New AMD Agesa Combo V2 1.0.8.1 update potentially improves latencies between CPU cores

asturur

2020-09-14 13:51

mentor07825:

Hey Alex, welcome to Guru3d. And yeah, I'm on the same boat. 3700x with an x570 board, water cooled. I don't go anywhere near the 4.4 ceiling. The best I can do is a 4.3 boost with PBO turned on, but obviously it cannot be sustained. I think all core is 4.17 under sustained benchmark load. At this point, I think the biggest wins on our platform that we can manually tune is RAM overclock.

I think 4.3 is perfectly somewhere near the 4.4 ceiling. Also 4.17 all core is very good.

#5829516

asturur

2020-09-14 13:52

magiQx:

Hi, got the same CPU on an x570 (taichi) and water cooled as well. sometimes while gaming i see the frequency reaching 4.4GHz but only for some seconds and not continuously. Did not yet try to benchmark all cores to see the max frequency. unfortunately i had no luck with manually tuning my RAM and had to rely on XMP. Manually i could never reach the default ram specs (15-15-15-15 @3600MHz).

Which ram offer default specs of 3600@15-15-15-15? i would be interested in buying them. For 3600 the best i can find is 16-18

#5829520

magiQx

2020-09-14 14:02

asturur:

Which ram offer default specs of 3600@15-15-15-15? i would be interested in buying them. For 3600 the best i can find is 16-18

16GB G.Skill Trident Z SKU F4-3600C15D-16GTZ. but my mistake, it's 15-15-15-35.

#5829521

Silva

2020-09-14 14:02

Good thing about AMD is that they keep working on the code after releasing a product, improving it not only for the next generation but for the current generation.

#5829533

blackguard

2020-09-14 14:44

Cidious:

I think you're also on overclockers.net. I solved it somehow by raising VDDG to 1.050v I used to run it at 0.950v long time ago without issues but every bios update changes stuff. ran it on 1.0v for a while without issues on A40 and A50 bios. but now it needs more... Might be related to this posting saying internal latency dropped. Maybe internally something is more voltage demanding because of that. Who knows... EDIT: I just checked my MSI webtickets and they had responded with the same suggestion to adjust VDDG. You can try something higher. For me it's 1.050v that does the trick.

Yes I am, though haven't been there in a while and I don't post much anywhere these days. Good shout on the VDDG. Interesting that MSI would advise users to raise voltage.

#5829548

schmidtbag

2020-09-14 15:00

I'm curious how this would affect the performance of games that exceed 100FPS. I'm sure it narrows the gap between Intel (though I highly doubt it's enough to match Intel).

geogan:

FFS. I was reading this and wanted to know which version of AGESA BIOS I actually have now on my ASUS Hero VIII x570 - haven't upgraded for a long time. So I tried to run the "ASUS Update" app and it procreating well blue-screened the PC completely. Buggy POS.

That's why I never depend on Windows tools to perform such updates. Too risky. Besides, those tools are often bloatware anyway.

#5829552

Astyanax

2020-09-14 15:09

blackguard:

MSI claimed this AGESA improved memory OC headroom, and given reports of lowered inter-core latency, we're looking at Infinity Fabric changes.

you realize you just agreed with me right? Error 19's are not caused by memory instability, its uncore instability in the cache lines.

#5829578

Undying

2020-09-14 16:07

It totally borked my xmp profile. Nothing is working beyond 2133mhz. Im switching back to F21 from gigabyte.

#5829591

blackguard

2020-09-14 16:24

Astyanax:

you realize you just agreed with me right? Error 19's are not caused by memory instability, its uncore instability in the cache lines.

Is agreeing with you a bad thing? I wasn't saying XMP was causing it either, just that data suggested enabling XMP is enough to cause issues, and pointed to the new microcode. Two people can reach the same conclusion independently, through different thought processes.

#5829601

DeskStar

2020-09-14 16:49

Undying:

It totally borked my xmp profile. Nothing is working beyond 2133mhz. Im switching back to F21 from gigabyte.

Which board do you have? I've got the Xtreme (570 & tRX40) and was just about there at flashing until I read your post!

#5829604

Krzysztof

2020-09-14 16:53

Hi . Guys i have asus prime x470 pro and 2700ryzen cpu.Now my Bios is 5406 from 2019/11/25. my Ram is XMP 3200mhz with bad score latency 16-20-20-38. My set is DOCP (xmp) ,and offset voltage cpu 1.238v with x40 ratio. stability 100% . Thats my score in AIDA cache mem : I see 2 version newer Bios for my mobo ,and take it last one : 5603 from 2020/08/10. And when updsate this Bios i write how going on ,and gives you my new score in AIDA cache mem test. Maybe tomorow....

#5829612

Undying

2020-09-14 17:20

DeskStar:

Which board do you have? I've got the Xtreme (570 & tRX40) and was just about there at flashing until I read your post!

Gigabyte x570 gaming X. Basically xmp are none working after the flash. F21 is working fine i just went back to it so keep that in mind.

#5829625

DeskStar

2020-09-14 17:59

Undying:

Gigabyte x570 gaming X. Basically xmp are none working after the flash. F21 is working fine i just went back to it so keep that in mind.

Right on. Are you actually "OCing" the memory or just running the XMP? Or just that XMP is completely shot? I have a 3600 128gb NeoZ running exactly like it should be and 3800 32gb just a 200mhz jump to match a 1900 infinity fabric.

#5829630

Undying

2020-09-14 18:15

DeskStar:

Right on. Are you actually "OCing" the memory or just running the XMP? Or just that XMP is completely shot? I have a 3600 128gb NeoZ running exactly like it should be and 3800 32gb just a 200mhz jump to match a 1900 infinity fabric.

I've tried xmp and manual oc with no luck, 2133mhz stuck. Tried rising vdd, dram and soc voltages nothing. Can you confirm that or still deciding what to do? 🙂

#5829634

DeskStar

2020-09-14 18:21

Undying:

I've tried xmp and manual oc with no luck, 2133mhz stuck. Tried rising vdd, dram and soc voltages nothing. Can you confirm that or still deciding what to do? 🙂

Oh hell I haven't plunged into it personally. And the damn example as to what it is supposed to do (BIOS update) is for memory compatibility and nothing about a new AGEISA what so ever. I'll keep off of this one for the time being until it gets hashed out. I like my settings and she's solid as a rock as of now. Just got done tweaking my 5700XT in here and couldn't be happier with it. 2080 super performance practically!! Just not sure I want to even mess with a secondary BIOS setting as I've had my board just say "F" off in the past and loads whatever it wanted to after crashing. All to be seen that the BIOS doesn't even have any saved settings, so then I have to start all over.

#5829639

Astyanax

2020-09-14 18:32

Reminder that some bios updates require a NVRAM wipe because of changes to c_var names that gigabyte and msi have historically been stupid enough to make.

#5829640

Robson

2020-09-14 18:35

Slightly faster with same Bios and Ram settings.

#5829642

kapu

2020-09-14 18:41

Change nothing for me . Whats Your best results for memmory latency ? Or there is different test to see it actually works ? 😀

#5829696

willgart

2020-09-14 21:32

kapu:

Change nothing for me . Whats Your best results for memmory latency ? Or there is different test to see it actually works ? 😀

its not about the ram access its the core to core communication, so 1 thread communicating with another one on another core. there is a tool to measure that, I dont remember the name.

#5829710

waltc3

2020-09-14 22:03

1.0.8.1 F-30a bios runs fine in my x570 Aorus Master rev 1.0 mboard. I added 2x8GB more ram to populate 32GB in all four slots (4x8) @ 3733MHz using a combination of XMP + faster-than-stock ram timings of 16-19-19-19-39-68 1T Gear Down (CL is set to 15 in the bios which =16 CL at boot with GD.)