• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

AMD Ryzen 9 9950X3D Performance, Overclocking and Cooling Thread

Update, still stable with the exception of one issue.

Either memory stick in slot 1B will not work when doing a system reboot. A shut down and start up works every time. This was not an issue with the Corsairs and for giggles I swapped them back in and instant errors in OCCT within 7 seconds with no other changes at 6200.

I also noticed another bug with this board, doing a CMOS reset will cause voltages and fans speeds to show incorrectly in HW Monitor section of the bios as well as making the fans run at minimum speeds until the PC is unplugged and plugged back in.

So looks like ill need to get new AMD cert ram, ill see what Micro Center has to go pick up that is also on the QVL list.

EDIT: Might be a winner here: https://www.microcenter.com/product...top-memory-kit-f5-6000j3040g32gx2-tz5nr-black
 
Last edited:
LOCO LAPTOP said:
I slapped my Intel Gskill set from my 14900k setup, F5-6400j3239g32gx2-rs5k (SK Hynix A-die) and it actually boots at 6200mhz (and 6400mhz but not stable) Ran 10 minutes of OCCT with no errors then raised the FCLK to 2133 and same thing. I know, that doesn't mean it's stable. But I'm already ahead with this set and finally in the 60's for latency.

Gonna stress test this overnight and see how it goes.

View attachment 724546
Click to expand...
There is no way even your other kit should have been running so hot and unstable, maybe it was actually bad? Seems you found the main issue though.

Personally, I'd try and snag a 6400 (or better) kit and attempt 6400 1:1 at CL30. If you can run 2133 FCLK stable, you might luck out with the mclk and uclk with the right SoC voltage. Mine takes 1.275V and passes OCCT for hours at 6400 1:1. I attempted 6600 at 1:1 multiple times and that was a no go for me. My FCLK starts to crap out above 2167Mhz.

Is it mind blowing? No. But it will help those 1% lows ever so slightly and if your IMC can do it, some bragging rights on the 9950X3D.

The X3D CCD really does make the ram almost irrelevant though as long as you have decent latency.
 
So if anyone is curious, here is where I ended up with my PBO and Cinebench scoring... pretty nice boost compared to "Stock" or default PBO settings. I eventually landed on -20 CO, 5X Scalar, +100Mhz FBoost, tested with OCCT and P95, 100% stable.... finally! :)

CineR24_12.jpg
CineR23_12.jpg
 
  • Like
Reactions: blurp
like this
Update:

Ended up getting 6200 to work on the Corsair kit after some timing changes, Passes with flying colors on TM5 and OCCT. 6400 is not gonna happen with this CPU unfortunately.

Once I got it stable for a 1 hour pass I change the timing a few more times. Below is what I ended up with that is a known 8 hour pass with TM5 and 5 with OCCT. I think I found the limit for my setup. Any lower timings in most options= crash or higher L3 latency.

1745277091741.png


1745276473648.png
 
LOCO LAPTOP said:
Update:

Ended up getting 6200 to work on the Corsair kit after some timing changes, Passes with flying colors on TM5 and OCCT. 6400 is not gonna happen with this CPU unfortunately.

Once I got it stable for a 1 hour pass I change the timing a few more times. Below is what I ended up with that is a known 8 hour pass with TM5 and 5 with OCCT. I think I found the limit for my setup. Any lower timings in most options= crash or higher L3 latency.

View attachment 725099

View attachment 725098
Click to expand...
Nice! What PBO settings did you land on as well?
 
LOCO LAPTOP said:
-20 CO, 5X Scalar, +200Mhz PBO

System is stable on 10X but didn't see a positive on running it that high besides a few points in 3dmark.
Click to expand...
Yeah, 5X felt like a happy place to leave it for gaming. 10X provided nothing in games, just some extra points in cinebench and 3dmark. It mostly added a little bit of heat that high. 5X seemed to be a decent enough jump from 1X though without any real heat or voltage. I think at 10X it only adds .025V, so I'm assuming 5X is maybe .0125V. It looked roughly the same as 1X to me logging the voltage tho, so I'm happy. Glad you got it tuned in finally! This CPU is a computational beast!
 
I moved to the 9950X3d from 7950X3d - the jump in both multi threaded productivity and gaming was bigger than I expected. I'm fully GPU limited now with my 4090 in VR gaming which wasn't always the case with the 7950x3d, and it's reduced photogrammetry rendering times by around %20. That's significant with a problematic dataset as I might have to run 4 or 5 iterations to get the cameras to align and create a good model, which is real time saving - basically I can run eg 6 iterations in the time it would have taken to run 5 previously. I have done nothing besides load the latest bios, tune the ram using buildzoid's guide and install the latest AMD chipset drivers. I have yet to test it's performance in llm inference but that's very ram bandwidth limited so I'm not expecting much.

Temperature wise it's about the same as the 7950x3d if not a little cooler, which I attribute to fresh TG paste.
 
bobzdar said:
I moved to the 9950X3d from 7950X3d - the jump in both multi threaded productivity and gaming was bigger than I expected. I'm fully GPU limited now with my 4090 in VR gaming which wasn't always the case with the 7950x3d, and it's reduced photogrammetry rendering times by around %20. That's significant with a problematic dataset as I might have to run 4 or 5 iterations to get the cameras to align and create a good model, which is real time saving - basically I can run eg 6 iterations in the time it would have taken to run 5 previously. I have done nothing besides load the latest bios, tune the ram using buildzoid's guide and install the latest AMD chipset drivers. I have yet to test it's performance in llm inference but that's very ram bandwidth limited so I'm not expecting much.

Temperature wise it's about the same as the 7950x3d if not a little cooler, which I attribute to fresh TG paste.
Click to expand...
The major change was the move of the cache to under the CPU die itself, allowing it to cool much better and allow proper overclocking and better PBO behavior. That's an awesome uplift you got from only 1 generation behind. I came from an AM4 5950X! lol.
 
bobzdar said:
I moved to the 9950X3d from 7950X3d - the jump in both multi threaded productivity and gaming was bigger than I expected. I'm fully GPU limited now with my 4090 in VR gaming which wasn't always the case with the 7950x3d, and it's reduced photogrammetry rendering times by around %20. That's significant with a problematic dataset as I might have to run 4 or 5 iterations to get the cameras to align and create a good model, which is real time saving - basically I can run eg 6 iterations in the time it would have taken to run 5 previously. I have done nothing besides load the latest bios, tune the ram using buildzoid's guide and install the latest AMD chipset drivers. I have yet to test it's performance in llm inference but that's very ram bandwidth limited so I'm not expecting much.

Temperature wise it's about the same as the 7950x3d if not a little cooler, which I attribute to fresh TG paste.
Click to expand...
That's exactly what I was expecting this cpu to do. Glad to see it pan out.

In terms of inference, if you're running torch with avx512 you should see decent cpu based performance. That full 512 bit simd bus is pure gold.
 
You must log in or register to reply here.
Back
Top