Overclock.net banner
1,201 - 1,220 of 14,355 Posts
So i have received my second 2x16GB SR set and have done a full performance comparison between the different sets individually and mixed together for a dual rank setup (4x16GB), guess i can share the information ive learned by playing around with these for a few days now..

My main interest for this comparison were SR vs DR as i have seen little data on this so far.
Some keydata first:
  • All benchmark were run at a static CCD0-5000mhz / CCD1-5000mhz cpu clockspeeds
  • All the different memory configurations were maxed out on my system. (both timings and FCLK) (still looking for a better CPU)
  • All memory sticks in this comparison is Hynix m-die
  • Performance is compared in Aida, dram calc, linpack xtreme, Corona 1.3, Performancetest 10.2, OCCT, y-cruncher 2.5b and the XMR Monero miner
First off we have my EXPO set:
VENGEANCE® RGB 32GB (2x16GB) DDR5 DRAM 6000MT/s C30 AMD EXPO Memory Kit
Font Screenshot Technology Software Number

My rog crosshair x670e hero could not unlock this PMIC -> limited to 1435mv maximum.
Memory chips on these sticks seems to be the best binned as they could run the tightest timings at 1435mv, but since this is the limit they lose overall in ultimate performance.
Maximum SPD HUB temp after ~5hours karhu ram test @ 1425mv with two noctua 40mm fans blowing on them = 39 degrees with stock cooling.
Full performance screen:
Product Computer Font Screenshot Software


Next we have my XMP set:
VENGEANCE® 32GB (2x16GB) DDR5 DRAM 6600MHz C32 Memory Kit
Font Screenshot Software Technology Electronic device

Unlocked PMIC
1530mv needed to run these timings with stability intact .
1 hour karhu ram test @ 1530mv with two noctua 40mm fans blowing on them = 35 degrees with barts custom copper heatsinks
Performance screen:
Screenshot Font Software Terrestrial plant Computer


Lastly dual rank 4x16GB (both sets mixed)
Product Font Screenshot Technology Software

1435mv since locked PMIC
Could only run FCLK @ 2067mhz, and needed to increase VSOC to 1400mv to keep stability. (higher did not help)
Performance:
Font Screenshot Software Computer Technology


Conclusion:
There seem to be no performance benefit in running dual rank memory on this AM5 platform, atleast not with 4x16GB (which should be exactly the same as 2x32GB)
FCLK get stressed much harder with this memory amount and needed to be reduced to keep stability.
Ultimately my 2x16GB XMP set with unlocked PMIC performed the highest when totally maxed out, thanks to higher voltages allowed. I'm back to running this set solo again.

-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Can share some other settings/screenshots also since i'm already posting: (Better latency since not limited to static 5ghz anymore)
Highest Aida numbers i can put up: (unstable with GDM disabled)
Font Screenshot Electronic device Multimedia Software
Font Screenshot Electronic device Technology Software


Highest Aida with full stability in y-cruncher all stresstests enabled
Font Screenshot Technology Software Electronic device

Full stability-testing screen:
Azure Font Screenshot Line Software
 
I'm running the Asus rog strix-e x670e. Bios version 0705.

Very similar to the hero.

When I set my vid and my vddcr CPU voltage, it seems like I get weird behavior in hwmonitor. 1) vid bios sets the vddcr CPU voltage reading in hwmonitor. 2) vddcr CPU voltage in bios sets vcore for mobo in hwmonitor.

Also, setting llc seems to have no affect. Tested levels 3 and 5, and vdroop never happens.

Has anyone had behavior like this?
 
  • Rep+
Reactions: xeizo
@domdtxdissar why bios change between 0803 and 0045?
I have also the Hero, can you share the 0045 Bios? Did not find that :D

This is possibly my best stable setting currently with 7900x and Hynix A-Dies:
Karhu 10k with ~293 MB/s
View attachment 2580345
View attachment 2580346
View attachment 2580347
Sorry i cannot share it, 45 it just a earlier testbios for selectable tREFI.
It perform 100% same as now released 0803..
(i'm using 0803 now)

I'm sure you understand seeing as you are running ZT 1.2.8 :)

This is possibly my best stable setting currently with 7900x and Hynix A-Dies:
Karhu 10k with ~293 MB/s
Wow thats fastest a-die ive seen in this platform yet (y)
I also have a pair of green PCB china a-die laying around but haven't really tried them since PMIC seem to be locked currently (your seem to be the same?)
Hoping for maybe future bios revisions can unlock these PMIC's 😇 (also my expo memory above)
 
I'm sure you understand seeing as you are running ZT 1.2.8 :)
Nothing special with that. There is the discord of the maker of the tool, where I have it from ;)

pmic is not locked
I could also boot 6200 @ CL 26, but at least for now it would probably consume more than 1.5v on memory, that’s why I stopped testing there.
Highest clock stability was 6400 at bios 0801
Best results with trefi enabled on 0803 with 6200, unfortunately this bios struggles with 6400 and did not find the culprit yet

bootable is also 6600 but not stable, sometimes not even Aida stable
It gets a bit easier when uclk is half, but not invested too much time into that since 6400 also not stable yet
 
Discussion starter · #1,206 · (Edited)
Nice seeing some more extensive memory testing, looks like ~90GB/s is the upper end of what most can expect at the moment. Yeah, it's distracting seeing Asus validating DDR5-10000 for Z790 but it is what it is, one has to look at the platform as a whole. At least we have no 400W load CPU in our rigs ;)

I'm happy my HyperX Fury Beast at least have unlocked PMIC, 1.46V in bios is 1.47V in HWINFO64, it's been completely stable as long as I don't try to raise FCLK which looks weak on my CPU. Hmm, haven't tried async BCLK yet, but it has caused problems with M.2 for me on all previous platforms so I'm not that eager even if that problem may be solved by now.
 
Discussion starter · #1,207 ·

Now leads to a new test bios, 0804

"meanwhile updated links in first thread to 0804 -> Patch A since some people had issues with Patch C, + D for just some thunderbolt updates. "
 
  • Rep+
Reactions: LtMatt

Now leads to a new test bios, 0804

"meanwhile updated links in first thread to 0804 -> Patch A since some people had issues with Patch C, + D for just some thunderbolt updates. "
I think I'll try this new 0804. Wonder what the issues were with patch C+D?
 
Discussion starter · #1,210 · (Edited)
I think I'll try this new 0804. Wonder what the issues were with patch C+D?
New bios is patch A +D, patch C is removed. I suppose it is to reinstate the more enthusiastic boost behavior of 0611. Or maybe not, the good boost was even before patch A from what I recall. I read some had instability with patch C, could be that.
 
Discussion starter · #1,211 ·
Best latency inside normal Windows 11 so far, without going into safe mode

Font Electronic device Screenshot Technology Software
 
  • Love
Reactions: LtMatt
Discussion starter · #1,213 ·
FCLK 2067, a little better bandwidth and passes 32 core Threadripper in PhotoWorxx

Font Electronic device Screenshot Software Multimedia

Computer Font Personal computer Rectangle Technology

Font Technology Screenshot Multimedia Electric blue
 
  • Rep+
Reactions: domdtxdissar
Discussion starter · #1,215 ·
@xeizo trefi higher not possible for you?
Or do you have problems?
I haven't tried higher as domdtxdissar reported 60380 was the highest he could run(stable), so I spared myself, I want stable not benchmark
 
Discussion starter · #1,216 ·
Looks like there are some problems with test bios 0804, just beware, it's no replacement for a stable bios. Just saying. Moving in the right direction though, waiting for the next test bios! (y)
 
  • Sad
Reactions: LtMatt
Discussion starter · #1,220 ·
  • Wow
Reactions: LtMatt
1,201 - 1,220 of 14,355 Posts