I have a Ryzen 3700x that I replaced the stock cooler on. After doing so I ran Prime95 to see how well the new cooler worked. I found that all of the tests fail on Worker #3 and #4 every time.
I’ve read that this can sometimes be related to memory issues. I’ve tried stock and XMP settings with the same results. It also doesn’t make sense to me that only one core would be affected if memory was the issue.
I get different numbers for the rounding error each time, but this is an example of the output:
[Nov 2 04:50] Worker starting
[Nov 2 04:50] Beginning a continuous self-test on your computer.
[Nov 2 04:50] Please read stress.txt. Choose Test/Stop to end this test.
[Nov 2 04:50] Test 1, 480000 Lucas-Lehmer in-place iterations of M753663 using FMA3 FFT length 36K, Pass1=768, Pass2=48, clm=2.
[Nov 2 04:50] FATAL ERROR: Rounding was 0.499909461, expected less than 0.4
[Nov 2 04:50] Hardware failure detected, consult stress.txt file.
[Nov 2 04:50] Torture Test completed 0 tests in 0 minutes - 1 errors, 0 warnings.
[Nov 2 04:50] Worker stopped.
Do I have a bad core and need to RMA my CPU or is something else wrong?
CPU is at stock speeds. I haven’t tried anything more than turning on PBO in Ryzen Master in the OC arena.
@Kla3 all four slots are populated with an 8 GB stick.
@MisteryAngel under full load I’m not seeing voltages higher than 1.275 V. Sitting on the desktop at idol they’ll jump up to 1.5 V for a fraction of a second though. The board is a Gigabyte x570 Aorus Pro WiFi.
The way my case sits I have to unplug everything and pull the case out of the desk to open it. So I haven’t tried that yet. If this is for sure a RAM and not a CPU issue I can try that.
Yup so its likely throttling probably the reason why the worker failed.
You could try to set a fixed Vcore of like 1.35V ish and use llc level 2 or 3.
Then test again.
Of course monitor temps and voltages during tests.
@Kla3 I’ll hopefully have time this week to try running with each of the sticks of RAM individually. I did run the full MemTest86, 4 passes of 13 tests, without any errors.
@MisteryAngel I turned on PBO which brought the average voltage up to 1.3 V with max between 1.337 V and 1.363 V. That bumps all but the core that isn’t working up to an average of 4.1 MHz, and I’m maxing out at 75.5C. Same workers still fail. Should I still try a manual Vcore?
75°C is basically fine for a p95 stress test.
So yeah give it a go, fixed voltage like 1.35V ish / 1.356V,
LLC level 2 to start.
Maybe disabling pbo at some point when it keeps failing,
and try some manual upping on the cpu ratio.
I think that a 3700X has issues with hitting their max core speeds on all cores.
So i wouldn´t be suprised when you end up like 4.1GHz on all cores.
3800X is a better bin in that regards.
Well, I have a fresh install of Windows now. I managed to corrupt my boot drive beyond recovery.
I’m unable to get CCX0 over 4025 and CCX1 over 4125 Prime95 stable. I’m using the Peak Core Voltage in Ryzen Master. I can keep that speed all the way down to 1 V. I went up to 1.475 V trying to get another 25 and decided to stop.
I haven’t figured out where the LLC setting is in my bios, so I didn’t mess with that.
Turns out the above results were with an old bios that the mobo reverted to at some point in my screwing around. That low voltage is not stable with the latest bios.
So, stock settings are no longer Windows stable. I’m thinking more and more that I have a failing CPU. When I first got it I was able to run all the Passmark CPU benchmarks. Then the physics one wouldn’t complete starting a month or so ago. Now if I use my manual CPU settings it will run.
I bought some stuff to upgrade my HTPC on Black Friday. So I was able to test the CPU in a different system and it still errored out on Prime95. The RMA is approved and will go out in Monday’s mail. We’ll see how long the turn around will be.
The 3200G is a decent processor, but there is a noticeable difference under load vs the 3700x.