AMD 2:1 IMC Latency Penalty Fixed?

[ir_cow]

I'm redoing all my memory benchmarks with the 9950X. To my surprise, the 2:1 penalty is gone with new MSI BIOS and AGESA PI-1.2.0.3a.

It use to be around 10ns, now its practically zero. Not sure what magic MSI is doing... I have a feeling some Auto settings is messing with the tREFI and a few sub-timings to achieve this.

 

[R0H1T]

What's the actual difference in other, CPU heavy, workloads?

 

[ir_cow]

Hmm. it gets stranger... ugh this what I hate about reviewing stuff. Can't be simple at all. This is getting increasingly harder to explain why 2:1 is suddenly good, but It might just be this Motherboard doing funky stuff. why is 7950X latency across the board better?

 

[R0H1T]

Could that that random thing i.e. Silicon lottery.

 

[ir_cow]

Could that that random thing i.e. Silicon lottery.

I'm going to say no because the 9800X3D has the same latency.

I think I will scrap it all and start over again. New OS install, BIOS, drivers, etc.

Edit: The motherboard reviewer is getting similar latency with DDR5-6000

And I was getting much lower...

I might have figured it out. Previously I was using Windows 23H2, now I am on 24H2. Gained 10ns just by switching to the newest OS.
Not the OS. I swapped back and latency is the same.

Can it really be just the 9000 series / new AGEAS?

 

[blinnbanir]

Are you saying that the 6000 sweet spot no longer matters? If this is true and becomes known fast RAM will be even more expensive. Does it also means that it widens the gap vs Intel? It would mean that was the last thing that Intel has in advantage (RAM speed) would suddenly be moot.

 

[R-T-B]

Hmm. it gets stranger... ugh this what I hate about reviewing stuff. Can't be simple at all. This is getting increasingly harder to explain why 2:1 is suddenly good, but It might just be this Motherboard doing funky stuff. why is 7950X latency across the board better?

View attachment 391022 View attachment 391024

I think the 7950x vs 9950x latency increase has been a known thing for a while, without a clear explanation from AMD. It's part of the reason 9xxx series doesn't always win gaming benches without using something like AVX-512... Or am I wrong?

Can it really be just the 9000 series

Pretty sure that's it. Also they did do some logic changes to the latency defaults in AGESA 1.2.02a. Possibly part of it as well.

 

[blinnbanir]

I think the 7950x vs 9950x latency increase has been a known thing for a while, without a clear explanation from AMD. It's part of the reason 9xxx series doesn't always win gaming benches without using something like AVX-512... Or am I wrong?


Pretty sure that's it. Also they did do some logic changes to the latency defaults in AGESA 1.2.02a. Possibly part of it as well.

As far as the 9000 series goes I found something today. I have a ton of mods installed for AMS2. The latest truck update deleted those files. There is a program called AMS2CM (Content Manager) that handles adding the mods (back) to the Game. I had Task manager running and when I put the GUI on performance, every single core was running to the same clock. It was kind of insane seeing every single core boost (Those little Graphs) be the exact same in every core Window. Are we talking about launch day latency for 9000? I thought there was an Agesa update that resolved that months ago?

If it can also benefit from faster Memory these will be the fastest chips on desktop period.

I have a 9900X3D with 64GB of 6000 Gskill memory.

 

[ir_cow]

I think the 7950x vs 9950x latency increase has been a known thing for a while, without a clear explanation from AMD. It's part of the reason 9xxx series doesn't always win gaming benches without using something like AVX-512... Or am I wrong?

Pretty sure that's it. Also they did do some logic changes to the latency defaults in AGESA 1.2.02a. Possibly part of it as well.

I did not know this.

Even though latency is now higher, FPS is as well.

 

[Onasi]

There were speculations prior and at release that the 800 series chipsets paired with 9000 series CPUs could potentially yield better memory performance, even though the IMC is technically unchanged. Maybe that’s finally bearing fruit after AGESA updates?

 

[danc]

did you compare 1:1 also with the new bios? Do not engage EXPO, MSI AIB are likely to tweak timings.

So you can use zentimings to check out what parameters MSI changed between EXPO 2:1 vs 1:1 vs JEDEC 2:1 vs 1:1.

 

[ir_cow]

There were speculations prior and at release that the 800 series chipsets paired with 9000 series CPUs could potentially yield better memory performance, even though the IMC is technically unchanged. Maybe that’s finally bearing fruit after AGESA updates?

I agree with this assessment. Same SoC, but clearly they did something.

did you compare 1:1 also with the new bios? Do not engage EXPO, MSI AIB are likely to tweak timings.

Yep that's the 4800

So you can use zentimings to check out what parameters MSI changed between EXPO 2:1 vs 1:1 vs JEDEC 2:1 vs 1:1.

I have. Disabled all optimization lead to 1-2ns higher. Slightly lower tREFi, but nothing drastic.

 

[R-T-B]

I thought there was an Agesa update that resolved that months ago?

Improved but its still a thing vs previous gen, I think. Not really a big deal anymore though to my knowledge.

I agree with this assment. Same SoC, but clearly they did something.

Yeah, but I think it lies more in firmware. Do the 600 series boards behave similarly?

Even though latency is now higher, FPS is as well.

An AMD engineer said that'd be the case on reddit with AGESA 1.2.0.2a plus. Wish I could find the post. I'll keep looking.

 

[_roman_]

I miss the zentimings screenshots.

It's all in the numbers and not in the 6000 MT/s

 

[R-T-B]

numbers

Bandwidth and FPS metrics means more to me than just raw timings, honestly.

 

[ir_cow]

Yeah, but I think it lies more in firmware. Do the 600 series boards behave similarly?

haven't tried a 9000 in the 600 Chipset yet. Might be a while though. Limited time to work on reviews between day job and family.

I miss the zentimings screenshots.

It's all in the numbers and not in the 6000 MT/s

I didn't upload any. Still gathering data.

 

[A Computer Guy]

Hmm. it gets stranger... ugh this what I hate about reviewing stuff. Can't be simple at all. This is getting increasingly harder to explain why 2:1 is suddenly good, but It might just be this Motherboard doing funky stuff. why is 7950X latency across the board better?

View attachment 391022 View attachment 391024

Is this an off script example of AMD's "Fine Wine" effect?

 

[danc]

DDR5 각각의 램 타이밍이 게이밍 FPS에 미치는 영향 (9800X3D)

1. 시스템 정보1) CPU: 9800X3D (5.225 GHz 고정)2) GPU: RTX40903) M.B…

quasarzone.com

Here is shown TREFI and TRFC have biggest impact in FPS. So yours may be also showing in CS2 with MSI tweaked timings in EXPO

 

[mkppo]

Pretty sure it's related to the funky stuff some AGESA does specifically with AIDA latency. Does the same happen in other latency tests?

For gaming generally 6000 yields better performance but 8000, when tuned, is faster in applications and can match a tuned 6000 in games.

 

[DirtyDingusMcgee]

Sounds like it is time to update my bios and give that Mushkin 6400 cl30 that my board and cpu didn't like another try.

 

[ir_cow]

DDR5 각각의 램 타이밍이 게이밍 FPS에 미치는 영향 (9800X3D)

1. 시스템 정보1) CPU: 9800X3D (5.225 GHz 고정)2) GPU: RTX40903) M.B…

quasarzone.com

Here is shown TREFI and TRFC have biggest impact in FPS. So yours may be also showing in CS2 with MSI tweaked timings in EXPO

All extra memory stuff that the motherboard can do is turned off. I also manually enter the tRFC if it's not inputted correctly (asus sometimes does this).

I wish it was that simple