Intel Thread

[majestic]

Hey, the 14900KS is here!

Well, and no one cares, except maybe for overclockers. For them the extra money for (more or less) eliminating silicon lottery might be worth it.

[majestic]

Wendell: "It's weird, if there really would be a 50% failure rate, people would be climbing the walls."

Intermittent issues with no rhyme or reason and no obvious connections and no microcode updates from Intel as a fix. The only reasonable* explanation for that behavior is a manufacturing defect leading to silicon degredation under load, and the lack of more widespread outrage over failing CPUs in game servers might point to it being localized. Wouldn't be surprised if the affected CPUs come from the same fab (maybe even the same lithography machine). 50% of all CPUs going belly up would be noticable, and not even Intel could keep a lid on that.

Intel should be able to trace the issue with batch numbers and shipping manifests. The caveat here being that this is all based on Wendell looking through crash dumps and server logs. Wendell mentions that in the video, that is sample size is not exactly fantastic.

*Well, there's the explanation that Intel is selling unstable bins for the heck of it, but that seems unlikely, as Intel cannot afford to alienate server providers, and the CPUs pass checks when first activated and then eventually degrade under load. I could see management ordering laxer binning standards for the consumer market, but not for CPUs shipped to companies. Too much at stake.

Edited July 13 by majestic

[Humanoid]

6 minutes ago, majestic said:

Wendell: "It's weird, if there really would be a 50% failure rate, people would be climbing the walls."

Developer Alderon games, who may or may not be one of the companies discussed here, claims an almost 100% failure rate.

[majestic]

30 minutes ago, Humanoid said:

Developer Alderon games, who may or may not be one of the companies discussed here, claims an almost 100% failure rate.

That's probably what Wendell was talking about, also regarding his sample size and how he got the information. It's an indie dev allegedly experiencing problems with every single one of their CPUs in every area of use. Many YouTubers, including GamersNexus, use 13/14900K based systems for video editing and rendering have done so for years now. Whatever is going on at Alderon's can't affect all CPUs, even Wendell's analysis of the data suggest 50% - which is insane enough. 's a bold statement that could bring Intel's litigation hammer down on them, although if even half of what they describe is true it is understandable.

It could, however, affect all CPUs shipped to them assuming they buy from the same supplier (which is a reasonable assumption, even for smaller companies) and they got a batch with a manufacturing defect. Anyway, from that statement:  

Quote

Over the last 3–4 months, we have observed that CPUs initially working well deteriorate over time, eventually failing. The failure rate we have observed from our own testing is nearly 100%, indicating it's only a matter of time before affected CPUs fail.

Well, it would make sense that 100% of all affected CPUs fail. That wording is very unfortunate.

Edited July 13 by majestic

[Zoraptor]

Probably difficult to tell how widespread it is when one of (the?) most common errors for gaming rigs made it look like a video card issue.

Very interesting that the issue happens even when locked to 125W by server MBs.

[majestic]

Surprise! It was/is a manufacturing defect, plus a bug in voltage control microcode. 

Intel's communication and response has been atrocious. Manufacturing defects can happen, but it looks really bad when not responsing properly.

[Bartimaeus]

It's one thing if they're just unstable and the problem can eventually be fixed, but it kiiind of sounds like all these problems are contributing to permanently damaging two generations' worth of CPUs, which is, uh, not great. I guess we'll see when Intel rolls outs its updates in August to see if all the already unstable 14th generation CPUs continue to be broken, or if they're actually able to salvage at least that. Regardless, Intel's communication and transparency has been pretty terrible.

[Bokishi]

Yeah I haven't had any issues yet, but jeez this is a big L for intel. I will probably undervolt just to delay the inevitable and go AMD next round

[Zoraptor]

On 7/24/2024 at 9:55 PM, Bartimaeus said:

It's one thing if they're just unstable and the problem can eventually be fixed, but it kiiind of sounds like all these problems are contributing to permanently damaging two generations' worth of CPUs, which is, uh, not great.

Seems to be confirmed now (via non official channel, but not denied by Intel) that the damage is permanent once it occurs, and no microcode/ BIOS or other update will fix it.

[majestic]

Seems logical. The actual problem is a manufacturing defect, patching the faulty voltage regulation microcode might prevent affected CPUs from ever becoming unstable in their normal lifespan (that remains to be seen though), but it can't patch out physical damage.

[Humanoid]

"Normal lifespan" probably being the warranty period.

[Malcador]

https://www.cnbc.com/amp/2024/08/01/intel-intc-q2-earnings-report-2024.html

Yowch

[Bartimaeus]

Yikes.

As much as I would heartily laugh if Intel were to ever fail entirely, it would be quickly followed by some very grave contemplation. But hey, if AMD survived [insert unending horde of problems they faced for many years here], I think Intel will be okay.

Edited August 2 by Bartimaeus

[majestic]

Doesn't bode too well for consumer graphics, more than anything else. Something tells me Pat Gelsinger will have a hard time convincing shareholders that Arc is worthwhile in the face of that earnings call. Who knows, perhaps I'm wrong and Lunar Lake is going to be so good that it encroaches on AMD's SoC monopoly. Meteor Lake sure didn't when you look at The Claw's performance. I want to be wrong because really, we need some competition in the consumer graphics space, but eight ball says outlook not good.

[Humanoid]

To be honest I just want something, anything, to disrupt the de-facto monopoly they have in the premium laptop space.

[Bartimaeus]

Some synthetic benchmarking of the new micro-code that supposedly fixes the power degradation issues (scroll down): https://tweakers.net/reviews/12320/hoeveel-trager-worden-intel-processors-door-de-nieuwe-default-settings.html

tl;dr: Typically a small performance dip (5-10%) versus prior benchmarks, occasionally more significant, occasionally less or even nothing at all...also, occasionally actually better! Mildly reduces clock speeds across the board, reduces power draw significantly in the 14th gen (but weirdly not the 13th). I hope chasing top performance no matter what was worth all this, Intel.

(e): Actually, I misread: this doesn't seem to be the new micro-code that just came out (which Asus was the first to release as a beta version of earlier today, and these tests were done with a Gigabyte board anyways), but rather the cumulative micro-code changes that Intel has already made over the past year. Still useful information to know given initial benchmarks of these chips versus benchmarks you might see if they were tested today, but I guess hold onto your horses to see what the latest will actually do. Guess that's what I get for seeing "Asus releases new BIOS beta with Intel's micro-code changes" and then "these guys re-tested and compared 2023 to latest bios update" right after each other and making assumptions from that.

Edited August 8 by Bartimaeus

[majestic]

Some videos in the interwebs from channels that claim to have "sources" inside of tech companies did not age all to well. Like this one from MLID:

Well, what do you know. Manufacturing defect/microcode bug in voltage control/stability issues aside, the 13900K rebrand turned out to be a much better Zen 5 competitor than expected. *snort*

[majestic]

AMD: Here's Zen (notquite)5%!

Intel: Hold my beer!

Granted, if even half of the claimed power savings are correct that'll be interesting, but unless Intel can't into 1st party benchmarks, the incentive to upgrade to Arrow Lake just went away.

Unless the 265K is proportionally better in comparison than the 285K, which I am not really interested in anyway, the only upside would be going back to air cooling's much better idle noise. Not gonna lie, that pump is annoying as f.... when it is quiet in the room.

 

[Sarex]

44 minutes ago, majestic said:

Not gonna lie, that pump is annoying as f.... when it is quiet in the room.

Why I went back to the Noctua NH-D15. But nowadays reviews are pretty good about pump noise, so you can find something decent.

[majestic]

On 10/10/2024 at 9:20 PM, majestic said:

AMD: Here's Zen (notquite)5%!

Intel: Hold my beer!

With testing this time:

Well, that is underwhelming. Guess that's one generation I'll be sitting out then. Especially since there are rumors now that the other LGA 1851 CPU generations have been scrapped. Pity, I was looking forward to Arrow Lake, but that gaming performance is just, uhm... not good, and I really don't need the productivity gains, and as Steve puts it, even with gaming as a full time job, which I don't do obviously, it will take years to get the price difference in with the lower power draw.

edit: weird though, looking at the released benchmark scores on other reviews, the single thread and multithread performance of Arrow Lake in e.g. Cinebench outclasses everything by a more than decent margin, it just translates into no gains or even worse performance in gaming. Guess that makes Cinebench and other synthetic benchmarks either worthless, or something else is not quite right. Bizarre, at any rate. Especially that performance drop in Cyberpunk 2077, where it is slower than a freaking 12600K.

edit2: In der8auer's German video, he gets completely different results from Hardware Unboxed with Cyberpunk, where the 285K is behind the 14900K, but still ahead of the 9950X (and obviously behind the 7800X3D, but that much was to be expected anyway).

Edited October 24 by majestic

[Bokishi]

Haha skipping this

[Bartimaeus]

7800X3D:

The beatings will continue until morale improves. The prices will continue to get higher until Intel improves.

[majestic]

I don't think I've ever seen such a steep divide between synthetic benchmarks and real world application and gaming performance*. Cinebench, 3D Mark, Geekbench, whatever you pick, the Core Ultra 285K is either on par with AMD's best or dominates the charts, and when it comes to actually performing, it falls way short. Except for productivity workloads, but even there, how can so much single thread benchmark performance lead to such terrible Photoshop real world performance?

Guess Intel does struggle a bit with glueing their CPUs together. Going to be interesting to see how Zen 6 will shape up, as AMD is also switching their way of glueing CPUs together.

1 hour ago, Bartimaeus said:

7800X3D:

The beatings will continue until morale improves. The prices will continue to get higher until Intel improves.

According to rumors, if all goes well, Nova Lake will come out with an additional cache tile. Intel will call that LLC (Last-Level-Cache) and is planned to basically be Intel's version of 3D V-Cache. Roadmapped for late 2026/early 2027. It'll be a while before prices for the X3D CPUs drop, there's just no incentive for AMD (or retailers) to do so.

*nVidia cheating with driver side optimizations when synthetic benchmarks were detected nonwithstanding. That can't really apply here. I mean, I hope there's nothing in Arrow Lake's microcode that detects if 3DMark is running just to produce better performance. That would be weird, even for Intel.

Edited October 24 by majestic

[Bokishi]

22 hours ago, majestic said:

According to rumors, if all goes well, Nova Lake will come out with an additional cache tile. Intel will call that LLC (Last-Level-Cache) and is planned to basically be Intel's version of 3D V-Cache. Roadmapped for late 2026/early 2027. It'll be a while before prices for the X3D CPUs drop, there's just no incentive for AMD (or retailers) to do so.

Just in time for GTA 6 benchmarks 

I can see where they're going with the better efficiency, but having no/worse performance over last gen kinda makes this DOA

[majestic]

19 minutes ago, Bokishi said:

Just in time for GTA 6 benchmarks 

I can see where they're going with the better efficiency, but having no/worse performance over last gen kinda makes this DOA

Intel has two primary motivations here, one is fighting off ARM's encroachment on their laptop market with efficiency gains, and the other is to be able to slap more cores onto workstation CPUs to take the wind out of AMD's Epyc. Still, it's a massive disappointment that the Lion Cove p-cores of Arrow Lake apparently can't even match the Raptor Cove p-cores. At this point in time it would have been better to slap Raptor Cove p-cores with a node shrink on Arrow Lake and call it a day.

Point in case being the upcoming 288 core Sierra Forest (the 144 core variants are already available), although those still use the older Crestmont e-core architecture. Arrow Lake has the new Skymont e-cores that are largely the reason why the 285K can compete with the 9950X in heavily multithreaded workloads even though it has 8 threads less and the p-cores are, well, let's say, clearly not doing so well compared to the old Raptor Cove architecture.

Skymont e-cores have roughly the same performance as Raptor Cove p-cores at the same clock speeds (they just clock lower, obviously), i.e. Intel's IPC gains on their e-cores are massive, I wouldn't be entirely suprised if future Intel CPUs are just going to ditch the p-cores because the e-cores are going to pass them in performance. The team developing the e-cores is clearly doing something right, and the other teams aren't. We might be looking at another Core 2 moment in a not very far off future.