News AMD microcode improves cross-CCD latency on Ryzen 9000 CPUs — Ryzen 9 9900X and Ryzen 9 9950X cross-CCD latency cut in half to match previous-gen m...

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Did they change the I/O die or infinity fabric speed? I think that is determinative of the latency, no? To be clear I am more of a layman with the minutia of CPU architecture.
You missed this in the article, a?

"Ryzen 9000 uses the same IOD and Infinity Fabric configuration as Ryzen 7000."
 
  • Like
Reactions: bit_user
Phoronix has now tested this and found a negligible overall difference across a set of 385 benchmarks (0.14% improvement between geomeans). However, within that is quite a bit of variation. The greatest improvement was 16.5%, while the greatest regression was 10.3%. Most of the big differences seem to be in a positive direction, so it's probably a lot of smaller regressions that are dragging down the average.

Power consumption is about 0.2% lower, which seems at the edge of statistical significance.
 
Phoronix has now tested this and found a negligible overall difference across a set of 385 benchmarks (0.14% improvement between geomeans). However, within that is quite a bit of variation. The greatest improvement was 16.5%, while the greatest regression was 10.3%. Most of the big differences seem to be in a positive direction, so it's probably a lot of smaller regressions that are dragging down the average.

Power consumption is about 0.2% lower, which seems at the edge of statistical significance.
just looking through the comparison chart posted I'm curious what AMD was originally optimizing for. It does seem like most results don't have much of a change in general.
 
It seems that everyone has seemingly forgotten a simple rule of thumb when dealing with new CPU architectures (regardless of brand), and that is that new architectures always require a certain amount of time to tweak with bios/AGESAs--otherwise known as "teething"...😉 My Zen 2 AGESAs took nine months to iron out after it shipped in 2019. This is Zen5 and it's not Zen 4. I imagine there will be more changes made in the next few months, although not as many as from Zen 1 to Zen2, certainly. . I suppose people are a bit confused because it's the same AM5 socket? But it's still a new architecture. We will see more new architectures on AM5 I'm sure.
 
@valthuer @bit_user @helper800

Updated the BIOS last night. Most benchmarks had improvement but it was negligible... less than 5%. There was one that was a pretty insane improvement though.

Before/After update:

TimeSpy 24270

TimeSpy 29193

Not sure what happened. System is running with PBO and ram at 6000mhz. 4090 is stock. Maybe I OC and break 30k? 🤣

Either way... I consider it a definite win. More performance no matter how big or small.
 
@valthuer @bit_user @helper800

Updated the BIOS last night. Most benchmarks had improvement but it was negligible... less than 5%. There was one that was a pretty insane improvement though.

Before/After update:

TimeSpy 24270

TimeSpy 29193

Not sure what happened. System is running with PBO and ram at 6000mhz. 4090 is stock. Maybe I OC and break 30k? 🤣

Either way... I consider it a definite win. More performance no matter how big or small.

Niiice!!!

I know jealousy is a bad trait and all, but I'm definitely jealous right now. 😀😀



Revenge time! 🤣 OC strikes again! 🤣
 
I should probably OC... but don't see much of a need.

Yeah, you 're right about this one.

Just tried TimeSpy after a long time, with the latest Nvidia drivers.

OC, can definitely make a difference in 3dmark:


but in gaming... not so much.

The games in which my 4090 struggled before OC, she still struggles after.

In 4K Ultra RT native Black Myth Wukong, i went from 22 to 23 FPS.

Guess i'm gonna have to wait for 5090 for any sort of substantial difference.

If you ever wanna give OC a try, download the NVIDIA APP (that is, of course, if you haven't done so already):


It's probably the safest and easiest OC for your card.
 
Last edited by a moderator:
@valthuer @bit_user @helper800

Updated the BIOS last night. Most benchmarks had improvement but it was negligible... less than 5%. There was one that was a pretty insane improvement though.

Before/After update:

TimeSpy 24270

TimeSpy 29193

Not sure what happened. System is running with PBO and ram at 6000mhz. 4090 is stock. Maybe I OC and break 30k? 🤣

Either way... I consider it a definite win. More performance no matter how big or small.
The difference in timespy is from the gpu score, the cpu scores pretty much the same.
 
The difference in timespy is from the gpu score, the cpu scores pretty much the same.

Odd how it changed... because I didn't change any settings. I literally ran it and then updated the BIOS and ran it again. Either way I'm not struggling for performance.
 
Dunno, but your cpu score is a bit low. This is mine with a 12900k, you should definitely score higher than this

https://www.3dmark.com/spy/32427102

I'll have to check drivers. If you are OCing that might be part of the reason... because I'm at stock. But still... my GPU score is 4k higher with the same GPU... but CPU score is 5k lower.

Makes perfect sense.
 
Check your CPU temps and clocks under load. May have to repaste the AIO if the TIM cracked.

Good idea. I'll give that a try.

Temps haven't been outrageous IMO... 60-70c under load and peaks at 95C under stress tests. Haven't seen higher than that.