-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Poll: Ryzen 7950X3D, 7900X3D, 7800X3D
- Thread starter kindai
- Start date
Associate
Anything I should be doing with the 7950x3d on stock atm
I'll answer my own question! It failed Linkpack after 3 loops at -30. It's survived 10 hours at -27, has upped it's load clocks to ~4750, and stayed under 78 degrees the whole time. Total gains of about 5% vs out of the box, which isn't a lot, but it's also not nothing.
Then failed Prime 95 within 2 minutesI guess I'll start adding some back and see where it ends up.
Edit: sadly failing Prime at -22 also. Oh well, keep going up until it passes I guess. Weird that Linpack would run all day then Prime can't do 30 mins...
Or stop running unrealistic stress tests?
Man of Honour
Might be HWINFO64 related, every now and then i will click the shortcut to open it and the stat window will not appear but it shows active in the taskbar.
Caporegime
Disable thunderbolt in the BIOS and thank me later.
-15 all core is a decent place to start. Y Cruncher 1-7-0 for at least 12 hours was used for testing the cores. You can use core cycler too, but for me I found better stability with y cruncher but some people report different so YMMV.
Idle random restarts likely mean one of your two best cores (can be on either CCD depends on if the workload if using the cache or frequency CCD) has too much negative offset at idle/low load, so dial one core back a few notches. May take a few attempts before you either guess the correct core that **** the bed, and reduced the margin enough to stop the restart.
Disabling Thunderbolt has fixed that instantly, thank you.
I started off all core 15, left overnight on Y Cruncher as you recommended, and also SHA3 which passed. Made sure I tried idle things also, and games. I got all core 25 the same testing and passed. Then tried -30 and crashed instantly. I am now doing per core and testing one at a time -30 and leaving all others on -25 until I complete.
I have followed all your recommendations and have had no issues with games choosing the correct CCD etc. Everything running perfect.
Associate
I don't think any stress tests are unrealistic.
Either the CPU works whatever is thrown at it, or it doesn't. If it doesn't, sooner or later you're going to have crashes whenever the conditions are right.
This, really. If it fails under any scenario ever, it will eventually fail in normal usage - and I will be quite cross if it fails while writing an image file I've been working on for 8 hours and corrupts it. Given that I'm messing with undervolting, I of course have no problem just bringing it up until it's solid. First time in my life I've had the option of building it in the spare room and leaving it on the 'test bench' while I continue using my daily driver - it's nice!
In the end, I settled on -18 in curve optimiser. It's not as much as I'd hoped, but it is Linpack and P95 stable for the 18 hours I gave them. P95 runs oddly cool, never passing 70 degrees, while Linpack just barely touches 80. Cinebench meanwhile instantly goes into thermal throttling territory.
Not a problem, Cinebench isn't exactly among my daily programs, and I suspect the gaming load will be 50-60 degrees, but it's interesting what seems to stress this CPU and what doesn't. I'm used to my 8700k, where the hottest test is P95 and everything else is a few degrees cooler. Curious as to what it is about Cinebench that loads Ryzen so hard.
Good shout, I will give it a pass with Y Cruncher too. Had forgotten that one!
Last edited:
I am finding idle a little bit harder to find instability. I passed overnight Y Cruncher, multiple runs of SHA 3 and today when idle just on the desktop it shut down.This, really. If it fails under any scenario ever, it will eventually fail in normal usage - and I will be quite cross if it fails while writing an image file I've been working on for 8 hours and corrupts it. Given that I'm messing with undervolting, I of course have no problem just bringing it up until it's solid. First time in my life I've had the option of building it in the spare room and leaving it on the 'test bench' while I continue using my daily driver - it's nice!
In the end, I settled on -18 in curve optimiser. It's not as much as I'd hoped, but it is Linpack and P95 stable for the 18 hours I gave them. P95 runs oddly cool, never passing 70 degrees, while Linpack just barely touches 80. Cinebench meanwhile instantly goes into thermal throttling territory.
Not a problem, Cinebench isn't exactly among my daily programs, and I suspect the gaming load will be 50-60 degrees, but it's interesting what seems to stress this CPU and what doesn't. I'm used to my 8700k, where the hottest test is P95 and everything else is a few degrees cooler. Curious as to what it is about Cinebench that loads Ryzen so hard.
Good shout, I will give it a pass with Y Cruncher too. Had forgotten that one!
This is going to be a very long process for me.
Caporegime
See my previous post. It's most likely your two fastest cores on CCD1, possibly even just one of them. Find your best core and reduce the margin by -5 and see if it it happens again. If it does, try the same with the second best core and repeat the process. Eventually you'll figure out which one it was and you can further fine tune.
I am slowly progressing. Do you know any test for idle? That’s where I am finding it more difficult to test. I can be stable in stress tests then crash just browsing.
Also Matt did you do a negative curve optimiser on both CCD or just the 3d?
Then you should run the tests forever. Sooner or later the system will fail.
Caporegime
No, there is no test but using the system under light load, browsing etc for a few days.
Ok Thanks.
Would you recommend doing curve optimiser for both of the 3d and non 3d? Most users seem to just do the 8 3d cores and finish.
Caporegime
I'd suggest doing the Cache cores first, to maximise margin as those will see most usage, depending on the games you play.
Funnily enough my main game is Football Manager 2023 and game bar makes the game use the frequency cores in this title as the game performs better with higher frequency.
Most of my other games all run solely on the cache cores though.
I was very concerned coming from only ever using Intel CPU about people having trouble with the 7950x3d. It couldn’t be any easier coming from my own opinion with this.
The only thing I really want to achieve after I’ve tuned the curve optimiser is bringing the memory latency down a little. Maybe you can help with this when I reach that part?
Thank you again Matt.
Caporegime
AM5/X3D has been the easiest platform/CPU I've tuned. Very easy to get top tier performance without even doing anything.
And another 10-15% on offer if you want to tune things, all at very low power and temps.
Happy to help with suggested timings, just share a Zen Timings screenshot.
The temps are a big surprise consider I’m only using an aio. In Warzone I’m in the 50c with super flat frametime apart from in the plane at the start.
I’m currently testing render worker count on that game and I’m unsure between 7 or 8 it is very close. Have you tested Matt?
Thank you. I will hope to be at the memory tuning stage soon enough. Maybe a week.
I also have not found SHA3 as good as Y Cruncher for finding instability but I’ve seen people saying is best. For me not so much. Y cruncher is better but I do some after completing anyway.