Share your AIDA 64 cache and memory benchmark here

[megaclite]

11600k
12600k
9900x

 

[FilipM]

Not the happiest with the L3 cache performance, especially with the latency...maybe too little EDC, but not a bad result at all Also re-did some RTT/Cad settings again


Im within 1ns off my 2 sticks CL15 run at the same speed, so Im happy

 

[tabascosauz]

A few tests below at 3800. 4ns lower with 3rd party apps disabled and also safe mode results.

View attachment 245150

Nice! Thanks for doing the test.

After some snooping I only had to disable internet + reboot (no difference unless reboot, I'm guessing whatever service has already started by then), to get my usual results back. Roughly 1.8ns difference for me.

I'm thinking it must be a new service/process native to 21H2. Have not had this happen from the first pre-release 11 build all the way to 21H1. Not sure what it is though, as my running services list is already very trimmed of unnecessary stuff, and I always bench with minimal background stuff. I didn't see anything in there, and nothing irregular is hogging CPU usage either.

 

[Dia01]

Nice! Thanks for doing the test.

After some snooping I only had to disable internet + reboot (no difference unless reboot, I'm guessing whatever service has already started by then), to get my usual results back. Roughly 1.8ns difference for me.

I'm thinking it must be a new service/process native to 21H2. Have not had this happen from the first pre-release 11 build all the way to 21H1. Not sure what it is though, as my running services list is already very trimmed of unnecessary stuff, and I always bench with minimal background stuff. I didn't see anything in there, and nothing irregular is hogging CPU usage either.

View attachment 245258

Curious, what's you're DIMM voltage?

 

[tabascosauz]

Curious, what's you're DIMM voltage?

1.55V for 3800 14-14-14, 1.42V for 3800 15-15-15, 1.42V for 3600 14-14-14.

Forgot to mention, the other half of my problem was corrupt chipset drivers that affected CPU perf as well. The latest chipset drivers (March 14) are pretty broken, the installer throws a bunch of errors about dependencies not found. It'll still finish installing though, I think my mistake was that I canceled it midway, so they've been all fudged up before reinstalling. Could be worth a try.

 

[Dia01]

1.55V for 3800 14-14-14, 1.42V for 3800 15-15-15, 1.42V for 3600 14-14-14.

Forgot to mention, the other half of my problem was corrupt chipset drivers that affected CPU perf as well. The latest chipset drivers (March 14) are pretty broken, the installer throws a bunch of errors about dependencies not found. It'll still finish installing though, I think my mistake was that I canceled it midway, so they've been all fudged up before reinstalling. Could be worth a try.

Using Motherboard specific chipset drivers or straight from AMD?

Decent tuning BTW

 

[tabascosauz]

Using Motherboard specific chipset drivers or straight from AMD?

Decent tuning BTW

From AMD website, 4.03.03.431

https://www.amd.com/en/support/chipsets/amd-socket-am4/x570

My 3800 latency is always high. 15-15-15 is often only 0.2ns slower (bandwidth is normal though). I've thrown all the tricks in the book at it but 54.7 is as far as it goes. I can't run 3866 without Bus/interconnect WHEA (but can still boot 4000) so I'm guessing the latency is from being on the edge, idk

I'm curious to see what you can do at 3800cl14 on 5950X. I wonder if it's just a dual rank thing?

 

[FilipM]

From AMD website, 4.03.03.431

https://www.amd.com/en/support/chipsets/amd-socket-am4/x570

My 3800 latency is always high. 15-15-15 is often only 0.2ns slower (bandwidth is normal though). I've thrown all the tricks in the book at it but 54.7 is as far as it goes. I can't run 3866 without Bus/interconnect WHEA (but can still boot 4000) so I'm guessing the latency is from being on the edge, idk

I'm curious to see what you can do at 3800cl14 on 5950X. I wonder if it's just a dual rank thing?

Increase Vsoc and see if latency goes down

 

[tabascosauz]

Increase Vsoc and see if latency goes down

Ach, I have tried. More VSOC, VDDG, VDIMM, nothing helps. It just kinda runs into a wall where latency doesn't scale. Can't remember exactly but I think membench results still scale normally, just AIDA latency.

 

[The King]

Ach, I have tried. More VSOC, VDDG, VDIMM, nothing helps. It just kinda runs into a wall where latency doesn't scale. Can't remember exactly but I think membench results still scale normally, just AIDA latency.

Few things you can try RFC 256, Dropping SCL to 2s and WTRS 3 and SD DD both 4s and 6s

 

[tabascosauz]

Few things you can try RFC 256, Dropping SCL to 2s and WTRS 3 and SD DD both 4s and 6s

I should clarify, it will always go tighter with tRFC, but that's not the issue. And tRFC doesn't seem to move the needle either at 3800CL14, latency just kinda...... stuck. Most importantly, 15-15-15 with nothing else changed except 2T and corresponding bump in tRFC, is usually only 0.2-0.3ns slower or even the same.

It's a peculiar thing. iirc @freeagent is in the same range with his 5900X. Quick google search for other much tighter 3800CL14 profiles on 5900X, all of them seem to land around the 54.5ns mark. I can do 54.2 once in a blue moon on Unify-X, but still it barely moves. Probably a few outliers I haven't seen yet, but everything I've seen so far is 54.3-54.9.

Notably I saw a very tight reddit one, 14-12-13, CWL 10, 120ns ish tRFC.........54.4ns. By conventional tRFC wisdom (and those primaries too!) we shouldn't even be in spitting range of each other's results. None of them seem lacking for VSOC, though some are a bit low on VDDG (but that's more a WHEA problem)

 

[tjtremor999]

Nice! Thanks for doing the test.

After some snooping I only had to disable internet + reboot (no difference unless reboot, I'm guessing whatever service has already started by then), to get my usual results back. Roughly 1.8ns difference for me.

I'm thinking it must be a new service/process native to 21H2. Have not had this happen from the first pre-release 11 build all the way to 21H1. Not sure what it is though, as my running services list is already very trimmed of unnecessary stuff, and I always bench with minimal background stuff. I didn't see anything in there, and nothing irregular is hogging CPU usage either.

View attachment 245258

54.x is about the max for 3800cl14 on dual ccds. As far as being ~54.7 that's due to your 260trfc. ~242trfc should drop to 54.5ns

It gets to 52.x around 2000IF and you get the +3GB/s speed.

 

[damric]

Try 101 base clock. If that works, try 102, ect. You may be surprised that you can exceed 1900FCLK by doing so and still remain stable.

 

[freeagent]

I think the only time my 5900 has been in the 52ns range is with fclk at 2000. It is a whea generator at that speed so I don’t run it

 

[tjtremor999]

I think the only time my 5900 has been in the 52ns range is with fclk at 2000. It is a whea generator at that speed so I don’t run it

If it's the whea 19 warning, that might be just agesa bug. I tested 4000cl16 2000IF multiple times since I was doing CO no crash or performance loss.

Funny enough, my event viewer was spamming DCOM errors(my store apps are broken) like crazy. Was up to 7% cpu usage from event viewer in task manager under load.
Having 300+ whea 19 over 3mins of usage didn't even generate DPC latency ,Event viewer usage like DCOM spam.

I just disabled the event viewer for now, too lazy to backup for fresh install of windows.

 

[freeagent]

I’ll try when I get home

 

[zx128k]

Testing this at the moment.

 

[lawood]

If it's the whea 19 warning, that might be just agesa bug. I tested 4000cl16 2000IF multiple times since I was doing CO no crash or performance loss.

Funny enough, my event viewer was spamming DCOM errors(my store apps are broken) like crazy. Was up to 7% cpu usage from event viewer in task manager under load.
Having 300+ whea 19 over 3mins of usage didn't even generate DPC latency ,Event viewer usage like DCOM spam.

I just disabled the event viewer for now, too lazy to backup for fresh install of windows.

I find this interesting.
Over 1900 FCLK I seem to be 100% stable, this entails prime small ffts, linpack, occt, burntest, I can go hours without crashes. Also gamed a good amount without problems.

BUT I get his warning on event viewer, the whea 19 warning:

A corrected hardware error has occurred.

Reported by component: Processor Core
Error Source: Unknown Error Source
Error Type: Bus/Interconnect Error
Processor APIC ID: 0

The details view of this entry contains further information.

I can also see them on hwinfo.

I since then dialed back my ram to 3800mhz to eliminate whea errors.

 

[tjtremor999]

I find this interesting.
Over 1900 FCLK I seem to be 100% stable, this entails prime small ffts, linpack, occt, burntest, I can go hours without crashes. Also gamed a good amount without problems.

BUT I get his warning on event viewer, the whea 19 warning:


I can also see them on hwinfo.

I since then dialed back my ram to 3800mhz to eliminate whea errors.

that one is an actual error on core 0 of your cpu; whea 19 warnings show as self correcting with no relevant source attached to them

the whea19 error on core0 was typical on my 3900x if I used undervolt on the vcore, maybe 1 or 2 every month would show, also 100% stable under testing,apps,games.

 

[zx128k]

I find this interesting.
Over 1900 FCLK I seem to be 100% stable, this entails prime small ffts, linpack, occt, burntest, I can go hours without crashes. Also gamed a good amount without problems.

BUT I get his warning on event viewer, the whea 19 warning:


I can also see them on hwinfo.

I since then dialed back my ram to 3800mhz to eliminate whea errors.

whea errors = unstable system.

 

[tjtremor999]

whea errors = unstable system.

it's not that simple, you can have 1 or 2 whea errors & be 100% stable

the opposite is also true, you can have 0 whea errors, pass every stress test out there & crash during idle, or long gaming sessions

 

[zx128k]

it's not that simple, you can have 1 or 2 whea errors & be 100% stable

the opposite is also true, you can have 0 whea errors, pass every stress test out there & crash during idle, or long gaming sessions

idle crash is caused by too low voltage at idle its 100% something bad but you can work around it. Overclocking wise always go for 100% stable, which means 0 WHEA errors. A stock system should have 0 WHEA errors.

 

[Dia01]

3800C14
1.480V DIMM

 

[lawood]

whea errors = unstable system.

Yes, I agree.
I dialed down the frequency to 3800 and 1800 fclk and it looks good now.
My system really doesn't like anything above 1800.

 

[Ibizadr]

3800C14
1.480V DIMM

View attachment 245425

Twr =2 X trtp

Yes, I agree.
I dialed down the frequency to 3800 and 1800 fclk and it looks good now.
My system really doesn't like anything above 1800.

1800 wasn't 1=1 to 3800. You need to run fclk==mclk. Só mclk at 1900 should do ram at 3800. If it boot 38xx with whea errors it can do 3800 stable for sure