Share your AIDA 64 cache and memory benchmark here

[Simkin]

@Simkin post up a Zentimings screenshot. The latency looks about right for 18-19-19, but your bandwidth is way down at least on copy where I'd expect 3600CL16 to be. I can understand a 1-2GB/s loss due to looser timings, but not 5GB/s. At 4000 16-16-16 on the 4650G:
I did consistently lose a tiny bit of read and copy on evey profile going from 4650G to 5600G and 5700G. But overall Cezanne enjoys ~3ns gain, and a gain in membench as well. AIDA is so fickle I generally only run it once and move onto membench

Here it is

 

[tabascosauz]

Here is is

How about this:

• tRRDS 4
• tRRDL 6
• tFAW 16
• tWTRS 4
• tWTRL 12
• tWR 12
• tRFC 320
• tRDRDSCL 4
• tWRWRSCL 4
• tRTP 10

160ns tRFC is pretty lax and shouldn't require much more VDIMM than you use currently, but if necessary give it a 0.01-0.02V bump (assuming your current settings are completely stable)

 

[Simkin]

How about this:

• tRRDS 4
• tRRDL 6
• tFAW 16
• tWTRS 4
• tWTRL 12
• tWR 12
• tRFC 320
• tRDRDSCL 4
• tWRWRSCL 4
• tRTP 10

160ns tRFC is pretty lax and shouldn't require much more VDIMM than you use currently, but if necessary give it a 0.01-0.02V bump (assuming your current settings are completely stable)

Thanks mate, it looks alot better now! i bumped dimm voltage up to 1.38v also.

 

[mclaren85]

Here is my monster without core performance mode and overclocking:

 

[Simkin]

Some small adjustment to my 32GB b-die on my old 5960X. Adjusted all the sub-timings, and tRAS down from 34 to 32. Its over 80k read now. Latency under 51ns also. This old X99 system still hangs in pretty well

 

[freeagent]

This old X99 system still hangs in pretty well

Oh yeah those old Intel's rock, latency was never a problem.. heck my 3770K can make it to mid 30s with old ddr3

 

[mclaren85]

Some small adjustment to my 32GB b-die on my old 5960X. Adjusted all the sub-timings, and tRAS down from 34 to 32. Its over 80k read now. Latency under 51ns also. This old X99 system still hangs in pretty well

View attachment 213578

How is your's is fast even my system is better on the paper?
That's not fair man:/

 

[freeagent]

That's not fair man:/

You just gotta learn to tweak!

 

[Simkin]

Seems like my 4000Mhz is not stable. It runs Aida fine even many times in a row, and a quick memtest (TestMEM) shows no error. I know this is not an indication of stability though.

The PC just reboots without any BSOD, just a black screen and im back at the login screen. Happend two times now.

I suspect the 2000 FCLK as i have not adjusted any other voltages than the vdimm.

Back to 1933/3866 to see if its stable without any voltage adjustment.

 

[freeagent]

That speed is not stable for me even at 1.65v that’s 2v territory I think.

Edit:

I can get the speed stable but not with those timings. Try flat 15s that might work with about 1.5v

Edit again:

You will have to tweak voltage sometime soon

 

[Simkin]

That speed is not stable for me even at 1.65v that’s 2v territory I think.

Edit:

I can get the speed stable but not with those timings. Try flat 15s that might work with about 1.5v

Edit again:

You will have to tweak voltage sometime soon

Im back at 1933/3866, still vdimm 1.38v with 18-18-18-36 and tweaked sub-timings. Aida now is 58900-58900-52700 and 55.0ns.

I just want to see how far i can push this without touching the voltage on the 5700G, its my HTPC only so no big deal anyway

The FCLK on these APU's can, as far as i have read, go as high as 2500Mhz, but of course not without raising voltages.

 

[INSTG8R]

Recently moved from 16GB to 32GB despite needing loose timings I'm pleased with the result and definitely saw improvements in other benches

 

[tabascosauz]

Seems like my 4000Mhz is not stable. It runs Aida fine even many times in a row, and a quick memtest (TestMEM) shows no error. I know this is not an indication of stability though.

The PC just reboots without any BSOD, just a black screen and im back at the login screen. Happend two times now.

I suspect the 2000 FCLK as i have not adjusted any other voltages than the vdimm.

Back to 1933/3866 to see if its stable without any voltage adjustment.

Im back at 1933/3866, still vdimm 1.38v with 18-18-18-36 and tweaked sub-timings. Aida now is 58900-58900-52700 and 55.0ns.

I just want to see how far i can push this without touching the voltage on the 5700G, its my HTPC only so no big deal anyway

The FCLK on these APU's can, as far as i have read, go as high as 2500Mhz, but of course not without raising voltages.

From that description it's definitely VSOC. Benching at 2500MHz (hell I'd say 2350MHz you're pushing it) is probably only possible if you don't use the iGPU. TM5 *can* catch IF errors once in a while but don't count on it. Frankly the iGPU is your best IF stress test, nothing comes close

My B550I Aorus Pro AX has always had a problem with any Renoir and Cezanne APU in that its LLC implementation sucks ass, and as soon as >10A of current flows through the SOC domain (ie. any significant iGPU load), VSOC drops like a brick. When that happens and VSOC isn't sufficient to sustain the IF at given speed, then instant reboot, no BSOD. Not sure if your Steel Legend suffers the same problem, you'd have to watch HWInfo.

Solution is just to set higher VSOC to raise the vdroop floor a bit. I'm at 1.256V SOC on 2300 iGP/2166 IF because it droops from 1.256V to 1.17V in game......I know someone who runs north of 1.3V SOC to sustain 2400/2200

The problem with trying to get vendors to try and fix something like this is that the only people who will have this issue are those that seriously use the iGPU. Anything (even Prime95 Large FFT or TM5/HCI) pales in comparison to iGPU usage in terms of VSOC current.

 

[Simkin]

From that description it's definitely VSOC. Benching at 2500MHz (hell I'd say 2350MHz you're pushing it) is probably only possible if you don't use the iGPU. TM5 *can* catch IF errors once in a while but don't count on it. Frankly the iGPU is your best IF stress test, nothing comes close

My B550I Aorus Pro AX has always had a problem with any Renoir and Cezanne APU in that its LLC implementation sucks ass, and as soon as >10A of current flows through the SOC domain (ie. any significant iGPU load), VSOC drops like a brick. When that happens and VSOC isn't sufficient to sustain the IF at given speed, then instant reboot, no BSOD. Not sure if your Steel Legend suffers the same problem, you'd have to watch HWInfo.

Solution is just to set higher VSOC to raise the vdroop floor a bit. I'm at 1.256V SOC on 2300 iGP/2166 IF because it droops from 1.256V to 1.17V in game......I know someone who runs north of 1.3V SOC to sustain 2400/2200

The problem with trying to get vendors to try and fix something like this is that the only people who will have this issue are those that seriously use the iGPU. Anything (even Prime95 Large FFT or TM5/HCI) pales in comparison to iGPU usage in terms of VSOC current.

Yep, i assume VSOC. however, if i decide to raise it, i only need to raise VSOC? As there is more voltages to adjust like VDDG etc.

And do you think 1933 fclk will be ok without touching any voltages?

When it comes to the Steel Legend, it has a 10-Phase Power Supply, and the top boards from ASRock have 14 and 16 Phases, so i dont know how much this will affect the power delivery, vdroop etc "50A Dr.MOS" and "Premium 60A Power Choke" Seems to be the same on my mATX Steel Legend as the top ATX boards from ASRock.

 

[tabascosauz]

Yep, i assume VSOC. however, if i decide to raise it, i only need to raise VSOC? As there is more voltages to adjust.

And do you think 1933 fclk will be ok without touching any voltages?

When it comes to the Steel Legend, it has a 10-Phase Power Supply, and the top boards from ASRock have 14 and 16 Phases, so i dont know how much this will affect the power delivery, vdroop etc "50A Dr.MOS" and "Premium 60A Power Choke" Seems to be the same on my mATX Steel Legend as the top ATX boards from ASRock.

I know, I had the Steel Legend mATX for two days lol, sent it back because none of the early BIOSes were stable and finished, it's not a bad lower midrange board at all.

Less to do with the VRM parts, more to do with the LLC implementation. By all standards the Aorus AX is one of the strongest ITX boards, but the B550I Gaming Edge Wifi has really tight VSOC LLC. Different VRM parts, but about as capable. But yeah, might want to keep an eye on your VSOC. Mine used to crash just watching a movie when I was trying to run 2000 IF at 1.15V.

Does just seem to be VSOC really for the APUs. I don't think they have nearly as conplex CLDOs functionality as chiplet CPUs (obviously, VDDG_IOD can't exist etc.).

 

[Simkin]

I know, I had the Steel Legend mATX for two days lol, sent it back because none of the early BIOSes were stable and finished, it's not a bad lower midrange board at all.

Less to do with the VRM parts, more to do with the LLC implementation. By all standards the Aorus AX is one of the strongest ITX boards, but the B550I Gaming Edge Wifi has really tight VSOC LLC. Different VRM parts, but about as capable. But yeah, might want to keep an eye on your VSOC. Mine used to crash just watching a movie when I was trying to run 2000 IF at 1.15V.

Does just seem to be VSOC really for the APUs. I don't think they have nearly as conplex CLDOs functionality as chiplet CPUs (obviously, VDDG_IOD can't exist etc.)

Ok, i might try 2000 IF again, what VSOC should i set?

 

[Akira1364]

Managed to decently improve on my previous best results out here in DDR3 land. It's completely stable for daily use, too.

 

[freeagent]

Just messin around..

 

[tabascosauz]

Ok, i might try 2000 IF again, what VSOC should i set?

1.15V might be a good place to start, but more important is to know how much your board droops with iGPU.

 

[Simkin]

1.15V might be a good place to start, but more important is to know how much your board droops with iGPU.

In the BIOS i have;

SoC/Uncore OC Voltage(VID)
Specify the SoC/Uncore voltage (VDD_SOC) in mV to support memory and Infinity Fabric overclocking. VDD_SOC also determines the GPU voltage on processors with integrated graphics. “SoC/Uncore OC Mode” needs to be enabled to force this voltage

External Voltage Settings and Load-line Calibration:
VDDCR_SOC Voltage
Input voltage for the processor by the external voltage regulator

VDDCR_SOC Load-Line Calibration
VDDCR_SOC Load-Line Calibration helps prevent VDDCR_SOC voltage droop when the system is under heavy loading

Where should i set the 1.15v? And do i leave LLC Auto?

 

[Simkin]

VDDCR_SOC was the correct one

 

[ThrashZone]

Hi,
Apex rabbit hole

 

[ThrashZone]

How about this:

• tRRDS 4
• tRRDL 6
• tFAW 16
• tWTRS 4
• tWTRL 12
• tWR 12
• tRFC 320
• tRDRDSCL 4
• tWRWRSCL 4
• tRTP 10

160ns tRFC is pretty lax and shouldn't require much more VDIMM than you use currently, but if necessary give it a 0.01-0.02V bump (assuming your current settings are completely stable)

Hi,
A lot of those numbers look familiar

 

[glnn_23]

Looking to reduce vdimm here and I'm not sure that Aida 64 really suffers that much going from 3800c14 down to 3800c16.

 

[Braegnok]

F4-4000C15Q-32GVK in XMP.