Intel Griffin Cove Leak: Killing AMD Zen 7 without E-Cores! (+ ARC Battlemage Update)

[@MooresLawIsDead]

[SPON: Thanks to Notion for sponsoring this video! Sign up for Notion AI: ntn.so/MooresLaw ] #NotionPartner #Notion #NotionAI

[@jimtekkit]

A blind man has more sense of direction than Intel with their AI efforts. It's like they're asking ChatGPT what to do.

[@tringuyen7519]

Agreed. Intel’s unified architecture for Griffen Cove is basically copying AMD’s Zen5 & Zen5c architecture. Right?

[@aberkae]

@jimtekkit Now, they join forces with AMD on the future of X86.

[@godnamedtay]

I bet u have AMD cpu, u fanboys can be seen by a blind man. Weird mf’ers bro

[@Capeau]

You just described this channel imho... its obviously such bs

[@visualpref]

@Capeau go to Cuba

[@bfish9700]

Intel not sucking is better for the market

[@lucasrem]

Play games ? Market for gamers ?

[@hellcoreproductions]

@@lucasrem Don't assume everything is for gamers.

[@zee-fr5kw]

why does intel sucks but its just as good as AMD i don't understand, how is just as good sucking

[@tstager1978]

@@zee-fr5kw they are not just as good when they have to either use more power or a better more expensive node to be just as good. With these advantages, they should be better than AMD but they aren't.

[@florin604]

​@@zee-fr5kwit reached 101C of heat in stress tests... That's over the water boiling point... How do you cool that?

[@WillFuI]

If zen 6 doesn’t get a new memory controller and a wider fabric link to the cores I’m going to be annoyed the fabric is really limiting the performance.

[@Noisy_Cricket]

I think at some point AMD is going to move away from the infinity fabric. Not sure when or how though. I think it's starting to become clear that they're reaching the limit of that design for chiplets.

[@tringuyen7519]

@@Noisy_CricketIt’s Zen 6. Zen 6 will look more like Apples Ultra Fusion architecture than AMD’s infinity fabric.

[@WillFuI]

@@Noisy_Cricket if they made the link wider it would solve the problem. they already do this on the server chips. On

[@TheGuruStud]

Then it consumes more power. I remember on old zen that IF was half of the power consumption.​@@WillFuI

[@godnamedtay]

Y’all thought the SAME thing about zen 5. AND again with the 9800x3d being the “savior” lmao foh bro, AMD is trash this gen, accept it, u have NO choice.

[@Qamar92]

If battlemage core can provide similar performance like CUDA in ML or fine- tuning model, while keeping the VRAM high and the price affordable, i wouldn't mind buying Intel dGPU.

[@Toshinben]

Nvidia has a functional monopoly on a lot of the graphics market because of CUDA. I suspect they'll eventually be forced to license it out.

[@edwxx20001]

@@Toshinben The question would be how long would it take to force them to license it, who would make the cards, and will it still matter with a timeline stretching into the 2030s?

[@jaynorwood2]

@@Toshinben Triton is being billed as a better CUDA. It is an open language that is supported as a backend for pytorch. Intel's roadmap shows them moving to Triton for the transition from Gaudi to Falcon Shores.

[@MaxIronsThird]

G31 16GB - $380 (4070/7800XT performance) G21 12GB - $220 (3060Ti/6700XT performance) Imagine those cards in April instead of November, these prices would've looked much better.

[@Violet-ui]

7700 xt is 350 and only about 10% slower than the 4070 G31 would need to be at most that price too make sense

[@102728]

G30 - $300, G21 - $200. At those prices it can sell with the performance Tom leaked. Arc fights an uphill battle in brand recognition, brand loyalty, efficiency, driver support and they have to recover from the horrible alchemist launch. If their drivers are good enough for a casual buyer now, they need to stand out in some respect, and that'll have to be value. That way reviews can be actually positive about their products and they can start building mindshare. I'd happily replace my wife's rx6600 with a $300, 4070-level of performance Arc card if it just works. If I can go Radeon or Geforce with similar performance at the same price, I see no reason to pick Arc.

[@bmqww223]

you can see in geekbench with "Intel(R) Xe Graphics RI" there are 3 benchmarks for the g21 , it is definetly not going for rtx 3060 ti performance, since it already trades blows in the metric with rtx 4060 and we hve to be mindful how amd is bad with raytracing and intel already had a good base to improve upon, navi 44 with 128 bit bus surely be very much handicapped in that segment but it can fight it in price due to the microscopic die size , so at the end of the day we pc gamers will finally get mainstream affordable cards.......price and the raytracing performance would decide the results because with lunar lake we already see how battlemage trade blows with the top end of the rdna 3.5 offering and lets be honest rdna 4 is a bandaid to rdna 3 , with just better raytracing performance we won't see drastic raster performance......and if they can release before rnda 4 , then they can put a huge dent on amd since the rnda 3 has a oversupply of card and they are not even selling as fast with the price drops given how shitty this generation has been , they will be left with a pile of rdna 3 cards where they can't just come out and release rdna 4 this year.....

[@bmqww223]

you can see in geekbench with "Intel(R) Xe Graphics RI" there are 3 benchmarks for the g21 , it is definetly not going for rtx 3060 ti performance, since it already trades blows in the metric with rtx 4060 and we hve to be mindful how amd is bad with raytracing and intel already had a good base to improve upon, navi 44 with 128 bit bus surely be very much handicapped in that segment but it can fight it in price due to the microscopic die size , so at the end of the day we pc gamers will finally get mainstream affordable cards.......price and the raytracing performance would decide the results because with lunar lake we already see how battlemage trade blows with the top end of the rdna 3.5 offering and lets be honest rdna 4 is a bandaid to rdna 3 , with just better raytracing performance we won't see drastic raster performance......and if they can release before rnda 4 , then they can put a huge dent on amd since the rnda 3 has a oversupply of card and they are not even selling as fast with the price drops given how shitty this generation has been , they will be left with a pile of rdna 3 cards where they can't just come out and release rdna 4 this year.....

[@MaxIronsThird]

@@Violet-ui G31 has 16gb of Vram though, Intel can't sell the card at a loss

[@Sean-km7ht]

With ARC, I really hope they keep producing ARC generations, but mainly products in a similar class as the A310. I want to be able to throw an affordable add-in card that brings the latest generation of QuickSync and the encoders that Intel does very well, with dedicated video memory, on systems with older or AMD CPUs. I don't see them being particularly competitive in the gaming segment. With a tile design and already sharing the Xe cores with the iGPUs, it should be fairly cost effective to make a card using the current iGPU and NPU tiles.

[@1xXNimrodXx1]

I hope INTEL finally bring ECC to cosumer boards and cpus, then there would be no need for those add in cards and you save money and powerdraw but yeah. currently I use AMD CPU with ECC + A310 for transcoding

[@Sean-km7ht]

@1xXNimrodXx1 I'm doing the same for my TrueNAS, but I'd still kinda like to see it to allow upgrading to newer encoders without a full rebuild and likely platform change.

[@godnamedtay]

@@Sean-km7ht arc made them no $. Therefore arc is essentially dead, just like he said. They might have something for battlemage, but I have heard from several people that much of it has been canceled as well. He is 100% right imo, but we’ll see. Intel has canceled SO many projects as of late, it’s laughable.

[@knabbagluon]

They have to. Who will buy APUs without a good GPU?

[@fatjawns3671]

A310 is great for NAS systems as well

[@raikoh05]

If intel stops developing gpus it will be their own mistake again

[@blahorgaslisk7763]

Just like killing off Beastlake was their own mistake. "We don't need high performance cores anymore." was something said at the time. All so they could spend the money on AI cores instead.

[@erichrybinczak8894]

Unifed cores like strix will be the way. Good cpu, with offloading to a gpu or npu for tasks. Npu will start picking up lot more of the lower performance cpu tasks. A pure cpu is no longer the processors we need for home market. for servers and high performance computing is different.

[@knabbagluon]

@@erichrybinczak8894 I think so too. And the thread director could be a key here to give tasks to different specialized tiles. But a good GPU is still key.

[@qweasdzxc]

@@erichrybinczak8894 what lower performance tasks will npu pick up?

[@RetroPaul6502]

Glad to hear ARC hasn't been canceled yet. They need to pair up with AMD like CPU has done.

[@TubeYouDSilvy]

Griffin cove - 14A for Intel will be going up against presumably Zen 6 on TSMC 2nm or some variant. I think the real question is going to be power efficiency and whether Intel comes close delivering their roadmap on time - because it’s incredibly likely that TSMC and AMD will be reasonably on track.

[@tringuyen7519]

Intel 14A is a pipe dream. Intel has already delayed ordering more high NA EUV machines from ASML which means 18A & 18A-P is already delayed!

[@TubeYouDSilvy]

Hi-NA not used in production for 18A - but I agree Intel as a viable/going concern fab can never be taken for granted

[@avatarion]

Zen 6 is probably N3X.

[@lightward9487]

not, 14A Griffin killing zen7

[@DreadyBearBoi]

@@tringuyen7519 That’s absolutely not what that means whatsoever especially considering the tools they’re using for that they already have bud bud. Not to mention 18A is not high NA, 14A is. Also if you’re talking about the ASML headline they detailed Samsung as the main factor not Intel.

[@Aaron-mn2ro]

4070 performance from their second go around isn't terrible at all, and I am sure the price would have been great

[@Ziplock9000]

It is when you need to compete

[@Pro720HyperMaster720]

@@Ziplock9000if it has a good enough price and specially drivers which I would even think is the most important to reviewers and people thinking on recommending on a budget. Then you have a card that is worth to buy, if they hit 4070 performance I’ll buy one, I hope drivers continue on a good trend. I’m not expecting the most massive success but I really would like to turn out better than Alchemist and sell enough to continue the Dedicated GPUs. I don’t think Intel should close Arc as Tom suggest and use the money for something else. I would vote to constrain the ARC GPUs to the entry level/mid like what battlemage will do with 1 or 2 dies like it is doing, to at least keep it alive and not fall too much behind so they are in a good position if they are economically better in the future. Worst case scenario I would cancel the Celestial die, focus all the resources on the architecture, so the Tiles show the best possible jump (and maybe releasing the Tile as a standalone discrete very entry level GPU) and then on Druid or the next arch E-whatever come back again to the discrete GPU on entry and mid range, basically getting the best possible architecture advances while on life support

[@kevinerbs2778]

@@Ziplock9000 Why when Nvidia's making a 12Gb 5070 for $700??? it's a terrible card, terrible price since it's really a 5060/ti.

[@hellcoreproductions]

Second go around? Do you realise how long they have been in the graphics market?

[@darkfire3691]

@@kevinerbs2778 Oh you already know 5070 price and performance, good for you. And you probably already know intels performance and price. Not even announced cards too.

[@MrQuashu]

I'm still hoping for Battlemage, we need some competition. If they will have 4070 performance ( that's it's quite good enough to play comfortable 1440p ) and good price they might sell better.

[@e2rqey]

Yeah no. Especially since you could just buy a used 4070 at worst. Or one of multiple other options

[@TheDarksideFNothing]

Hard to be price competitive when you're shipping significantly more silicon for the same performance as your competitors. I'm crossing my fingers too but don't expect much.

[@jaynorwood2]

Intel's integrated GPUs will get gaming support because they are still around 80% of the laptop market. A second reason is that they are very competitive in matrix processing. They are reportedly upping the integrated GPU core count by 50% on Panther Lake. I think the increased gaming support will save their GPU program just because of the stability.

[@TheDarksideFNothing]

Yeah they will keep on existing for laptop integrated. Our best hope is they manage to do well enough there to warrant spinning back up their dGPU development.

[@CyberneticArgumentCreator]

@@TheDarksideFNothing Spin what back up? It hasn't been canceled...

[@TheDarksideFNothing]

@@CyberneticArgumentCreator Most of the dies were cancelled. They absolutely gutted their GPU division. They're basically on life support and it looks like most of what they're doing is in support of GPU tiles for their laptop producs. Look at the old road maps... We aren't getting anything even REMOTELY close to that. So yes, much of it has indeed been cancelled and there is much to "spin back up." What I mean by "Spin back up" is returning to actually having a full lineup and trying to be competitive in something other than entry level or low end.

[@NicolasChanCSY]

If you can go back in time, would you use another phrase instead of "effectively cancelled", Tom? Don't get me wrong, I understand what you mean because I have seen you explaining it again and again and again. But I still think that if you used "Arc (discrete) is getting lowest priority" or something, you could have saved yourself from numerous explanations, because not everyone watches all your related videos, instead of news outlets quoting you partially, before they enter a discussion about Arc.

[@Cadriella98]

I wonder how dropping e-cores will affect their server roadmap. Im guessing the e-cores will continue in some form as their zen-dense competitor. Think that removing hybrid architecture will save a lot of software engineer time though. And maybe they can bring back avx512

[@lharsay]

The real competition of E cores were high core count ARM based servers gaining ground over both Intel and AMD in certain applications.

[@prem3548]

If it launched at the start of this year, then it would have honestly been a pretty good $450 and $350 cards. With RDNA 4 though, they are gonna have to be $350 and $200 cards.

[@Psychx_]

g31 below 400USD would actually be a sweet deal. Much better price/perf than we had for years. If the GPU got positive reviews and the drivers became more solid than now, it could grab quite some marketshare.

[@MechAdv]

If they sold a 4070 equivalent card with 16 GB for 300$, they would move units. If RDNA 4 lands at the high end of leaked performance with raster on par with a 4080, and the low end of pricing at 500$, then a we can reasonably expect a 300$ card from the RDNA4 lineup to match it, and Intel is cooked. Again.

[@Violet-ui]

The 7700 XT is 350 and only about 10 percent slower than the 4070 BMG G31 would need to be 299 to make sense by the time rdna 4 launches

[@prem3548]

@@Violet-ui OOF you are right. I hope at launch Intel Drivers are good enough to not cripple its performance. If it has good drivers, better RT, and 16GB of VRAM then I can see it being a good card at $350

[@NoSpamForYou]

@@prem3548 They changed the way Battlemage works so it should have good compatibility without custom day 1 driver hot fixes for every game. Sadly still only 32 XE cores on G31. 56 or 64 would have been a beast for ray tracing and AI / rendering.

[@ivantsipr]

The real problem with Intel is not the dgpu business, on the contrary that's where they should continue working, Alch was a dumpster fire but had way better ray tracing than anything AMD has, in Battlemage it can only get better, if they continue they have a real chance on been really good in a couple generations, now AI that's what Intel should be careful, that's a area that Nvidia and AMD have a huge heads up so if Intel insists in getting into that race is a huge gamble, the AI bubble might burst before they have anything built and they would have just wasted a lot of money on nothing.

[@mytech6779]

I would like a GPU card for use as a general accelerator for hobbyist scale HPC.and intel seems best positioned to make such a thing with their oneAPI stack (Incorporating their SyCL implementation and the intel C++ compiler, both are open cross-platform standards, among other things.). AMD hip-ROCm is a dumpster fire as far as the lack of dependable commitments to product support timelines or even clear statements of current product compatibility, just bad management in that department. CUDA is dependable even on 10 year old cards but it is proprietary and drivers are not well supported beyond Windows. AI focused accelerators are all about low precision for high quantity (FP4 FP6 FP8) to save memory bandwidth while most scientific HPC, and related stuff like CADD rendering/simulation, is focused on full single or double precision math and some integer work (FP32 fp64 int32 int64). Graphics seems to be about single or half precision fp32 and fp16 and ray tracing matrix math.

[@Viking8888]

Not having intel in the GPU game really sucks. Having good intel competition was supposed to bring the cost of graphics cards down to sane prices, or more sane at the least.

[@patrickweaver1105]

That was never going to happen. People will buy from whichever company makes the fastest and most expensive cards. Because they're insane.

[@Viking8888]

@@patrickweaver1105 I guess you have to be on a strict budget after saving for years to be the type to buy the best you can afford regardless of brand.

[@Demonoid1990]

The bulk of GPUS people seem to be buying are midrange, they tend to have good price to performance. I don't think Intel could remain relevant if they're stuck to the low end budget zone. Who knows maybe Intel can let ARC simmer in the background and try again later once they get back on their feet financially.

[@arenzricodexd4409]

The flaw with that thinking is for cheap price to happen material and labour coat also need to be cheap. The last 2 quarters JPR report that intel market share is at 0%. That's because intel only ship extremely low quantity of card. For intel they rather not selling those card and lose market share rather than have to sell those Arc below the cost.

[@bruhirl1023]

Well theyre leaving cause they just cant compete

[@Miley_00]

I love my A770! it does everything I want in 1440p 165hz! I hope the newer drivers will push it even further!

[@TheGuruStud]

Bot

[@RafitoOoO]

enjoy it while it lasts

[@Anukinihun]

I'll believe all of this when I see it.

[@ze_chrome]

Sad that the higher one got canceled 4070ti/4080 perf for 500-600 bucks would have been game changer

[@Roodoca]

Isn't that what the top end Navi 48 / RDNA 4 card suppose to deliver though, that performance and pricing? Just from AMD not Intel, don't see the big deal. Hope FSR 4 with AI, launches with the RDNA 4 and not ~8 months later.

[@mckinleyostvig7135]

​@@Roodocahardware encoding. The ARC GPUs do a much better job competing with NVENC than anything AMD does. If there was an ARC that had much more raw power it would be massive for content creation which right now pretty much requires NVENC for a single PC/GPU setup.

[@ze_chrome]

@@Roodoca Xess looks much better than FSR to me and Intel competing with high end Nvidia GPU for half the price would force both AMD and Nvidia to lower prices and release much better prices . Yes AMD could do smth like that , but more actors mean more competition, more competition leads to lower prices ( often )

[@RafitoOoO]

Just wait for the RX 8800 XT then.

[@ze_chrome]

@@RafitoOoO More competition, less expensive GPUs . Seeing a battle for mid range could be amazing for the regular gamer

[@conza1989]

G21 should do 16GB, if they want to be really competitive and low pricing, shaving it to 12GB, but there's also bandwidth considerations, hmm.

[@Stars-Mine]

ram amount is based off of the bus width. you don't just go with less ram unless you want to demolish bandwidth. Bus width is one of the most expensive things in terms of die space, so someone at the architecture level thought that die required it for performance, and the die space was worth putting in the bus rather then more cores.

[@lharsay]

192bit bus= hex channel, you can't put 16GB in hex channel mode. Anyways it will compete well against 8GB RTX4060 and 4060Ti cards.

[@Hobo_X]

The battlemage cards seem like promising performance uplifts, even if they won't be very competitive it's a good increased over Alchemist. If they can keep that pace then by Druid they may actually have a reasonable choice at certain price points. If they just commit to not bailing out of the competition too soon, their software/drivers will get mature enough to _eventually_ pay off

[@edwxx20001]

the problem is they are catching up, at the end of a generation, with a more advanced node, to midrange. it feels to me like they would need to hit a home run on their own nodes, A18/A14 and make the cards on that, using a node advantage to catch up with Nvidia in real time. its alot of moving parts that have to hit at the right time, then you just need to get all the drivers to work right and still sell it at a 20% markdown from NVidia to get people to buy it.

[@Hobo_X]

@@edwxx20001 My point is its not realistic to expect to instantly be competitive with nvidia and AMD right away. Intel needs to plan this as a long term investment not short term

[@edwxx20001]

@@Hobo_X Yes! they need to continue to work on this. my point is they aren't going to have a chance to catch Nvidia unless they have some silicon advantage that Nvidia doesn't have. That means either find a way for TSMC to drop Nvidia, unlikely but not impossible, or have the Intel fabs hit a home run and design the cards on that. Which is why they need to BOTH get the fabs/nodes on time, but also design the chips on time as well. its a hard target, but its the only way I can see intel making headway in the space. AMD has had struggles in the gpu market, but they have carved out the lower-cost price to performance segment. they look to be focusing on that next gen with cards that will not compete with the top end, but have cheaper silicon and lower price points in comparison to NVidia. Intel's best chance was for AMD to try and push to beat Nvidia leaving a market segment open for them. this isn't happening, so intel at the moment dosen't have any easy market segment to target like they did when Alchemist was originally meant to launch. Tough times at intel, but I believe they can right the ship, and if the balloon ever goes up in China, having fabs stateside will save the company.

[@evilhunter64]

Whew lad. Early on this one some good watching at work

[@FilmMonger]

Intel GPUs for creator software are incredible bang for buck. Update after update, the A770 reached RTX 3080 performance levels, beating it and the 3090 on some workflows. It offers better, faster encoding/decoding than Nvidia and AMD, and across more codecs. Not only that, but it allows you to have Intel Quicksync support on AMD CPU machines. If Intel really markets these for creators and gets some big partners to market and bundle with, think Adobe (in which the card excels), DaVinci Resolve (which can leverage the winning aspects from Arc GPUs, along with an AMD's or Nvidia's computational prowess to deliver even better performance), or even Avid, (which might help bring its costs down, a driver in productions moving away from it), and I think they will have something beyond viable. Maybe not the way gaming GPUs are, and they might have to stick to just one model for some time, but remember that for years, Nvidia Titan GPUs were pretty much the only option for professional production. If they find the partners necessary to embrace this, I think that at the very least, the project can stay alive and allow behind-the-scenes development to catch up more in gaming performance.

[@LprogressivesANDliberals]

Proud owner of both founders edition of battlemage and all models of sparkle intel gpus 💯❤️ all you can do is hope they succeed

[@vigilant_1934]

You mean Limited Edition Alchemist right (Nvidia calls their reference cards Founder's Editions not Intel or AMD) Unless you somehow got Battlemage before it's been released or you're trolling idk what you're talking about.

[@MrHav1k]

Would be smart if they based this unified core on the E-core design tbh. The Skymont (Chadmont lol) E-cores are punching way above their weight and are giving Raptor Lake like IPC with a quarter of the area. Would be great for Intel's margins if they could base their future designs around that. Would be way better in perf/watt server too. Just a thought.

[@nempk1817]

"with a quarter of the area" what about the ISA supported?

[@mewmew5143]

e-cores sucks

[@quantumdot7393]

​@@MrHav1k tests made on lunar lake show these e cores having like 75% the IPC of the big cores. But in lunar lake they don't have access to the L3 cache so I want to see if they perform better in arrow lake . But they can't really clock high so overall performance is like half . Also I would say they are a bit bigger than quarter the size. more like 30% .Either way you are not wrong. The p cores need to be stronger given the area they occupy.

[@hellcoreproductions]

Scaling is a thing, IPC is only a one part of a design, something at a certain power envelope and transistor budget often doesn't scale to higher frequencies, you can make something in a certain power window work really well but it won't ever clock well etc . There are reasons for Intel making monster e-core chips for servers but not client and big e-core counts on client CPUs really not being that impactful for a lot of actual client workloads. Desktop workloads are still very stuck in a tall vs wide dilemma, while servers live in the very wide domain.

[@klyplays]

@@nempk1817 raptorcove and skymont has ISA-parity in Mainstream components, otherwise they wouldn't need to laser off AVX-512 in Goldencove

[@Dudewitbow]

if battle mage was indeed a 4070 competitor, its sad they missed the bus had it released earlier in the year. a 4070 card for ~500 but had 16gb vram would have at least been a decent option vs AMDs 7800XT and the 4070 due to Xess' slightly better image quality vs FSR, and is less burdened by RT than AMD at a given price point, but would have the 16gb vram availability vs Nvidia's options(which require to be upsold to 800$ for the 70 ti super, or downgraded to the 60 ti)

[@quademasters249]

If the prices is right and it outperforms my 3080, I'd consider buying it. Most of the market is in mid-range GPU's. You just need to check steam charts.

[@vigilant_1934]

@@quademasters249 For it to outperform the 3080 it would have be quite a bit stronger than a 4070 since your 3080 is already faster than that.

[@ChairmanMeow1]

My trust in Intel is still totally broken. Gonna need at least one generation of chips with no issues for me to consider them ever again. Im gonna let others be the guinea pig for Intels Core line.

[@NetflixForeign]

Intel didn't help matter with new chips only being able to run Windows 11 instead of 10.

[@zee-fr5kw]

maybe stop watching doomer youtube channels, didn't even know intel was "bad"

[@pf100andahalf]

@@zee-fr5kw There's a big difference between watching doomer youtube channels and being completely oblivious to what's going on as you seem to be. Intel 13th and 14th gen cpu's are degrading as I type this because they pushed them too far with too much voltage, heat, and power just so they'd beat AMD. Then Intel refused RMA's for almost 2 years and finally admitted there was a problem but wouldn't do a recall but I hear they're accepting all RMA's now. I won't buy an Intel cpu for a few generations myself because I don't trust them now.

[@oggilein1]

yep, says something when I congratulate someone for going with intel 12th gen over 13th or 14th

[@jazeenharal6013]

​@@zee-fr5kwSo, you've been completely unaware of about 70% of their entire lineup being prone to degradation due to the Vmin voltage fault? In what world part of the internet have you managed to miss the biggest CPU news of the last 30 years? "Doomer channels" lol...being completely uninformed doesn't make you some kind of sage, it just means you haven't the first clue what's going on

[@conza1989]

Just back on, Intel GPUs, for long term prospects, I'm glad they're trying to pursue the big money in datacenter and AI, because, if they can crack in there with a decent and stable market share, that will make gaming GPUs from reverse binned(?) dies, an easier and more competitive proposition. I appreciate the workloads aren't the same, but that's the same problem that AMD and Nvidia sort of have - AMD are doing this, they're doing bare bones minimum for the desktop DIY consumer, phoned in poor marketing etc etc, their trying to increase their economies of scale against Nvidia - I hope Intel go this route too but I hope they're successful so we have a better gaming landscape in 3 or so years (for the Nvidia 6000 series roughly maybe in 2027).

[@conodigrom]

Dear kernel and scheduler developers, dear compiler guys, thank you for writing stuff for the E-Cores sparing us the work. Now mark them obsolete. - Sincerely, Intel

[@Musikvidedo]

You're aware that there's a pretty big overlap between the groups "kernel devs working on intel scheduling", "compiler devs working on intel compilation" and "intel guys"? Edit: also don't forget that intel definitely aren't the only ones ever making heterogeneous chips.

[@conodigrom]

@@Musikvidedo Yes I'm well aware of that

[@samson7294]

9:27 all due respect, Tom! Don't tell me how to spend my money. Not that it matters but I buy Arc for collection purposes.

[@MooresLawIsDead]

Hey I got an A770 for that reason too! (I may have cut me saying "it's fine for collecting" just to make it flow better)

[@Lokotraktor]

Is that a train horn in the background? :D 🎉

[@pedrohenrique-wb3ed]

2025 is looking Bright for casual buyers of hardware

[@Voidkitty_]

Tom this is the 7th time you've leaked arc is dead to the class this week

[@lharsay]

Well it doesn't seem like store shelves are gonna be flooded with with Asus, Asrock and MSI made high end Arc cards any time soon.

[@mika2666]

Would be amazing to see a grand return to single core performance

[@bozilla9975]

I read the title as "Coffin Lake" and thought, how fitting.

[@hellcoreproductions]

In his own words, Jim was cancelled before Pat got there, he wanted engineering excellence and Intel wanted someone to be a good boy for the finance team. Funny how that is one of the many things that helped messed them up. They chose doing business short term over doing good business. Further more, it seems Pat is basically going for more of the same BS, not surprising given how he worked when he was on the exec suite before he left Intel last time. A little reading on pretty much anyone vs Intel in the courts for the past few decades shows what kind of operation he was part of.

[@Lemard77]

The 18A architectures are going to use backend power delivery? that should be quite good for efficiency

[@rabb1tjones921]

Good job!

[@Fk8td]

Battlemage having 20gb of ram will work well for creators on a budget.

[@daysgone3950]

Thank you. Lots of exciting info in this one.

[@MageeMC]

I really hope that Intel has a competitive path forward. That would push AMD to start pressing on the gas again. They haven't been exactly coasting, but neither have they been pushing limits.

[@bastiangugu4083]

I hope that Intel is able to achieve their goals with these architectures. We need competition or we end up in the same pickle we were in before Zen 2.

[@Miparwo]

One Intel neuron says "double hyperthreading!", and the other neuron says "scrap hyperthreading!".

[@edwxx20001]

all three have potential. dropping hyperthreading can reduce power, increase single tread performance, and reduce core size on silicon. it only costs one thread. on the other hand you can spend more power, reduce single thread performance, increase the silicon size of the core, but get 4 threads instead of 2 per core.

[@lharsay]

So would you rather have 24 real cores with no HT or 16 cores with HT?

[@mytech6779]

Intel had 4 way HT from 2010 to 2020 on the xeon-phi "Knights …" accelerators. With AVX512

[@edwxx20001]

@@lharsay It depends on workload, if you want ultra high frequency and ST performance, drop HT and optimize the core for it. If you need more and more cores to do many smaller tasks, 4x hyperthreading is a possible route. the intel E-cores accomplish a similar task, while the compressed Zen-c cores also do the same thing. It comes down to the right silicon for the right task, but it seems like the big 2 have both moved from 4xHT to a different solution

[@cr4zyg047]

Easily my favorite channel on KZbin, you've come so far!

[@NetflixForeign]

Royal Core sounds amazing at the beginning, like another Ryzen moment.

[@blinspeed]

I have completely lost track of intel's laptop and desktop architectures lmao, every day I hear some new "Reaper Cove" style name.

[@cajampa]

What i really want to know is when we will get DDR6. Because all of these cores are starting to get heavily memory limited. Or even better they start offering on package faster memory support.

[@grospoulpe951]

Thanks for those good news :) Good competition = good products :)

[@devindodge8648]

What is your take on ARM and AMD's apparent "openness" to experimenting with new software? I would love to see AMD finding a competitive edge against CUDA in AI (and Ray tracing) just to keep Nvidia on their toes a little bit. Could Nvidia become some new competitor with their own O.S.? It feels like compatability is such a limiting factor in evolution.

[@nickbarkas5774]

When is intel going to offer desktop/mobile avx 10.2 cpus?

[@aaronlaw7981]

i am satisfied with my a770 16G. the hardware/value ratio is hard to pass. the updates occur weekly. if I was intel i would announce longer update support. they planned/built it to be more competitive than it was. probably still alot of room for optimization. i plan on buying flagship battlemage. the future probably will be accelerators facilitated either by arm or x86 cores so intel can't afford to exit an accelerator race.

[@FirestormX9]

That's just ridiculous

[@HexerPsy]

Its shame royal core 2.0 isn't coming. It sounds like it would ve been fun to play with and tune. Intel has tackled hybrid designs and solved the scheduling issue. 6+0, 4+8, 2+16 ... I like tinkering with hardware - just wondering how you would go about tuning that, I have so many questions! Ultimately, I think splitting and fusing cores may have come at a serious latency penalty. And scheduling might've been a nightmare... Intel wouldn't just need to show their manufacturing is reliable again, they also need to cut power, and support their sockets longer. AMD now let's you buy in on first gen in a socket, and three gen later you buy the best thing that came out on your socket with just a CPU upgrade. Its hard to beat that strategy.

[@TonyMasters-u2w]

It looks like AMD and Intel are focusing on AI rather than Gaming nowadays that's why we have 2% generational increase...

[@haukionkannel]

AI is much more important so… ofcourse they put money where they can get most money back!

[@TonyMasters-u2w]

@@haukionkannel idk, i spoke to my colleagues, nobody knows why we suddenly need AI, looks like an artificial importance

[@Sir_Rift]

I really hope we start seeing Intel GPUs primarily compete in the SFX space. The low profile A380 is a fantastic value for a SFX computer. I plan to use it as an HTPC and bring it in my camper as well. If they made an 8gb+ low profile card, I'd buy it in a heartbeat! I'm even gonna get the 6gb A380 and upgrade later because it's 3 times cheaper than the low profile 4060 8gb, so I don't have to wait.

[@oleksandrbilonoh2137]

To be price competitive, Intel should have used Intel fab. Moving to TSMC was a mistake. As we see in Arrow Lake, TDP saving is not huge and probably more comes from arch/frequency changes.

[@MooresLawIsDead]

IFS is uncompetitive.

[@jamegumb7298]

If Battlemage is legit good, I will buy it. I fully expect Nvidia as well as Intel and AMD to design cards for compute, and sell the not perfect die off as desktop gpu. Or at least get AI/compute cards with minimal alterations to cut cost in the design.

[@Cooldude1991c]

Thank you for giving useful information

[@RafitoOoO]

I guess now we know why Pat was CTO for almost 10 years and never got a promotion to CEO lol.

[@benjaminlynch9958]

He was CTO at a time when Intel was kicking the absolute shit out of AMD.

[@qweasdzxc]

Is Nova Lake on LGA 1851?

[@Tarvoskemwer]

Everyone will still be cautious to be early adopters, Intel have placed themselves in the "we'll believe it when we actually see it"-category. With any new Intel product I'd wait at least 6 months to see what problems arises. Does it break, does it underperform, is it a space heater?

[@AugmentedGravity]

Doubt it but it would be nice to see some real competition again

[@trevoro.9731]

Are they going to make one more defect generation of Ice lake cores with insane L1 latency of 5 clocks ?

[@lolmao500]

To me all those predictions of IPC gains years before even a single silicon exists is kind of stupid... even the engineers working on it have zero idea how its gonna come out.

[@benjaminlynch9958]

The predictions aren’t created out of thin air. The engineers design the chip using EDA software where they get a very good idea of performance, and then they make a sample (usually just 1-2 cores, which is fine for measuring IPC) using FPGA’s. It’s not any more ‘stupid’ than civil engineers calculating the load limit of a bridge before a single brick has been laid or concrete poured. There’s a science behind what they’re doing, and they’re not just pulling numbers out of their ass for public consumption. Remember, all these IPC leaks are meant for internal use only and aren’t supposed to see the light of day publicly. But the figures are shared widely enough internally that inevitably someone leaks them to people like Tom.

[@mytech6779]

IPC gains are pretty easy for them to simulate. The trouble is the achievable frequencies and voltage limits. Aside from that, they are not years away from engineering samples, they are only years away from finished retail packages..

[@tsclly2377]

Hyper-threading probably is the reason for cancellation.. I think that this concept will be revived with an integrated AI GPU set for the consumer inferencing (spy in the die). A 6 core with L2 cache, chiplet L3, GPU (probably dual) with the tensor like core all on the die, chiplet L3 and shared hybrid or separate DRAM.. I still can see e-cores (limited instruction type) in a limited function to run the feedback through USB, Ethernet as long as it has adequate L3 and does not impact the main DRAM bus.

[@mikeb3172]

Where are you getting 5-13% IPC for 18A... it's the first backside power delivery (PowerVia) product so they can jack the clock speed right?

[@mytech6779]

It would help if you bothered to lookup the definitions for any of the words you use.

[@Psychx_]

If Intel prices Arc competitively (sub 400 USD), especially the G31 card, then battlemage may actually be a compelling deal. 4070 performance is good enough to comfortably run all games at non-potato graphics.

[@GustavoNoronha]

17:04 that is all really exciting, but the numbers you were being told for Arrow Lake implied a very high IPC increase that did not come to be; you were also being told great things about MTL before that, so it's hard to know if your Intel sources are being able to get good information. We can hope, but at this point I would lower expectations compared to this.

[@fullerfit93]

Honestly I think that Intel should keep going with how they did with lunar lake. SOC based computing. Low end 4p 4e/4cu-8cu (ultra 3-5) and high end 4p 8e/12-16cu GPU (ultra 7-9).

[@JusticeGamingChannel]

It's not up to me to tell Intel where to spend their money, it is up to them. If they want to make a competitive GPU, I'm all for competition, I actually hope they continue and try to push for it. If, however, they decide not to, that's entirely on them. You should also never ever buy a product, even if its bad, just to support a company, that definitely sends the wrong message. So don't just buy something up, unless it actually provides a good value for you, and works well.

[@Ivan-pr7ku]

Cool! The "benchmarking" E-cores are going bye-bye.

[@lorenzwiedemar1837]

These videos about future intel cpu's increasingly sound like an nature documentary straight from old David Attenborough.

[@Speak_Out_and_Remove_All_Doubt]

Is the rumour still that Nova Lake's CPU tile will be produced using TSMC's 2nm? Or now that Intel is hopefully back on track at it's fabs will it use in-house 18A or even 14A?

[@henrythegreatamerican8136]

Any leaks on what is going on with their Foundries?

[@AlfonsoASecas]

16:32 I thought the sound of the train was near my house haha

[@josephmazor725]

I’m starting to be cautiously optimistic

[@jonogrimmer6013]

Cool very interesting! Hope intel comes back for competition to push down prices

[@neti_neti_]

Any news about RTX5070Ti 18GB 280W GPU ?

[@BusAlexey]

It's a 5070, not 5070 Ti though

[@robertbolzicco9995]

Zen 7, were looking for into the filter here.

[@bankaimaster999]

Wait is there any performance news for the iGPU on the Ryzen Ai Max+ 395? Is it really going to be good or at a 4060-4070 mobile dGPU?? Will it truly have up to 96GB or VRAM???

[@nempk1817]

Thats almost confirmed with leaked transport documentation(the VRAM amount atleast). I think he recorded this video before the "new" leaks about the naming

[@halrichard1969]

Thank You Tom, for yet another informative video. I have to tell you that for many years now I have viewed Intel as more and more like the Budweiser of CPU Manufacturing. That translates to "just an average CPU with tons and tons of great advertising". Intel doesnt have a Mulvaney moment but they have alot of little moments when they delay and then delay again on their launch dates, and or come out with a Paper Launch on top of it. The CPUs having some kind of fatal flaw or problem. NO THANKS!

[@nuurnwui]

In which factories will these be made? TSCM or own?

[@benjaminlynch9958]

The dGPU’s discussed will be made at TSMC; the CPU’s will be made internally at Intel.

[@Daniel-zr4pk]

why would they stop using E cores. they started using them like 2-3 years ago

[@lharsay]

It seems they want to beef them up to the point where they become more dense P cores with lower clock speeds.

[@Dangerman-zg3ui]

Isn't Razor Lake after Nova Lake? Raichu on Twitter leaked it.

[@sebbbi2]

Battlemage has more traditional GPU register file and wider SIMD architecture similar to every other modern GPU design. And they solved various HW bottlenecks in modern workloads. This is a completely new GPU design. Much bigger changes than single AMD or Nvidia generation. Bigger than GCN to RDNA. And it performs admirably in Lunar Lake, even at very strict TDP budget. Intel has finally a modern GPU architecture. I am sad that they aren’t pushing it harder to dGPU space. Intel already has pretty good drivers for Battlemage in Lunar Lake. I don’t think drivers are the correct excuse as they must do the driver work already for their Battlemage iGPUs in Lunar Lake and all their forthcoming laptop chips. AMDs RDNA iGPUs are using pretty much the same driver as their dGPUs too. iGPU vs dGPU is mostly difference in GPU core counts (and external VRAM). But Battlemage is late and Intel can’t afford to sell dGPUs at zero profits to grow their market share. Intel brand isn’t as valuable anymore. And ARC hurt their brand regarding GPUs. They would need to outperform competition and price it more aggressively at the same time to reach financial success, which is a problematic combo. It tool Intel way too long time to get ARC into the market and Battlemage is already late. I guess they fear similar failure again. But their new architecture is much better this time. It’s a shame they aren’t willing to even try.

[@RylTheValstrax]

The lack of finalization around griffin cove doesn't exactly inspire confidence in that 2027 timeline though... We're already at the end of 2024 and it doesnt sound like Intel is close to deciding what goes in yet.

[@SigmaHuman69]

Arc A380 from SPARKLE is a 1080p streaming/recording beast

[@Toksyuryel]

Before Intel can kill Zen 7, they first need to figure out how to kill Zen 4

[@jantube358]

I think Battlemage should be supported from a gamer perspective because the only alternative gamers have to Nvidia and AMD dGPUs are AMD iGPUs. Remember when GPUs were totally unaffordable for gamers during the pandemic and during the crypto hype? There could be another crypto hype and then a 4060 will cost $1000 again.

[@PohnnyRico]

will intel get there because of a ram upgrade?

[@Mark_Williams.]

15:14 - Removing some instruction sets etc. Might that mean this is when they expect X86S to hit?

[@J0xxe]

What you actually leaked was that Arc was axed after alchemist, multiple times. Guess now you can repeat that after battlemage?

[@BeSome1]

I've seen a lot of people suggest that if you're spending more than $1k on a Nvidia GPU, you should wait until 50 Series. However I am currently in the market for the 4080 super and it doesn't seem like anything is really going to compete value-wise in the first quarter of 2025. I'm on the AM5 platform w/ Rzyn 7600X & Rtx 3060 12gb. Opinions?

[@nostrum6410]

but how will it stack up against zen 8, because thats likely what its up agaisnt

[@minusinfinity6974]

Ah yes two cpus we know absolutely nothing about and one will kill the other according to Intel leakers. But yesterday Intel had no answer to Zen 6 so needs Panther Lake S apparently as a rushed response.

[@bugleboop]

will those new chips run on 1851 socket or will people have to go out and buy new motherboard for nova lake?

[@nempk1817]

the chances are a new mobo

[@bugleboop]

@@nempk1817 the fact they haven’t said anything regarding desktop panther lake and nova lakes chip set speaks volumes.

[@andersjjensen]

Until Intel says otherwise (and they haven't communicated anything yet) expect 2 generations per socket. 14th gen was a needless excursion into telling shareholders that Intel was doing "something" and should be considered a fluke (or an insult).

[@ItsAkile]

GPU and CPU projects should not be competing in that sense where a revolutionary CPU Project would compete with a potential mobile decent GPU, AMD it not that amazing at GPUs that I believe Battlemage should essentially be Intel already catching up to them in ipc/arch. Intel focusing mainly on the iGPU isn't even a bad thing nowadays, lots of uses for this if they ship alot of CPUs in the near future.

[@jaynorwood2]

Use of BSPD has been compared to two generations of EUV shrinkage. It's a big deal.

[@mentalplayground]

Of true, yeey for competition.