AMD AGESA 1.2.0.2 Update Fixes Ryzen 9000 Series Inter-Core Latency Issues

[AleksandarK]

According to new latest testing, the latest AGESA (AMD Generic Encapsulated Software Architecture) update, version 1.2.0.2, promises a significant boost in performance for AMD Ryzen 9000 "Zen 5" processors. This update is targeting one of the most crucial aspects of multi-core processing: inter-core latency. The AGESA 1.2.0.2 update addresses challenges initially reported in AMD's Zen 5 architecture, particularly in scenarios demanding rapid communication between multiple cores. Early reports suggest a remarkable reduction in inter-core latency by up to 58%. According to Overclock.net testing, older AGESA 1.2.0.1A showed the cross-CCD latency at around 180 ns. However, with the new AGESA 1.2.0.2 BIOS, the latency is seemingly around 75 ns.

Interestingly, the update has arrived on the ASUS ROG Crosshair X670E motherboard, with BIOS version 2401. BIOS updates with the latest AGESA 1.2.0.2 are still rolling out, so it will be interesting to see further testing and possible improvements. It could be that the cross-CCD latency has just been reported badly, so final testing will conclude the latency increase from Zen 4 to Zen 5 debate.



View at TechPowerUp Main Site | Source

 

[FoulOnWhite]

Nice indeed. Looking forward to some TPU user reports on this

 

[Macro Device]

Does this mean they will become better in gaming?
I'm no expert, this is a serious question. I wanna know.

 

[Onasi]

@Beginner Macro Device
Theoretically. Latency matters a lot for gaming workloads. How much this will pan out in practice in terms of actual frametimes is up in the air.

 

[aktpu]

Nice for benchmark wanking, doubt it'll do much for anyone else

 

[low]

Nice to see my future 9800x3D evolving.

On release the 9k series will be stable and first day error fixed.

 

[RootinTootinPootin]

Does this mean they will become better in gaming?

FPS wise, not much, but the "feel" would be alot better. this is meant for the multi CCD chips, ain't really for single CCD chips like Ryzen 7 and below. (correct me if I am wrong)

Nice to see my future 9800x3D evolving.

Duhhhh!! the 9800X3D (Ryzen 7 parts) are single CCD, this only applies to multi CCD chips like the Ryzen 9

 

[Bruno Vieira]

The performance with W11-24-H4, now the inter-core latency. If they solve the 2-cycle SIMD problem, zen5 will be a proper monster months post-launch (and hopefully very discounted)

 

[Onasi]

this is meant for the multi CCD chips, ain't really for single CCD chips like Ryzen 7 and below. (correct me if I am wrong)

Not necessarily, it says inter-core, not solely cross-CCD. So it should help single CCD chips too, albeit to a lesser extent, theoretically.

 

[R0H1T]

Fine wine AMD storefront open for business

Spoiler

 

[RootinTootinPootin]

Not necessarily, it says inter-core, not solely cross-CCD. So it should help single CCD chips too, albeit to a lesser extent, theoretically.

the other CCD present aside from the cores is the I/O die, pretty much where the IMC is. in his test he's clocked to oblivion 8k MT's, you should take that result with a grain of salt, either ways with normal plebs pretty much the improvement is Nigh but its still there, an improvement is still an improvement otherwise.

EDIT: Checked it on OCN (where I am a regular member as well) and its on the CCD cores only.

 

[tfp]

Nice improvement, to bad the software heat map looks the same with that level of change in the latency.

Not exactly sure why hardware companies are giving people beta firmware and drivers. These are all well known benchmarks and something AMD should have been able to prioritize. Intel's screwing up big but it's nice that AMD makes sure that Intel can be seen in the best light based on the current circumstances.

 

[R0H1T]

Heat map? You mean someone made them all green, instead of yellow/red or something?

 

[windwhirl]

Fine wine AMD storefront open for business

Spoiler

I wish we didn't have to wait for the "wine" to mature though, it generates a bad first impression that is hard to correct.

 

[tfp]

Heat map? You mean someone made them all green, instead of yellow/red or something?

Yes exactly, it looks to be based on relative difference and not factoring in any overall grading. The images that you posted the one with 180 - 210 latencies looks about the same or maybe better because it has more yellow. At least your version shows that there significant differences across die vs the green and light green graphics.

Not a big deal

 

[Darc Requiem]

AMD should have just pushed back Zen 5's launch a couple of months and ironed out all these issues. Intel is having issues right now, so I don't see the impetus behind the rush to market.

 

[Makaveli]

Does this mean they will become better in gaming?
I'm no expert, this is a serious question. I wanna know.

not sure as you generally don't want things bouncing back and forth between CCD's

AMD should have just pushed back Zen 5's launch a couple of months and ironed out all these issues. Intel is having issues right now, so I don't see the impetus behind the rush to market.

To be honest this isn't something that is new for new arch. Intel also had issues when they dropped Alder Lake that was fixed. They had ecore issues, thread director issues etc.

I think far too many people have forgotten this. There is always some growing pains when you are doing something new.

 

[_roman_]

This is just an excel function. The color is defined on the size of the numbers. Therefore two different tables can not be compared by color alone. Each different table has different numbers and different minimal and maximal about of numbers. I used that function to only see where are the bigger and where are the smaller number visually for one table each.

not sure as you generally don't want things bouncing back and forth between CCD's

How the scheduler works is a big project in my point of view in the linux kernel. Not sure if Microsoft does any optimizations at all. On which core you put processes and threads is a never ending story what i saw in the past years in the linux kernel. I do read the new introduced changelog options for the linux kernel to decide if i want to use them or not (yes or no).

 

[umeng2002]

And AMD wants to be a "software company" when they couldn't iron out these issue for the chip launch last month?

 

[Assimilator]

I think far too many people have forgotten this. There is always some growing pains when you are doing something new.

And there is this thing called "extensive quality assurance testing" that you do, as a responsible company, to ensure your users don't experience those pains. Unfortunately every company is making customers their beta testers nowadays.

 

[Makaveli]

And there is this thing called "extensive quality assurance testing" that you do, as a responsible company, to ensure your users don't experience those pains. Unfortunately every company is making customers their beta testers nowadays.

This is true but it seems other industries also do this like gaming.

 

[tfp]

This is true but it seems other industries also do this like gaming.

The problem to me is this isn't gaming. Intel and AMD need to deliver quality because CPUs and the other HW they make isn't just for "fun".

 

[Makaveli]

The problem to me is this isn't gaming. Intel and AMD need to deliver quality because CPUs and the other HW they make isn't just for "fun".

I know just agreeing with you that its an industry problem.

Everyone seems to be on the release first patch after boat.

 

[mkppo]

Won't really do much for benchmarks as most software doesn't cross a CCD. And this is coming from the dude who made this cross latency test..

But obviously there was a massive uproar over this with people claiming this bug reduces performance drastically bla bla.

Good thing they fixed it I suppose.

 

[csendesmark]

Great news!
Do we have a similar "map" for Threadrippers?