Share your AIDA 64 cache and memory benchmark here

[FilipM]

Thank you sir, and duly noted :salute:

View attachment 194250

Split IOD and CCD voltages, CCD likes to be low but IOD likes to be higher. VDDG CCD doesnt scale well with voltage


Some people running vSoc over 1.2V, although if that imposes Degradation issues is unknown.

PS. Test performance above 2000FCLK, it may be stable but performance can degrade - a known high FCLK bug. If that js the case, well, you gonna have to settle for less for the time being

 

[Final_Fighter]

getting whea errors at 2000mhz, backed it down to 1800mhz and dropped the timings to 16-18-18-18. will test from there.

 

[Cowboystrekk]

View attachment 194420

getting whea errors at 2000mhz, backed it down to 1800mhz and dropped the timings to 16-18-18-18. will test from there.

Try again, but refuce vddg ccd to 940mv and vddp to 900mv. They are way to high. Try 40-20-24-24 instead of all 24.

 

[Final_Fighter]

gave those settings a try and was still getting errors in event viewer. ill mess around some more.

 

[Cowboystrekk]

gave those settings a try and was still getting errors in event viewer. ill mess around some more.

Tried 1900? Whea then aswell?

 

[Final_Fighter]

these are the current settings im running now. ill work on dropping voltage once i find my max stable. no errors so far tho.

 

[Cowboystrekk]

these are the current settings im running now. ill work on dropping voltage once i find my max stable. no errors so far tho.

View attachment 194482

Okay. I urge you to try 940 ccd and 900 vddp as these often scale negatively above 940 and vddp trieves best at 860 to 900 on most system. You probably gain stability and reduce temp/consumption by lowering these 2. Soc and vddg iod are the ones that affect stability of infinity fabric. Vddg ccd is rekated to the cpu ccds and can affect cpu overclock while vddp is related to the imc which is very strong on zen 3.

 

[Final_Fighter]

Okay. I urge you to try 940 ccd and 900 vddp as these often scale negatively above 940 and vddp trieves best at 860 to 900 on most system. You probably gain stability and reduce temp/consumption by lowering these 2. Soc and vddg iod are the ones that affect stability of infinity fabric. Vddg ccd is rekated to the cpu ccds and can affect cpu overclock while vddp is related to the imc which is very strong on zen 3.

i may go ahead and start dropping voltage here sense im happy with the results. already getting a 5000 series proc to run on a b350 board is a good achievement in itself.


edit: dropped the volts down. windows booted and i have no errors as of yet. memtest is running in the back.

 

[Cowboystrekk]

i may go ahead and start dropping voltage here sense im happy with the results. already getting a 5000 series proc to run on a b350 board is a good achievement in itself.

Ah, thought you were on a 400 or 500-MB. Good job getting it to work! Voltages may be slightly different on 300-series, but vddg ccd and vddp mostly affects cpu itself and ram, not MB

 

[Final_Fighter]

cant seem to get my VSOC voltage down. just stays at 1.1875v. im trying to set it to 1.165. this is the same as SOC voltage?

edit: its probably a bios bug. ill just leave it sense the consensus is 1.2v is safe for daily but should not be pushed further. sense im below that ill just run with it.

edit 2: think i found my frequency limit for now. anything over pushes errors. time to work on sub timings then and more aida resaults.

 

[xPalis]

Crucial Ballistix 32GB (16x2) 3200 CL16

 

[Final_Fighter]

messed around with my pbo setting. +200mhz and cinebench stable. some aida results.

 

[Cowboystrekk]

Tried a different approach and got further The 21tras trick made low tRCDRD impossible somehow. This worked slightly better than last atempt.

messed around with my pbo setting. +200mhz and cinebench stable. some aida results.

View attachment 194495

You should reduce you tRFC, that will really speed things up. 250-300 should be doable depending on binning.

 

[FilipM]

Try this

From this, I altered tWTRL to 8 and tWR to 10. Next im gonna try tRRDL at 4.

1.46 VDIMM.

2T is better than 1T + GDM

 

[Cowboystrekk]

View attachment 194563

Try this

From this, I altered tWTRL to 8 and tWR to 10. Next im gonna try tRRDL at 4.

1.46 VDIMM.

2T is better than 1T + GDM

Remember I have a shitty bin Currently running 1.48V to avoid errors. But I can try.

Edit: Not a chance, booted, but unstable. Errors kept piling up even at 1.48V.

 

[freeagent]

2T is better than 1T + GDM

I haven't really experimented with GDM.. I am kind of surprised tbh.. its running pretty good. Hm.

 

[tabascosauz]

View attachment 194563

Try this

From this, I altered tWTRL to 8 and tWR to 10. Next im gonna try tRRDL at 4.

1.46 VDIMM.

2T is better than 1T + GDM

GDM is still faster, that's the point. If it's stability or odd CL you're after, well then yes 2T is better.

How come you're using the 1usmus not anta config for TM5?

 

[FilipM]

Except GDM is 2.5T and not 1.5T like everyone says Run them back to back on exactly the same config, you'll see what I mean

I've used 1usmus config, found it more reliable (consistent is the right word). Need to run them back to back.

 

[Cowboystrekk]

Except GDM is 2.5T and not 1.5T like everyone says Run them back to back on exactly the same config, you'll see what I mean

View attachment 194574

I've used 1usmus config, found it more reliable (consistent is the right word). Need to run them back to back.

Whar are the advantages of gdm vs 2t except for speed? 2t already has advantage with some oddnumbered timings. Is there none to gdm?

 

[tabascosauz]

Except GDM is 2.5T and not 1.5T like everyone says Run them back to back on exactly the same config, you'll see what I mean

View attachment 194574

I've used 1usmus config, found it more reliable (consistent is the right word). Need to run them back to back.

Still 1.5T on 4650G:


You are onto something though, because I can corroborate those results on my 5900X. Looks like AMD made some sneaky under the hood "fixes" to Vermeer.

Fortunately, it looks like just another reason why AIDA is a crappy benchmark - membench GDM is still noticeably faster:



About the same margin between 1T and GDM (~3sec), as between GDM and 2T (~3sec).

Knowing the level of consistency displayed in both applications, I think I would trust membench long before I trust AIDA lol. Funnily enough, GDM and 2T perform identically in membench on the 4650G but GDM is faster in AIDA, then 2T outperforms GDM on the 5900X but 2T is slower in membench.

 

[Final_Fighter]

lowered trfc timings and turned gdm off. set to 1t and running memtest now.

Edit. aida run for you.

 

[freeagent]

I am just using my DOCP profile for this speed, as usual, only difference it is @ 15-15-15-35 2T GDM off. I couldn't run @ 1T with GDM off, but I didn't mess with it too much..

I did run OCCT for a half hour the large data set with extreme setting. I played a little at 1900 15-15-15 on my XT a little bit and it seemed ok. I did more testing on it back then.

I'm pretty sure there is still room for improvement under the hood timings wise, but you guys got me interested for sure.

I didn't test it, just playing around..

 

[Final_Fighter]

getting that latency down. adjusted tfaw down to 16 amd a few others.

 

[FilipM]

Still 1.5T on 4650G:
View attachment 194587 View attachment 194589

You are onto something though, because I can corroborate those results on my 5900X. Looks like AMD made some sneaky under the hood "fixes" to Vermeer.

Fortunately, it looks like just another reason why AIDA is a crappy benchmark - membench GDM is still noticeably faster:

View attachment 194594View attachment 194593

About the same margin between 1T and GDM (~3sec), as between GDM and 2T (~3sec).

Knowing the level of consistency displayed in both applications, I think I would trust membench long before I trust AIDA lol. Funnily enough, GDM and 2T perform identically in membench on the 4650G but GDM is faster in AIDA, then 2T outperforms GDM on the 5900X but 2T is slower in membench.

Interesting...

If you can run something like Sisoft sandra or geekbench 3 back to back, that would be awesome

 

[tabascosauz]

Interesting...

If you can run something like Sisoft sandra or geekbench 3 back to back, that would be awesome

Memory score for GDM is marginally faster in GB3 but it doesn't mean much because GB3 likes raw freq and dual rank much more than anything else:

Just another tick in the "AIDA will be AIDA" column for me.

Honestly even more arbitrary than GB3. At least GB3 is more demanding and can function as a very quick and basic memtest for OCing. I mean, just look at the L3 cache "bug" for Ryzen 5000. I'm still surprised AMD even stooped so low as to release an AGESA version to "fix" AIDA's own deficiencies . Then there's also the fact that anyone can just as easily CPU overclock their way to better memory scores in AIDA which is just




Right, forgot to ask, what's up with the high VSOC on your comp?