Q&A with Wendell @ Level1Techs: Intel's Stability, AI PC, Q&A

/ level1techs
-----------------------
Need POTATO merch? There's a chip for that!
merch.techtechp...
more-moore.com : Sign up to the More Than Moore Newsletter
/ techtechpotato : Patreon gets you access to the TTP Discord server!
Follow Ian on Twitter at / iancutress
Follow TechTechPotato on Twitter at / techtechpotato
If you're in the market for something from Amazon, please use the following links. TTP may receive a commission if you purchase anything through these links.
Amazon USA : geni.us/Amazon...
Amazon UK : geni.us/Amazon...
Amazon CAN : geni.us/Amazon...
Amazon GER : geni.us/Amazon...
Amazon Other : geni.us/TTPAma...
Ending music: • An Jone - Night Run Away
-----------------------
Welcome to the TechTechPotato (c) Dr. Ian Cutress
Ramblings about things related to Technology from an analyst for More Than Moore
#techtechpotato
------------
More Than Moore, as with other research and analyst firms, provides or has provided paid research, analysis, advising, or consulting to many high-tech companies in the industry, which may include advertising on TTP. The companies that fall under this banner include AMD, Armari, Baidu, Facebook, IBM, Infineon, Intel, Lattice Semi, Linode, MediaTek, NordPass, ProteanTecs, Qualcomm, SiFive, Supermicro, Tenstorrent, TSMC.

[@420bobby69]

Wendell is the man. As a 13900k owner, I really really appreciate him doing the digging and going the extra mile to bring awareness to the issue. Mine isn't catastrophic, but it IS "quirky" sometimes and I've been noticing more quirks as time progresses. And I've been manually setting stock PL1 and PL2 long before any of this got covered.

[@ABaumstumpf]

You should be first looking into all the voltage-settings cause those are what is damaging the silicon. Running the CPU at 200W for hours each day, 365 days a year, will not cause problems. Running it a 90°C with +150mV overvoltage will.

[@420bobby69]

@@ABaumstumpf Yeah, I'm aware. I just wasn't gonna go deep into my BIOS setup lol. I run a -50mv undervolt, ASUS MCE disabled since day 1. The chip still hits 90c when I'm pushing it in Blender but that's "to be expected." You kind of can't avoid it with anything less than direct-die. Also, to be frank, I'm not sure anyone knows what is causing problems at this point.

[@leeloodog]

i have 2. One is perfect and one is defective. They made it kind of hard to work though the problems so i gave up. But ideally they'd just swap them out with something that isn't defective. it's annoying.

[@josef.hernandez9781]

@@ABaumstumpf Have you not seen Wendell's video? Even running below 200w at 85c max WILL damage the chip based on the data he received from game server companies.

[@ABaumstumpf]

@@josef.hernandez9781 i have seen it. And i have also seen that those boards too run high voltages.

[@wolpumba4099]

*Summary* Here is a bullet list summary of the video with starting timestamps: *Intel Stability:* * *[**33:21**]* Wendell's video on Intel stability has gained attention, with game studios confirming issues. * *[**34:49**]* Wendell has access to error report databases from multiple game studios, including sanitized data. * *[**39:02**]* The errors appear on various Intel CPUs, including KS, KFs, and regular models, across multiple generations (12th, 13th, 14th). * *[**42:39**]* Data suggests potential fundamental design or firmware issues, not solely related to motherboard partners. * *[**46:54**]* Concerns about accelerated electromigration due to possible programming errors leading to degradation. * *[**1:34:04**]* Wendell received "troubled" Intel hardware for analysis and disassembly, aiming to pinpoint the root cause. *AI PCs & Snapdragon:* * *[**04:21**]* Microsoft pushes Qualcomm, AMD, and Intel to enhance on-device AI performance for "co-pilot plus," aiming to reduce cloud reliance. * *[**20:30**]* Snapdragon X Elite performance falls short of initial promises, especially in the high-power 80-watt reference design. * *[**07:14**]* Developer experience with Snapdragon is frustrating due to fragmented tools, APIs, and lack of standardization. * *[**29:06**]* Microsoft's involvement appears to prioritize software functionality over hardware performance, causing issues. * *[**27:38**]* Concerns about Qualcomm's ability to establish a robust developer ecosystem and handle software challenges. * *[**29:17**]* Linux support on Snapdragon devices is in development but availability is unclear, possibly delayed due to devkit issues. * *[**18:34**]* Comparing Snapdragon's launch to early Ryzen notebooks, both competitive but lacking refinement in the initial stages. *General Discussion:* * *[**55:46**]* Lack of standardized benchmarking for NPUs makes it challenging to compare performance across different hardware. * *[**58:14**]* Cinebench is considered an inadequate benchmarking tool, particularly for mobile and AI workloads. * *[**45:40**]* Concerns about using consumer-grade hardware (W680 motherboards) in demanding server environments, despite potential benefits. * *[**1:16:48**]* Discussion about the longevity of high-end CPUs, advocating for prioritizing stability over chasing marginal performance gains. * *[**2:11:05**]* Exploring the use of AI and machine learning in chip design (EDA) and its potential to optimize performance and accelerate development. * *[**2:16:50**]* GlobalFoundries shifting focus to "legacy" nodes, driven by increased demand for ancillary silicon in devices like smartphones. *Personal Anecdotes:* * *[**2:18:36**]* Wendell shares his experiences with early dual-CPU systems and excitement for powerful hardware. * *[**1:22:23**]* Ian talks about the closure of his undergraduate chemistry department and potential video documenting its history. * *[**1:49:47**]* Both discuss navigating the challenges of mergers and acquisitions impacting their businesses and clients. *Closing:* * *[**2:23:44**]* The live stream was unplanned and exceeded initial expectations, encountering some technical difficulties along the way. * *[**2:29:01**]* Both creators emphasize their gratitude for viewer support and reiterate the importance of their primary jobs outside KZbin. i used gemini 1.5 pro to summarize the transcript

[@Oz-gv5fz]

Thanks 👍

[@daikucoffee5316]

This is the real „big boy“ press conference

[@korbensc7218]

This was one of the most informative YT videos I saw in a long time. Thank you guys.

[@Groovewonder2]

The thumbnail on this is gold. I love it.

[@killercage2]

Wendell is a great guest! Hope to see many more vids with him in the future!

[@Strykenine]

Excellent thumbnail.

[@AshtonCoolman]

This is one of my favorite thumbnails of all time

[@mentalplayground]

33:21 Intel Stuff Perfect Potato with level1 Love it

[@w0nd3rlu573r]

Excellent. I am confident Win10 was my last dip into MS products. When I got the one game I am playing to work under Linux, then I really felt I was breathing easier.

[@Kurukx]

Love when you guys chat

[@bobbybobman3073]

I feel like we need a lot more collaborations between these two ... Two Power nerds getting to power nerd together is cool.

[@mrtuk4282]

That was amazing :)

[@eltamarindo]

Sound gap at 1:43:03.

[@JS-kr7zy]

ends at 1:45:00

[@AdmV0rl0n]

BP6 and Celeron 300a's - That brings back fun memories. Great choice!

[@FloppaAppreciationSociet-ds7zf]

34:02 Why has it not been fixed in 14th gen? Because 14th and 13th gen are literally the same silicon, they even have the same steppings. There was no actual "refresh", only a name change.

[@erkinalp]

and a bit of product binning; don't expect any big improvements from Intel until they finalise the x86s (64-bit only x86) specification.

[@FloppaAppreciationSociet-ds7zf]

@@erkinalp We have no idea how long will it take for Intel to finalize x86S, and even then there is no guarantee we'll ever see a widespread adoption. The absolute fuckton of legacy software especially in the enterprise world is like the main reason we haven't all switched to ARM yet, x86S might end up being the worst of both worlds. In the meantime, Intel hasn't given up on producing new x86-64 architectures, Meteor Lake was actually a major architectural revision of the Golden Cove architecture, even though a failed one, as IPC is actually worse than before. Arrow Lake and Lunar Lake are supposed to come with new core architectures as well.

[@johnnytex1040]

38:58 Wait, so even the locked processors are failing?

[@punjabeeplaya]

I'm a simple man, i see Wendell, i click and watch

[@magfal]

As a 13980HX owner I hope this will be figured out.

[@TheEVEInspiration]

Almost all processors have microcode, down to the first x86. Not saying the capabilities have not been expanded however.

[@chtubbie]

Indeed. What the Pentium fdiv bug brought us was field-updateable microcode.

[@eltamarindo]

I love the building that they are in. It was always my favorite place to stay when Siggraph was in LA.

[@JJFX-]

I'd like to see the VID tables for the chips that start having problems. I think the elephant in the room has been staring at us this whole time. What's the difference between RL i5, i7 and i9? Voltage. To hit max boost clock a 14900K can request over 1.5V on the worst bins. The i7 is a bit lower and the i5 rarely even goes over 1.4V. This same voltage gets applied to P-cores, E-cores, L3, Ring, etc. Now imagine certain transient events causing notable spikes, sending another 100+ mV across the ring. Thermals would only make the situation worse but that's likely not as much of a factor for servers. I think voltage and design changes combined with node issues are causing these parts to slowly consume themselves faster than Intel predicted. A design flaw may accelerate the problem but at the end of the day I think it comes down to angry pixies bouncing around the die.

[@CompatibilityMadness]

If I had to bet on something, the primary cause of all this is Intel's voltage specsheet, that lists "1.6-1.7V" vcore as "safe maximum voltage" that can be utilized on Alder lake CPUs. When such high voltage is applied by [Auto] set VRM LCC at ANY point, it WILL degrade the core by very very tiny amounts for few miliseconds (maybe less). In such conditions you will get dead/damaged chips eventually (it's simply a matter of time). Being a dumbass in making a voltage specsheet will bite Intel for a long time to come. I wouldn't be surprised if voltage meant was 1.27V instead of 1.7V. Transistors are meant to be run on slightly lower voltage each time a technology makes them smaller (at least that was how it worked for 350nm - 90nm stuff). There are clever ways to bypass this (like better materials), but ultimately - you can't run current gen CPUs with voltages from early 2000s (note : 1.7V on Vcore was utilized as default voltage for Pentium 3 "Coppermine" generation [180nm process]).

[@ABaumstumpf]

"the primary cause of all this is Intel's voltage specsheet, that lists "1.6-1.7V" vcore as "safe maximum voltage" that can be utilized on Alder lake CPUs" That is not "safe maximum" but the absolute maximum ratings - which for any electronics part is not "safe" but the maximum it is ever allowed to see before having an instant permanent damage. The voltage-specs them self actually are not that bad - for a 14900K the maximum it requests from the BIOS would be ~1.41V - and then the boards decided that by default that should not be the 1.41V on the VRAM that the CPU requested but 1.45V at the CPU-pin - resulting in a "nice" >100mV overvoltage.

[@CompatibilityMadness]

@@ABaumstumpf Q : Does the guy programming BIOS (or makes V/F tables for Intel CPUs), knows that ? Because there are motherboards that yeeted voltage to just shy of 1.55V at Idle (one of them can be seen on video : "Shorter version of my EXPERIMENTAL fix" by Buildzoid).

[@ABaumstumpf]

@@CompatibilityMadness pretty sure they do - and then marketing and management say they should ignore that.

[@CompatibilityMadness]

@@ABaumstumpf Probably... About "the absolute maximum ratings - which for any electronics part is not "safe" but the maximum it is ever allowed to see before having an instant permanent damage." part : As far as I know, Intel NEVER specified at which voltage CPUs get insta-killed. However, if that was true for "maximum voltage rating", ALL CPUs from 7-th gen would instantly die or be permanently damaged (assuming non-POST here or IMC being incapable of working at rated speed), at just over 1.52V applied on Vcore. So, did "Operating voltage" change definition between 2017 and 2024 datasheets ?

[@FloppaAppreciationSociet-ds7zf]

Yeah, but I think Wendell did mention in his video that Alder Lake CPUs are not affected, only Raptor Lake. Something that his cloud gaming contact told him they never had such issues with their 12900K servers.

[@Pavlobot5]

I got a Surface RT too. Man its a great device with absolutely no support. Just let us install linux alread

[@WirrWicht]

I once had a lot of trouble convincing my superiors that a bug reported by one of our customers had to be a hardware failure, since the software and project-related data made it impossible to reach the point in the code where it crashed. The memory configuration looked like sometimes memory would run out and someone had found a loose memory module in a drawer and stuffed it in.

[@tremelai]

That looks like the Westin Bonaventure Hotel

[@DirkDjently]

thanks again for the Q&A! does anyone have a link to that "the war gaming blog post" Dr. Cutress mentions at 1:03:36 ? I've been trying to find it and google is not coming up with anything 🙃

[@WayStedYou]

Probably warframe by digital extremes

[@DirkDjently]

@@WayStedYou yep, that's the one. Thank you!

[@miinyoo]

They are in Wonka's factory and every so often a Charlie is freed into the stratosphere.

[@danielabbott9312]

Towards the end before I mailed mine back to Intel I was experiencing crashes just sitting at the desktop with random spikes up to 1. 8v at 5ghz all core on a 13900k The Chipettes never been pushed hard in an overclock or anything like that or experience thermal throttling and any significant way

[@jemo_hack]

That looks like Bonnadventure in L.A….. I think I saw a guy with the same shirt about two days ago… >D I f I would have know, would have stepped up and said hello! >D

[@SamanthaVimes177]

What time stamp is the 13/14900k discussion?

[@SamanthaVimes177]

33:15

[@ceasingendssnakeden7323]

im on my second 14900KS...had to do insane tuning to even get the RMA to remain stable. LLC AC/DC settings manual voltage tuning with offsets and changing cpu switching frequencies DRAM settings all kinds of stuff but no crashes in 1 week now @ 253 pl1&2 350a 56s getting 40k R23 scores @72 max temps.

[@abitler9445]

Should I use the 1.5tb octane drive as my main os drive vs something like a 990 pro

[@cracklingice]

first overclockable xeon? I'm running an E5 1660 V3 overclocked to 42x

[@TechTechPotato]

I meant to say first top of stack. People ofr ages asked for the 15/18 core overclockable

[@cracklingice]

@@TechTechPotato Oh yeah - if the 2600s were overclockable, I might be content staying on X99 a bit longer but with being limited to 8 cores - I've been debating X570 Taichi with 5700x3d since it's fairly reasonably priced way to get dual 8 lanes at least so I can have a GPU and 25gbe nic. I'll probably upgrade to a 16 or 18 core if it's worth doing so when I move the X99 to NAS. It's already got 256GB of registered ECC - fully functional too since it's on a Taichi.

[@WillFuI]

When did Paris use North America wall plugs?

[@blegi1245]

If you are at a pre briefing for a product launch you can't say you are there until the embargo lifts. It's amd zen 5 launch media pre brief btw.

[@WillFuI]

@@blegi1245 I know, They know, we know. Some people have posted pictures already.

[@robertmiller1638]

Maybe it’s in Paris Tennessee. Population 10,000.

[@WayStedYou]

When they cant say they are in LA for zen5

[@donnydarko7624]

you are in LA

[@Joniyah444]

The ring, the cpu cache speed and memory controller too hot and degrading ?

[@kevikiru]

The camera at about 1 hour 50 min has an annoying pulsating focus.

[@slim5782]

Please you two, start a podcast!

[@dirg3music]

1:59:45, Oh my God I think Wendell may be the literal only person in my entire life I've ever seen who does this too! I've always just memorized people's numbers instead of saving them under a name. Lmao. Feel very seen right now

[@squisherderheld]

How many decades does Microfsoft need to get the simple "unboxing" experience right? Most people dont buy a laptop as a tech gadget that needs permanent care-taking. It should be as simple as possible, take it out of the box, turn it and start using it. You only have once chance for a first impression. You cannot blow that. Don't sell shit that isn't ready for a general audience. They own Todd Howards now, so why doesnt it just work?

[@mrtuk4282]

One question please, why do you think AMD doesn't put 3DV Cache on both CCD's for the 9950X3D ????

[@WayStedYou]

Because it costs extra money and interchip latency negates any benefit vs having one use extra cache

[@egl12fen]

@@WayStedYou I'm not sure interchip latency is really a problem here... As I recall, in initial Zen 3 reviews, the 5950x and 5900x were the fastest Zen 3 chips when it came to gaming and those cpus also had dual CCDs. Any interchip latency seemed to be a non-issue, at least with those chips. The addition of 3d v-cache to the second ccd for the current Zen chips would most likely solve the issues with game's thread scheduling AMD was/is having with the 7950x3D and 7900x3D, as the problem here seems to be whether or not a game (which can benefit from the extra cache) is running on the 3D v-cache equipped CCD. Looking at the month to month Amazon sales average for the 7800x3D, AMD is seemingly happily selling an octo-core 3D v-cache equipped chip today for roughly launch price of Intel's 6 core 8700k from 2017. As for cost, they just charge you a little more. IMO, no one shopping for 12-16 core cpus in the $500-$700 price range would balk at paying 5%-10% more to get the best of the best from AMD with no 3D v-cache scheduling headaches. 🤷‍♀

[@eurocrusader1724]

Glad I've avoided the intel disaster..

[@Fractal_32]

32:01 to answer this question I had a neighbor who is going off to college for engineering I strictly told them “don’t even look at Qualcomm/X elite chips, buy AMD or Intel.”

[@woolfel]

recall is crap. I have zero plans of up grading windows.

[@chillidog8726]

Where can I find the wargaming pie chart

[@WayStedYou]

Digital extremes warframe not wargaming who make world of tanks and warships

[@NdxtremePro]

I wonder if Intel tried to use AI to design the chip?

[@1samm1]

There's a section in this video about their move to standard design tools, and those do include AI features (historically heuristics, now machine learning)

[@NdxtremePro]

@@1samm1 Yeah, I caught it at the end.

[@grizius4123]

... It might be actually the case

[@chromerims]

Willjum (U.K. gamer with 1.45M subs) probably is not using Ryzen: kzbin.info/www/bejne/oqXUfKOnba1mf6Mm20s Kindest regards, friends and neighbours. (P.S. Players of *Rust* can change their character's clothing but not any anatomical features as individually assigned day 1 by the game dev, FacePunch. So whatever your character looks like at purchase, that's what you play with. Forever. Not mod-able. That is why Willjum's narrating voice does not seem to match his Rust in-game character.)

[@StAlessa]

The problem is: the CPU are NOT getting ENOUGH voltages! The Power AC and DC load-line settings in the Bios are changing the voltages that the CPU gets, thus not providing enough voltages at certain frequencies under high loads. I can run the 12 gen CPU on Load line calibration LLC 1 fine, but the 14 gen needs LLC 8 to achieve higher frequencies because there is too much V droop at lower Load-line settings, and the CPU does not get enough voltages under higher loads.. 14 gen needs more power and voltages under high loads.. So Intel and Motherboard manufacturers just need to fix their Load line CPU Microcode and Bios settings ... And it is cool that intel lets us change the load-line settings so that we can under-volt the CPU, but if the under-volt is too aggressive crashes, errors, or instability happens. i tested this, and even the "out of video Memory" error was not present when i just was sending a bit more voltage to the CPU, but it was present if i was sending to little voltage. Apps and games were not starting or crashing with too little voltages, but not crashing anymore with a bit more..

[@TheEVEInspiration]

I disagree with the update cycle analysis as on the OS side Microsoft is actually KILLING the upgrade cycle. No-one in their right mind wants to go away from Windows 10 to Windows 11 for example. That is for many years now people not buying new PC's etc....just because of Microsoft. Them pretending to drive update cycles is delusional. Another company that is choking the consumer facing industry (instead of pushing it forward by innovation) is NVIdia with their pricing and deliberate hardware limitations, like undersupplying DRAM for inflated prices. It is costing everyone headaches and money and a future.

[@shanent5793]

So you just let him say that Intel invented microcode?

[@TechTechPotato]

I didn't say that

[@shanent5793]

@@TechTechPotato I fail to see how my comment implies that you said anything. Separated by a common language!

[@mitlanderson]

​@shanent5793 he never said Intel invented micro code, he said Intel was the first to make it updatable after the fact

[@saturnFIV3]

I never realized there were AMD cpu fanboys until reading some of these comments. Tribal allegiance to any corporation is whack af, especially if they don't even employ you.

[@Testbug000]

At least in the past, long ago, it was more anti-Intel Fanboys, and AMD was the only other option. Dunno how it has changed