's just cleaning house. He's spent huge capital in the past 2.5 years on foundry and design side IP and is now just reorganizing Intel Products to act like an efficient CPU/GPU design company by cutting off staff, exiting specific product markets, cutting entire product lines and simplifying the roadmap by a drastic amount. Royal developed a lot of IP for CPU uarch, which the new team (merger of P and E core) will now start to integrate over the following generations (say 2029 onwards). It's what Intel needs to do, act and behave like a startup. And deliver.

[@andersjjensen]

There is something really really really fishy with this "one core can split itself" explanation, assuming even distribution that is. As in one core is 100% and it splits itself into 2x50% or 4x25%. This, ladies and gentlemen, is exactly what the OS task scheduler already does when there are more software threads wanting to run than hardware threads available. For this to make any sense the there should be a performance gain from splitting. Like 100% -> 2x60% -> 4x40% or something like that. Then it would make sense if the OS is smart about it.

[@frommatorav1]

The one big core that can morph into smaller sections or turn itself off is revolutionary. I'm not an engineer, so I can't explain how it will work. It's different than hyperthreads because they are always there. The advantage would be in pure power in the big core and the efficiency when it dynamically changes for light multitasking.

[@TheHangarHobbit]

Anybody else get the feeling that Pat Gelsinger is to Intel what Rory Read was to AMD?

[@KarbinCry]

Rory Read saved AMD and put it in a path to a comeback.

[@Deytookourjurrrs]

He’s doing to Intel what Agent Orange did to America

[@SyrFlora]

If u looking carefully it not like magically lisa su bring back amd in just a few year.. The one before him lay down some ground work before.

[@Ojref1]

You're thinking about Hector Ruiz, forever may his name be cursed

[@delongzhai4887]

here is the honest story, nobody cared about having 4 threads per core, that is useless when GPU does things a lot faster in parallel, not to mention NPU, which is what all of those companies are focusing on nowaday. I kind of agree with Pat's undoing of RC. I think time has changed, and it's time to say good bye to that. Remember the push to more efficient cores, which let the birth of Ampere, which is founded by Intel employee, Renee. Server market just don't need high performance-oriented design because it's powerful enough, so does mobile (PC included). Yes high performance is important, but the generation of leap forward has come to an end due to process node slowed evolution, logarithmic price increase in wafer cost, and packaging.

[@tj_2701]

I know he's doing other things right now but I think it would be cool if Keller came back to AMD, where he first tried this idea, and made it a reality that beat his expectations. 😋

[@selohcin]

I can't believe Gelsinger cancelled this. This was EXACTLY what gamers have been asking for.

[@james...cardinal]

is there an hedt roadmap for cpu releases beyond threadripper 9000 or intel emerald rapids ?

[@albal156]

Tbh HEDT of old is dead. If you wanted such a thing now it would be LEW or Low End Workstation.

[@mlw636]

I miss them as well, but that model died due to intel failing to compete in 2019, SLI dying (on big reason to get one was 3+ GPUs), software still not catching up to core count (past 32 cores, it is hard), 1950x (now 16 cores on desktop), and greed (AMD gave those hints at massive scaling in prices for Zen 3 TR). Intel needs an HEDT that tops out at 32+ Cores for under $2k. They simply price them out of the market. Just buy an Epyc/Xeon now, prices are almost the same now.

[@nigelwoods6979]

Most CPU security exploits are a result of how Hyperthreading is implemented. The exploits possible when a single core "rents out" resources to whatever thread passes through?

[@Teste-gp7bm]

If this thing has legs, Keller will do it at Tenstorrent. They have been building RISC-V cores for a while. This doesn't seem much different than SMT and it would probably have similar problems. I actually like just removing SMT and using E cores for throughput. The new E core design, to me, is much more interesting than this (and its already here).

[@cracklingice]

wow. I used to think 'no way I'd vote to replace Pat' but if he killed their bread and butter of high performance cores to chase something they don't have a foothold in; I've lost faith.

[@anshulsingh8326]

I hope Intel doesn't shutdown. AMD gonna rip us up

[@user-wq9mw2xz3j]

id rather getted ripped by AMD than Intel 😂 (although that just means I wont buy any new cpu)

[@anshulsingh8326]

@@user-wq9mw2xz3j Well see new AMD launches. They already started it

[@napowolf]

Royal Core sounds like a nightmare for Windows scheduling and game engines tho... Just look back on how much trouble Windows and games have with high core count AMD cpu and Intel E cores a few years back.

[@quanlethienminh6002]

yeah the Arc investment certainly hurts. I mean even AMD have to sacrifice RDNA 4 flagship due to the current market. On another note, hot wings sound like a sick party! getting drunk with 2 visionaries in the computing industry while they tell you the secret ingredient 🤣

[@jessietomich8043]

Ok Intel lets get real. Stop trying to push the envelope until your fabs are up to speed. This current process node is still just enhanced 14nm until the new EUV machines are online. Just do your best and don't bother trying to keep the single threaded crown. Next go extra wide with the cores like Apple did. Finally put a focus on bandwidth and special purpose blocks. Quit trying to catch up by doing a "me too" with AMD. Keep your dignity and innovate on your own level.

[@dennismccoy9409]

I heard on someone else video that some of the people who left Intel Is starting there own open source CPU's. Have you heard anything like that?

[@rogerk6180]

Moving to risc-v development.

[@ablesam]

yeah, the people who created Royal core formed a startup called AheadComputing.

[@MrHav1k]

Intel is the GOAT of making the wrong bet at the wrong time, so watch datacenter inferencing on CPU take off and Intel wound up having killed the wrong product at the wrong time AGAIN... LMFAOOOOOOOO

[@cheslyhouser3864]

The evidence was clear, he was holding a bloody knife.

[@frommatorav1]

There's a small chance the knife was planted there. 🩸 🗡🩸 Tom does have a renderer now.

[@highdefinist9697]

Hm... this could, in principle, be someone spreading rumors as part of an internal power struggle against Pat... But then again, if it was confirmed by multiple independent Intel people, it is very likely true as it is.

[@GodRob]

As GPUs are getting stronger, we need bigger cores, not more of them...

[@trailduster6bt]

I believe 3D VCache was originally a feature that started with Broadwell and then Intel decided not to pursue to concept further. We all know how that turned out in the end… Lets hope there’s a similar story for these BEAST cores! Less cores and more IPC is what I like to hear!