Processors

[NewAMDGuy]

So i recently bought a new pc with following parts:

cpu: ryzen 9 3950x

gpu: rtx 2080ti

psu: beQuiet 750w (i replaced that because i thought my old PSU was not good).

bios: f33h

mobo: gigabyte auros pro x570 rev 1.0

Now the problem:

so when the PC was new, I was able to play and do everything with no problem, then after a while random blue screens appeared. That was before i updated my bios. I updated drivers and bios and those disappeared but the problem was bigger. I kept getting Event 41 power failure and I didn’t know why. So then I kept digging and forum users suggested me to replace my chieftec psu with a better one which I did. The thing is, I can barely replicate the random reboots reliably. Sometimes no random reboots occur for days and I play ever game on max and render alot of stuff for hours. But then randomly, or while texturing or editing in Photoshop I get random reboots. Sometimes even just by casual browsing. Its so unreliable. So I download occt and tested my GPU, CPU VRAM etc and it all worked with no problem UNTIL i tested the power. Whenever I test the power the computer would instantly reboot. I tried alot by now, update bios, fiddle the bios settings, disabling this and that but it doesn’t help. When I now stress test the POWER on OCCT it still immediately reboots. I googled and it seems to be an AMD issue and I dont have the warranty at hand. I was a intel guy before and while Amds philosophy is great and all, i never ever had those annoying problems with an intel powered machine. Its just so disappointing, I took a leap of faith once and got spat on... 

I really don’t know what to do, I don’t know much about bios or hardware as its not my expertise per se. But for me to literally do so much work to get my pc stable is just not a good look. I would appreciate any input. 

[turion]

Have you tried slightly increasing the voltage? I had a similar experience with an R5 3600. You should update the bios to F33, I would suggest not using the version with a letter at the end. Also make sure you have the latest chipset drivers and check your power settings in the control panel.

[quocbinhnguyen016]

Same issue with me. Really disappointed. Could anyone help me?

My specs:

• CPU: AMD Ryzen 9 3950x.
• Mainboard: ASUS TUF GAMING X570-PLUS.
• RAM Desktop G.Skill 32GB (2x16GB) DDR4 Bus 3200 Mhz F4-3200C16D-32GTZ.
• SSD 250GB WDS250G2B0C.
• PSU: Cooler Master MasterAir MA620P.
• VGA Gigabyte Geforce Gtx 1660 Super 6G.
• Power: Coolermaster MWE 750 BRONZE V2 230V.
• CASE SAMA DARK SHADOW.

[jglathe]

Here's my experience. There seems to be some variation in the chips where there are edge cases where they are not or almost stable. I had this issue with a R9-5950X and msi boards (Unify x570 and Tomahawk x570), which was relatively stable (with Linux) on an ASRock board, but not completely. RMA'd the chip, the replacement is very stable on any mainboard - now on the Tomahawk X570. I also have 2x R9-3950X running, since 2020. One developed a strange hang pattern I cannot really explain (complete machine goes unresponsive without log entries in low-load situatios, more than once a week). Its on the Unify X570 board with 128GB of RAM and lots of SSDs (database/VM work). A hang there is not really welcome. Changed the PSU (700->1000W), the 700W PSU got RMA'd (very nice of BeQuiet!, didn't really ask for it). The hang pattern was still there. The GPU got changed, too, no effect on the hang pattern. Exchanged the two R9-3950X chips against each other. The "edgy" one is now in an ASRock X570M Pro4, same OS, and is... stable, 24/7. The exchanged chip and the Unify board is also running stable 24/7. Both boards have the latest stable BIOS, the same brand of RAM (AData CL22). Still a little testy because of this, but seems okay now.

These chips are really demanding regarding power, yes. Maybe there lies the problem. A R9-3950X and a 2080Ti is definitely too much for a 750W PSU when you hammer CPU and GPU. I would go for bigger ones (1000W should be a good choice). The 5950X / RTX3090 machine has a 1200W Dark Power 12 PSU. No issues, but only after RMAing the chip.

From what you write I would expect aging components, which would be PSU and maybe even the VRMs. My choice was trying out other components in these edge cases, after ruling out under-specced PSUs and thermal issues.

Mostly seen in the R9 class (105W TDP), btw. R5, R7-3700X - rock stable and fast regardless of board. Yes, it's an expensive hobby.

[jglathe]

Update to add: It's getting weirder. Again the suspicious hangs (CPU, PSU, GPU already exchanged), now Mainboard changed. Same hang after a few days. Currently testing RAM with memtest86. Testing 128GB of RAM takes A LOT of time. That leaves - grounding issues with the case (Fractal Design R6 C), and the OS (Windows). I will change the power chord, just in case, but that's a pretty unlikely one. All the other machines in the room are not flaky, similar hardware, same OS. A bit of a mystey.

[jglathe]

Update #2: I thought I found it, a misconfiguration in the RAM timings. Nope. Today, I literally changed the freaking power cord  and use a different outlet, since I've run out of parts to change. WTH. We'll see. Anyway, I have now enough spare parts to do *another* build with a 5950X, 64GB RAM, MSI MPG Gaming plus (7C37), and a 1050Ti (yeah, I know. Nice card, though, for a "workstation"). Oh, and a few power cords.

[atmosphere]

At what speed is DDR4 running, and two or four sticks? 

If you run them on JEDEC-specs@2133, do the issues persist? 

You checked CPU temps & voltages right? Because .. 750W PSU<=>450W GPU, 50W mobo+ram+storage, 350W CPU (peak draw for all est.) gets close. If you got it swapped for 1KW PSU, it should not be an issue.    

[jglathe]

JEDEC specs for these are 3200... M/B auto-detects them with 2666. I ran them at this speed, too (no it was not the power cord, to everybodys surprise) - same issue. 4 sticks, 4x32GB. Temps all waaay down, 1kW PSU, top of the line air cooling. Issue persists irrespective of M/B, CPU, GPU, PSU. No flaky part. Now, I have changed the RAM to a different 64GB kit, which is 3200 XMP, 16-20-20-40. No issue so far, the machine got some use over the weekend. The remaining 4 sticks are now in the test setup and are getting tortured with memtest86 pro (Passmark). ~10 rounds so far, no errors, 40hrs of testing. I think I started with the autodetected 2666. Interesting tidbit is, the test setup has a 5950X, and the RAM benchmarks with 38GB/s, slightly slower than the L3 cache. On the problem setup it's a 3950X, RAM benchmarks with 18GB/s, slightly faster than the L3 cache. Go figure. In the end the test setup will swap to the work machine, I guess. A RAM edge case.

[jglathe]

The saga continues... The problem machine is now stable with the 64GB XMP 3200 16-20-20-40 kit. The 128GB RAM on the test setup ran for 10 cycles of memtest86 with no issues at all. I swapped the 5950X in this one out against a 3900X, and it's running now the same cycles again. In this case, too, the RAM bandwidth shown by memtest about halves against the 5950X - 18GB/s vs 38GB/s. That's some curious thing. L3 cache on the 3900X also says about 17GB/s, FCLK is 1600 (usually always is). The board is a MSI MEG X570 Unify.

[hrpuffnstuff]

IDK what to tell you.  My G.Skill 3600 ram has been flawless since day one.  Started on a 3900x and graduated last year to a 5950x.  Using an EVGA Supernova G2 1300w ps, overkill but I don't care, power is stable even with an xfx speedster merc 319 rx 6900 xt black at full tilt and all cores at 4.5ghz.

My biggest problem is the AGESA issues that have existed ever since 1.2.0.4 that have supposedly been address by 1.2.0.6b.  Those bugs made my system feel weird with random stuttering and usb shutdowns.

[jglathe]

Well. There is only one machine being "unstable" in my zoo of several, which is a bit odd. The RAM bandwidth thing I noted seems to be a measurement artifact (i.e. wrong algorithm/calculations), so of no concern. I bought a license for karhu RAMtest, and it verified for the "problem" build with the 64GB (Patriot Viper 4 Blackout 3200 16-20-20-40) with 10000% coverage.

The 128GB (Crucial 3200 22-22-22-52, no XMP) from the "problem" build is currently testing in the test setup. 8 hours through, no error. From the vibe I'd say I'll see no error. I managed to shuffle around my available kits and I can replace the 128GB with either G.Skill Ripjaws V 3200 or with Kingston HyperX 3600 kits. So... the problem kit will not go back into database and VM work. Too risky.

[jglathe]

Now that the dust has settled a little, conclusion. It's an Edge Case with these boards. The "problem" RAM is now working happily (? who knows) in an ASRock Board running TrueNAS and hosting VMs. No issues. The "problem" build has now 128GB G.Skill RipJaws V RAM and is working stable. Also no issues.

[elstaci]

If it reboots as soon as your run OCCT PSU Test which is running both the CPU and GPU Stress Tests at the same time means you have either a Overheating or under-powered or defective hardware (Motherboard, PSU, CPU, GPU, RAM, etc).

Try using a different GPU card if you have one around and see if it continues to crash using OCCT PSU Test.

When you ran either the GPU or CPU OCCT Test did you keep a close eye on the Temperatures and PSU Outputs and Fan speeds?

What type of CPU Cooler are you using for your 105 Watt TDP Processor?

Sometimes when the CPU and GPU are stressed it might produce high Power spikes that may cause the PSU to shut down the computer.