Update KB5007215 to build 22000.318 is being released as part of Patchday November, and the patch for the L3 cache of Ryzen CPUs is being made available to all users - this is the first time that cust... Windows 11 Build 22000.318: Microsoft rolls out AMD fix with KB5007215 to all users
I don't know if this made a difference for me ( 3700X ). In an old screenshot (Windows 10) I have for L3 cache: Read | Write | Copy | Latency 608GB/s | 550GB/s | 602GB/s | Trial version Now (W11 22000.318 updated) I have: Read | Write | Copy | Latency 575GB/s | 94000MB/s | 137GB/s | Trial version If somebody with a 3700X or 3800X running W11 stable has a much better result. Let me know so I check what's going on here. I don't care if it's a 5%-10% difference, but something bigger.
It has a lot of variance between runs so likely isn't a good way to validate latency like its being used.
Why are people still judging by some benchmark which doesn't DIRECTLY relate to performance level. Yes, AIDA can show some results, but does it change the performance if you see that vs what you see in W10 aparently? No.... right? What if, the rumours that it's AIDA-s fault are true?
Doesn't look like you where effected by the l3 cache issue either that or you already got the fix via the insider program as both your before and after screenshots show a latency of around 10ns My scores Pre Patch 38.5ns latency Post Patch 12.2ns latency
I downloaded KB5007215 patch day one, i remember before patch it was 50~98GB/sec read on L3 cache nows it's around 1050GB/sec but this does vary on run to run where sometimes it drops to like 600GB/sec but before the patch, it was consistent between 50~98GB/sec before, this is on a 3900X. As for latency can't never check cause I am using the free version and it always says trial on that section.
Those are 2 different PCs (RAM is different, BIOS is different...). What's the point of your comparison? The idea is to see one PC where only the fix is different, like @stereoman's post above.
Well that's throughput from 50 to 98GB/sec prior to patch and after patch over 1TB/sec, and you are right that is 10~20x more throughput. Although I hear the issue impacted processors the more cores CCD's and CCX's they have and technically the 3900X and 3950X have 2 CCD's and 4x CCX's. It almost looked like back in the old day when Ryzen first came out and the CCX traffic was hammering the memory controller where the process is running on a CCD but the data in the cache was part of another CCD. I did notice all threads were being used just running chrome which was kinda bizarre, to begin with.
After installing it all, my cinebench scores are back to where they were when on windows 10 2700x dropped to around 1400 in cb R15 back up to around 1750 stock speeds but to be honest with so much power around now a days I didn't notice unless benchmarking