Share your AIDA 64 cache and memory benchmark here

[FilipM]

1.14V is absolutely fine for FCLK 2000

I have passed TM5 at 1.125V too, however i havent tested the effects on performance. High FCLK speeds and low voltages may seem stable, yet performance drops off. I have followed this so far and never had issues

 

[tabascosauz]

1.14V is absolutely fine for FCLK 2000

I have passed TM5 at 1.125V too, however i havent tested the effects on performance. High FCLK speeds and low voltages may seem stable, yet performance drops off. I have followed this so far and never had issues

View attachment 194682

Ah, I was looking at the screenshot about GDM, didn't realize those weren't from your PC. Some freaky 1.24V SOC going on in those pics lol

1.16V is about where I run Renoir for 2000 as well

 

[FilipM]

Yea, they aren't my pics. But hence me saying a few posts back about high voltage on vsoc...if that is good or bad, we dont know yet. Amd doesn't provide technical data sheets like Intel, which is a shame

 

[freeagent]

I was playing a little more.. I finally got four sticks @ 1900. All it took was 2T and GDM disabled. Its not as high performance per se, but its still there.. just gotta dig a little it seems..

I haven't been able to crack 1933 yet

 

[INSTG8R]

No change in my normal settings but latest BIOS shaved a few ns of my previous 65ns

 

[Taraquin]

I was playing a little more.. I finally got four sticks @ 1900. All it took was 2T and GDM disabled. Its not as high performance per se, but its still there.. just gotta dig a little it seems..

I haven't been able to crack 1933 yet

View attachment 194802

Nice Again: Fix that tFAW. It should be 4x trrds. 23 makes no sense. Either try 5 trrds, 20 tfaw or change it to 16. 23 is a useless value

 

[FilipM]

I was playing a little more.. I finally got four sticks @ 1900. All it took was 2T and GDM disabled. Its not as high performance per se, but its still there.. just gotta dig a little it seems..

I haven't been able to crack 1933 yet

View attachment 194802

I am waiting on a 2nd kit of the Vipers myself, for 4x8

How difficult do you find it? Any big changes vs 2 sticks?

PS. TCWL at 14 will be much better

 

[Taraquin]

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

It worked fine at 3800\1900@1.44V Going above 1.45V is a problem for me due to heat and insufficient cooling. Still the 4000cl16@1.44V I run now is superior with 2ns lower and 3 sec faster in dram test. 4fps more on CPU-game at SOTTR aswell. I`m not thrilled about 1.17 soc and 1.1 iod vddg though. 3800\1900 runs stable at 1.06 soc and 1.0 iod vddg.

 

[freeagent]

Ok I will try those values out, but using some of those values recommended might cause a crash, could be too tight.. but this is a different bios from the last time I tried.. we’ll see. I have a bit of a theory but I don’t want to say anything yet.

 

[Taraquin]

Ok I will try those values out, but using some of those values recommended might cause a crash, could be too tight.. but this is a different bios from the last time I tried.. we’ll see. I have a bit of a theory but I don’t want to say anything yet.

Point is the tFAW-value you use now makes no sense, it might even cause instability as tFAW should be an even value that can be divided by 4 and is always 4x trrds. 4 and 16, 5 and 20, or 6 and 24 work great, but 4 and 23 makes no sense. Try 5 and 20 first, also raise trrdl to 7 or 8. If that works try 4 and 16 + 6 trrdl.

In my experience trrds and tFAW is barely voltage/frequency sensitive, but can improve performance significantly if tightened. It is also stable at both high and low voltage. CL, tRCD, tRP, tRC and tRFC are highly dependent on voltage and must be higher when frequency rises, some of them can scale positive or negative with voltage depending on ram-die. Other performance-significant timings like tRTP, tWR, scl's etc can be about the same at 3000 or 4000MHz.

 

[freeagent]

Point is the tFAW-value you use now makes no sense, it might even cause instability as tFAW should be an even value that can be divided by 4 and is always 4x trrds. 4 and 16, 5 and 20, or 6 and 24 work great, but 4 and 23 makes no sense. Try 5 and 20 first, also raise trrdl to 7 or 8. If that works try 4 and 16 + 6 trrdl.

In my experience trrds and tFAW is barely voltage/frequency sensitive, but can improve performance significantly if tightened. It is also stable at both high and low voltage. CL, tRCD, tRP, tRC and tRFC are highly dependent on voltage and must be higher when frequency rises, some of them can scale positive or negative with voltage depending on ram-die. Other performance-significant timings like tRTP, tWR, scl's etc can be about the same at 3000 or 4000MHz.

Makes no sense to you.. this is factory settings for 2133 ram. If I get creative in sub timings this board throws fits like crazy. Its a little to "optimized" for gaming. Same with the interconnect voltage, I am undervolting it compared to what the board wants to give, and I was being told it was too high. I'm going to finish waking up and I will dive back in and see what we can see..

I let it run last night, it was ok..

Edit:

Also going down on some of those values brings me slower times in spi32m.. because I did try more than once

Coffee is kicking in now.

 

[Taraquin]

It's not what I think, it's how it works. Up to you if you wanna try, but as I said, it is generally free performance that often don't require more voltage at all. Jedec/Xmp-settings can be all over the place and also the auto-setting on MBs can make some weird settings that is not optimal.

 

[freeagent]

I think for now I am going to give up on 4 sticks. I went to go into the bios shortly after my last post and it cycled on/off once. So I am back to two sticks for now. Thanks for the coaching guys, I really appreciate it. Your efforts are paying off, slowly but surely I have a pic of your post on my phone for when I dive back in. My kids are going to be up right away, spring break is almost over.. And she is at work today so probably wont be able to do much until this afternoon. I can distract them with food for some testing time

 

[Taraquin]

The 4000cl16-16-15-32@1.44V gets rare no 6 errors in TM5. According to guide this is imc due to too high voltage. Well infinity fab is unstable unless I use 1.16V soc in bios (actual 1.144-1.15) and 1.1v iod vddg (actual 1.098) so I guess there is no way around that error

I think for now I am going to give up on 4 sticks. I went to go into the bios shortly after my last post and it cycled on/off once. So I am back to two sticks for now. Thanks for the coaching guys, I really appreciate it. Your efforts are paying off, slowly but surely I have a pic of your post on my phone for when I dive back in. My kids are going to be up right away, spring break is almost over.. And she is at work today so probably wont be able to do much until this afternoon. I can distract them with food for some testing time

That 11 cwl timing might be a problem for stability. Many MBs prefer even numbers and it's derived from tCL so keeping it the same or 2 lower is often best Do you know if you have T-top or Daisy chain?

 

[FilipM]

Killed a load of services, got the same in Windows. God damn, Armory Crate is like 1ns on its own

1.5 VDimm this time round, calling it a day here for 24/7


PS. Tweaked secondary timings further

 

[Taraquin]

I think sub 50ns is impossible for me due to my shitty binned ram and insufficient cooling for ram, but I`m satisfied with this With better cooling I might be able to run 1.5V and get sub 50ns. For now stuck at 1.44V. For fun I compared running 4.4GHz vs 4.8 and got exactly 2ns more and 1000mb less on read and copy, 500mb less on write. Safemode gives me 0.5ns lower.

 

[FilipM]

80mm molex fan at 7V...silent assasin

 

[Taraquin]

80mm molex fan at 7V...silent assasin

Biggest problem is gpu blowing 65C hot air straight at ram. But getting a different 3060ti now at a okay price is impossible :/

 

[FilipM]

I never had issues with blower cards, yes they run slightly louder but nothing unbearable. Dont know why they changed that design

Asus and Gigabyte make a 3090 Turbo model, which has a blower style cooler on it, and they run surprisingly cool. There is a review on the internets, guy stacked 4 cards and tested every combo from 1 to 4 cards - 1 card installed, didnt go over 67C on stock fan profile.

Plus they make for good tables

 

[Taraquin]

I don't have blowercard, but second fan vents air through back right at ram.

https://websetnet.net/wp-content/uploads/2020/12/fe6b38e7-59e2-49de-b564-642b82de6f43.png

 

[freeagent]

Nice guys. So when you run TM5 do you have HWinfo open too? I can run my IF at 2066 and my mem is good. TM5 runs perfectly fine, shows no errors, but HWinfo does.. right at the bottom. For me its the IF and not the mems that are bad over 2k IF

 

[Taraquin]

I tried 4066/2033 with same ram timings and a bit more voltage, but got perf regression for some reason. Seems I can run soc at 1.12V and vddg iod at 1.04V without perf penalty at 4000/2000. If I try soc at 1.1 or below my CB20 loses 1% pr 0.01V, but everything is stable. Will try to lower vddg voltage more to lower temps and check if stable.

 

[FilipM]

Nice guys. So when you run TM5 do you have HWinfo open too? I can run my IF at 2066 and my mem is good. TM5 runs perfectly fine, shows no errors, but HWinfo does.. right at the bottom. For me its the IF and not the mems that are bad over 2k IF

I do check for WHEA both in HWInfo at the bottom and through event viewer. In my run, didnt get any.

A failure in TM5 can be infinity fabric instability too, so generally if it is stable in TM5 its allright.

You can run Prime Large FFT/OCCT for infinity fabric instability.

 

[freeagent]

Oh yeah I have 2k nailed down in occt, but having a hard time gaining stability any further than that.

 

[FilipM]

Increase VSOC as a starting point. Ive found that 40 50 and 65mV offsets work on mine:

EG. VDDP = 900 ; CCD = 940 (1x40mV up from VDDP); IOD = 1060 (3x40mV up from IOD); VSOC = 1140mV (2x40mV up from IOD)

or 50mV (900-950-1050-1150) might work


Play around