Intel issue might be a fabrication level defect, something to do with oxidation

https://youtube.com/watch?v=gTeubeCIwRw

>I boughted a 13700k last November :marseycry:

>It has never been more over


Leaks

  • leaker at "Intel Customer" timestamp

    • Over 8 million 13th gen CPUs possibly affected

    • Actual failure rate of 10%-25%

    • No information on 14th gen products

  • Other leaker timestamp

    • Expected to affect units from March of 2023 through April 2024

    • Infos

      • Fabrication issue where anti-oxidation coating is improperly applied

      • Intel working on microcode to decrease frequency, will not fix root cause but might work around it

  • Leaker 3 timestamp

    • Reducing max frequency for boosting was able to work around the issue?

    • Documents saying customer is purging its inventory as a result of issues

  • Allegedly leaked documents timestamp

    • Change to officially supported ram speeds DDR5-5600 reduced to DDR5-4800 ignoring XMP
  • List of affected customers includes hedgies? timestamp

  • Intel claims 0.035% failure rate in messages with OEMS timestamp

    • "This is in conflict with the OEM we spoke with which said 25%-50% failure rate" :marseyxd:
  • Leaker - "Either Intel is lying to us or they don't know the real failure rate. Until last month, they reported to us that 10% of their [production] was still having the 'oxidation' issue" timestamp

  • Multiple sources - Intel is beginning what it calls "Vendor Remediation" for OEM customers timestamp

  • "Medium-sized system integrator" timestamp

    • "We reduced out [harder to pass] failure requirements because of concerns of degradation. We're currently failing 12% of Intel CPUs during intake QA."

    • QA deets in this - certain tests are failing more often, this is why different companies are failing different %% timestamp

  • OEM source - considering limiting turboclocks to 5.4-5.5GHz to limit RMAs timestamp

  • General Platform Instability + Voltage timestamp

    • Microcode update could fix this?

    • The T series CPU failing doesn't make sense with this since it's low voltage or something

    • Potential memory Speed update timestamp


Root Cause

  • Root Cause according to leaked document timestamp

    • "The root cause of this mechanism is due to a random defect mode in the fabrication process of the Raptor Lake CPU during the via formation steps which could cause high resistance vias due to oxidation"
  • Possibly affected processors timestamp

    • Not copying all these down but even the 13600k(f) and 13700T are hit :marseyxd:

  • Start of Intel's duplicity, some quotes from customers and a quote from a Failure Analysis lab timestamp

    • Some details about ALD and how it works, possible failures that can happen during it
92
Jump in the discussion.

No email address required.

AMD is kept afloat purely by intels incompetence

Jump in the discussion.

No email address required.

AMD has legitimately excellent tech on the CPU side, they've been killing it since Zen2

black lives matter

Jump in the discussion.

No email address required.

They were dead cpu side literally till like 2018 lol no one could've seen zen being so good after they absolutely fricked bulldozer

Jump in the discussion.

No email address required.

You could tell as soon as the benchmarks for the first engineering samples of Zen 1 came out. The chiplet design was a god-tier idea, and it was their first iteration on a brand new core microarchitecture.

Jump in the discussion.

No email address required.

And AMD cheekily naming their chipset X399 right after Intel announced X299 was fun

Jump in the discussion.

No email address required.

Man Bulldozer sucked so much :#marseycringe:

To think they tried selling that shit for like 6 years

Jump in the discussion.

No email address required.

Athlon was so good bro... Who could've known the 8370 house fire wouldn't perform?

Jump in the discussion.

No email address required.

You could already tell when the benchmarks for the server variants of the bulldozer chips were released. There was a lot of cope that the desktop variants would be able to close the IPC gap with Intel via high clocks, but Global Foundries FD-SOI process didn't get them the clock speeds they were hoping for, and the FX-8150 was just a variant of those chips. It also didn't help that Intel's Sandy Bridge chips came out just months prior and had massive IPC improvements that were totally unexpected by AMD. Bulldozer was so bad that it had lower IPC than even Phenom 2. IIRC some lead engineer in the Bulldozer architectural team also died before tape-out, but I don't recall the details on this.

It was AMD's Pentium 4 Netburst architecture :marseysad:

I think it's neat that the process technology gap has flipped between Intel and AMD, and that AMD spinning off Global Foundries might have been the one winning move the company made under Ruiz' leadership (with everything else he did at the time being r-slurred)

Jump in the discussion.

No email address required.

First gen Ryzen was, in retrospect, pretty shit beyond the core count. Single thread performance was still far behind Intel, could've been up to 40% in some workloads, the cross CCX communication was horrible, tons of platform issues especially with RAM... It took until Zen 2 for them to catch up in 1T performance, even then they lost in games.

Zen 3 and Zen 4 especially are very solid though.

Jump in the discussion.

No email address required.

You still have cross-CCX issues, but it isn't as bad because the core-to-core latencies are a fair bit lower between them and the memory bandwidth is quite a bit higher. First gen Zen also had issues with achieving high memory clocks due to their kind of shitty memory controller, and the fabric clock was tied to the memory clock speed, so those issues were exacerbated. Extra L3 in Zen 2 + I/O die and 8 core CCX in Zen 3 helped quite a bit.

The CCX issue isn't totally solved, though - you can still get serious scaling issues especially on the really high core count SKUs because a lot of programmers are r-slurred and don't understand that there's serious performance overhead in constantly moving data between L3 cache domains.

Windows 10 for example has a performance bug in the kernel where they zero 4k memory pages across as many threads as the system has, which is terminally r-slurred. It doesn't even make sense because it's not a compute heavy operation, so I don't know why they're using more than a few cores at most to do it:

https://randomascii.wordpress.com/2022/07/11/slower-memory-zeroing-through-parallelism/

Jump in the discussion.

No email address required.

I recall price being a big win for Zen 1 though. Cost per core (real cores this time) was dirt cheap and shocked intel out of the 4 cores 8 threads for $600 coma it had fallen into

Jump in the discussion.

No email address required.

Price per core was good, especially right on launch when Intel only had Kaby Lake. In the mainstream segment, you got 3x the thread count (4c4t with i5 vs 6c12t with R5), those R5's aged in many ways better than i5's of that era, doubly so when you consider upgrade path. Coffee Lake launched like half a year later and was much more competitive though.

Jump in the discussion.

No email address required.

Early R5s were 4c8t but overall yeah, it was and kinda still is Zen's strength

Jump in the discussion.

No email address required.

Only R5 1400/1500, 1600 was by far the most common and was a 6 core, around the i5 pricing of $200

Jump in the discussion.

No email address required.

Oh, I didn't remember that

Jump in the discussion.

No email address required.

The Ryzen 1700 was a great value (cheap and you could OC it to near 1800X perf in many cases), and the TR 1950X destroyed Intel's Extreme series parts at a much better price as well. :marseythumbsup:

Jump in the discussion.

No email address required.

Strange that Intel are still behind after having been the obvious choice for a CPU for years beforehand

Jump in the discussion.

No email address required.

Link copied to clipboard
Action successful!
Error, please refresh the page and try again.