GPUs & CPUs & Enthusiast hardware: Questions, Discussion and fanboy slap-fights - Nvidia & AMD & Intel - Separe but Equal. Intel rides in the back of the bus.

Just gonna take this little tl:dr from elsewhere:

"
1.) Out of 1584 game decompression errors during the 90 day data retention period across both databases, 1431 of the errors are on Intel 13th or 14th gen CPU's. Second place (after 13900KS/K/KF and 14900KS/K/KF) is the i7-9700H with only 11 instances. AMD CPU's only had four crashes in total in this dataset.
2.) Crash rates occur on average every 2 hours of gameplay for affected users.
3.) Frequency of crash rates increase over time. (silicon or something else is degrading?)
4.) This is not due to overclocking.
5.) Datasets from game servers using either Asus or Supermicro W680 chipsets (no overclocking support, more conservative designs) show similar crash rates.
6.) Crash rates are also increasing over time on the W680 game server datasets..."

Oh ffs Intel, clean yourself up. Probably fucked something up getting even more desperate to compete.
 
It's also interesting it's only the 13900s and 14900s doing it, and not the 700s.

Makes you wonder what changed that could cause it.

Regardless, it's a really bad look for Intel. Those data centres would easily buy CPUs by the hundreds.
 
  • Thunk-Provoking
Reactions: Vecr
Imagine buying an intel processor post-2023 lol lmao.
 
Here's the Raptor Lake CPU to buy:
Intel Refreshes Dual-Core Raptor Lake With New 310 CPU, Just Two P-Cores At 4.1 GHz

I think the implication is their fabrication process has gone to pot.

Simple enough lets just all use TSMC chips, fabless intel here we come
If it's only affecting the 13900K/14900K variants, then it doesn't seem like it.

Lunar Lake is an all TSMC creation, but it had better not become a trend or Intel is in trouble.

Imagine buying an intel processor post-2023 lol lmao.
They still tend to be the best deals on the refurbished market because of volume and competing more at the low-end.

I'll pay $200 for a 12900T system, no problem. After a couple years on the market, it will be known if the chips are safe or not.
 
  • Thunk-Provoking
Reactions: Vecr

Level1Techs Wendell seems to think this is a proper issue
I like this new Wendell, Champion of Truth and Justice. Last week he was chewing out MS for their ARM support. Anyway, bad timing for Intel as AMD are getting very close to their Ryzen 9000 release which is looking very promising.


And review samples have apparently just gone out. Expected launch date, end of this month:

I'm seeing the usual "boorrrrring, 14% improvement is worthless" crowd emerge but I think this looks really promising. Efficiency seems much improved. Example the leaked scores show the 9700X 10% faster in Single Thread (4% faster in MT) over its predecessor 7700X but the 9700X is a 65W processor. Its predecessor is 105W. I could be wrong if the leaked scores aren't on stock voltage and similarly if the benchmark includes AVX-512 then that might throw it off from many real world scenarios. But to me this is looking really hopeful. Dropping CPU power down to 60% of its predecessor would be impressive even if performance was just at parity.

I'm assuming memory controller is the same though if it did let you use faster RAM, that would be awesome. 8000MHz RAM here we come.
 
@Overly Serious I think Zen 5 could have an unexpectedly high performance/Watt improvement, not sure how much but think 40%. Also lower stock temps on desktop from these TDPs/PPTs coming down, lower base clocks, and turbo clock speeds not/barely increasing. That's attributable to at least the use of a TSMC N4 node instead of N5, and other tweaks. It will be nice for mobile and Strix Point will beat the Snapdragon X Elite into submission.

The ~16% IPC improvement actually is boring (lol) and for gaming the 7800X3D will likely be tying or even faster than 9700X et al. Gamers want that 3D V-Cache and it would be better if both were launching at the same time. It sounded like there might be a full-width AVX-512 implementation but we should wait and see on that. AVX-512 could skew performance expectations but there are some legitimate applications for it. They should be adding some new AVX-512 instructions: AVXVNNI, MOVDIRI, MOVDIR64B, VP2INTERSECT, and PREFETCHI.

Based on this, you could run 8000 MT/s DDR5, but it's not going to be that helpful other than with desktop APUs. The "sweet spot" will be around DDR5-6000/6400 again. Keep in mind that there were already updates to AM5 motherboards last year that allowed them to support those higher speeds, but you aren't going to get much out of it other than wasting money:


Zen 6 is now poised to be the non-boring AMD release, because it could move away from the same chiplet design and 16 core maximum that has been seen since Zen 2. Look for it to have lower latencies from an interconnect change.

Intel can probably beat Zen 5 in gaming and other scenarios with Arrow Lake, but probably not Zen 5 X3D. Zen 5 X3D is said to have full overclocking support this time, and could have closer clocks to its non-X3D counterparts than 5000X3D and 7000X3D did. One way Intel can make Arrow Lake interesting is to stack Adamantine L4 cache on it. Arrow Lake should also have a better iGPU than the one AMD is reusing from the previous I/O chiplet, not comparable to APU performance but relevant if you buy them refurbished years from now.
 
I think Zen 5 could have an unexpectedly high performance/Watt improvement, not sure how much but think 40%.
There has been some testing of the 16-core 9950X at different PPTs (Package Power Tracking).

AMD Ryzen 9 9950X CPU Tested At 160W TDP, Blazes Past 230W Ryzen 9 7950X With 16% Performance Uplift (archive)

Here it's achieving a 7, 8, or 16% uplift in different Blender benchmarks vs. the 7950X while using less power. A naive calculation: (1.07 | 1.16) / (160 / 230) = +53-67% performance/Watt. That's probably an overestimate, but you can see how a 30-50% improvement is plausible.

AMD Ryzen 9 9950X “Zen 5” CPU Tested At 230W Power Limit, Up To 28% Faster Than 7950X & 34% Faster Than 14900K (archive)

Now, with both the 9950X and 7950X at the same PPT, 19-28% faster.

Strix Point mobile also has Zen 5c cores. From what I recall, the efficiency curve for Zen 4c vs. Zen 4 is not actually that impressive. The benefits are mostly in the smaller die area, and they can use slightly less power than a Zen 4 core would within a narrow frequency range. That's likely to happen because if the CPU isn't running a heavy multi-threaded workload, the scheduler will never prefer a 'C' core that boosts to around 30% less than the fast cores.

Zen 4c needs a higher core voltage to reach the same clock speeds as Zen 4. The VID (voltage identification definition) charts revealed that Zen 4 hits the Vmin (the minimal voltage that a processor requires for a workload at a particular frequency) at 2.3 GHz. In contrast, Zen 4c arrives at the Vmin below 1.5 GHz. The V/F (voltage-to-frequency) curve for both cores overlaps at 1.5 GHz. Zen 4c's power efficiency resides in between 1.5 GHz and 2 GHz. Zen 4c consumes less power despite the higher recorded voltage due to the more compact design.

Zen4c-Technology-for-Laptops-Press-Deck-8.png
 
Last edited:

My team at Alderon Games, working on the multiplayer dinosaur survival game Path of Titans, has been encountering significant problems with Intel CPU stability. These issues, including crashes, instability, and memory corruption, are confined to the 13th and 14th generation processors. Despite all released microcode, BIOS, and firmware updates, the problem remains unresolved.
We have identified failures in five main areas:
  • End Customers: Thousands of crashes on Intel CPUs on 13th and 14th Gen CPUs in our crash reporting tools.
  • Official Dedicated Game Servers: Experiencing constant crashes, taking entire servers down.
  • Development Team: Developers using these CPUs face frequent instability while building and working on the game. It can also cause SSD and memory corruption.
  • Game Server Providers: Hosting community servers with persistent crashing issues.
  • Benchmarking Tools: Decompression and memory tests unrelated to Path of Titans also fail.
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. This issue is gaining attention from news outlets and has been noted by Fortnite and RAD Game Tools, which powers decompression behind Unreal Engine.

Users are also receiving misleading error messages about running out of video driver memory, despite having sufficient memory.
 
Users are also receiving misleading error messages about running out of video driver memory, despite having sufficient memory.
Is this an incident of where Intel had a bad patch of memory for the cpu and shipped it anyway like the Zen 1 segfault issue? Or did Intel push their wafers so hard it's degrading the chips?
Anyways Intel's really fucked if they can't just bribe the OEM like last time.
 
  • Thunk-Provoking
Reactions: N Space and Vecr
Mm. Because this is hitting servers, data centres, etc, your big clients.
And having two generations of CPUs constantly failing's going to destroy their bottom lines, regardless of Intel incentives.

I can't imagine a data centre having a thousand 13900s go bad, get replacement 14900s and swap them in, and have them go bad, and being happy about it.

Sounds like a hilariously easy sale for AMD.
 
Weirdly, it looks to be just the 139 and 149s.

The 129's unaffected, and the other 13s and 14s haven't been mentioned.
 
Or did Intel push their wafers so hard it's degrading the chips?

Since it gets worse over time, and it only seems to be happening to the i9s with the super-sized die, it probably is the heat from being clocked to the edge degrading the chips. They pushed this thing as hard as they could to win internet pissing matches, and now they have a failing product. These rectangular chips have a flexing issue (an aftermarket bracket mitigates this), and I can't believe it's any less of an issue on the largest, hottest chips.

Weirdly, it looks to be just the 139 and 149s.

The 129's unaffected, and the other 13s and 14s haven't been mentioned.

An i9-12900 is basically an i7-13700 with slightly lower clock speed (and less cache? Don't remember and don't feel like looking it up)
 
Back