Literally burning through AMD Ryzen 9 5950X CPUs

[Krilega]

Hey guys,

 

as there are a lot of enthusiasts here, I wanted to know if anybody has experienced dying High-End Zen3 chips with a similar setup.

 

I literally keep killing a 5950X CPU every 2-4 weeks of moderate usage. Use cases of this machine are mostly programming / compiling, sim racing in VR (ACC) and streaming/watching YT/Twitch videos. These workflows are very often heavily single threaded with a couple of sub threads, sometimes parallelized. I honestly don't know, what I'm doing wrong here and I never experienced something like that in nearly 3 decades of upgrading and building my own systems. I never had a CPU dying in any of my systems before and as there is a pattern here, it has to do something with my workflow and/or the rest of my system.

 

The 5950X CPUs work fine after RMA for 2-4 weeks. Stupidly while sim racing and having a Twitch stream running in background, the system suddenly reboots (while it was working perfectly for several weeks before). If that happens, the CPU is basically broken and random BSODs and reboots occur in every workflow. This happened to me 3 times in a row with the same pattern. Reinstalling Windows 10 is not possible anymore at this point, as it keeps rebooting the system unexpectedly and aborting the setup procedure. The first RMA I swapped my PSU. The second RMA I replaced the memory kit with a QVL kit. The third RMA i didn't even bother to enable XMP and left nearly everything on default (except disabling a couple of useless (for me) motherboard features like Audio, TPM, Legacy/Compatibility Mode and Ethernet / yes, I'm using WiFi6 with ~1.5 to 2Gbit/s real downstream...). I specifically disabled PBO each time on the board and didn't use AutoOC, Curve Optimizer or similar features. I didn't touch voltage settings and didn't disable performance or power saving features like turbo or c-states as I expect them to run out of the box. Core temperature never exceeded 80°C in a permanent HWInfo64 monitoring (acoustic warning setup for core temp and pump speed).

 

Windows 10 event view does only report Kernel Event 41 - unexpected shutdown. No other errors except the typical boot up device error caused by the disabled motherboard features.

 

 

tested Hardware:

- Ryzen 9 3950X -> still works perfectly fine after 1 1/2 years of similar usage in this system and is my current backup CPU

- GPU (if important) RTX 3090 Reference

- "Overkill" watercooling setup (radiators 480x45mm push + 280x86mm push+pull)

- ASRock X570 Phantom Gaming X, BIOS 3.40, 3.80 and 4.00

- Memory: Corsair 4x8Gb SR 3600MHz CL16 @1.35V and G.Skill 2x16Gb DR 3600MHz CL14 @1.45V

- PSU: Corsair HX850 and HX1000

 

If anyone may have knowledge or suggestions what I can try or heard of similar issues, please let me know.

 

Thank you very much

Krilega

[Sauron]

I guess if you RMAd everything else the motherboard is the last remaining suspect... though it's odd behavior to be sure and I couldn't tell you why the 3950x is fine.

[emosun]

7 minutes ago, Krilega said:

This happened to me 3 times in a row

I think at that point I would conclude your board is just frying these cpus as cpu's..... are typically the most reliable component in any computer.

Your board could be feeding these cpu's with incorrect voltages BUT reporting itself as being fully working.

[jaslion]

25 minutes ago, emosun said:

I think at that point I would conclude your board is just frying these cpus as cpu's..... are typically the most reliable component in any computer.

Your board could be feeding these cpu's with incorrect voltages BUT reporting itself as being fully working.

I'd also say this especially since the 3950x is fine. The 3950x has different needs and the motherboard can handle that fine however when the 5950x is added into the mix it seems to start messing up. Might be as simple as a single pin or tiny little chip freaking out.

[ProjectBox153]

The entire issue seems to point at the motherboard. I'd swap it out if I were you. 

[emosun]

4 minutes ago, jaslion said:

Might be as simple as a single pin or tiny little chip freaking out.

exactly , there's a lot of components on a board that aren't monitored at all.

It's never impossible for some weird diode or resistor or chip to just be doing something wrong just barely. The board has thousands of traces and components , that fact that so many work without problems is more amazing than the ones that don't really.

[ForwardVoltage]

Asrock auto volts moment????

[Krilega]

Yep it's odd, since the 5950X should overall have less power requirements as the 3950X (~20W less PPT), including the SoC.

VCore  is typically near 1.5V at very low loads and in line with VID. VSoC runs at 1.2V since AGESA 1.2.0.0, before 1.1V (seems AMD brute forced FCLK stability again)

 

Since the board is rather old (july 2019) and the chipset cooling has been modified by me to fit a proper Noctua fan, there is nothing to RMA here...

 

Board did work fine otherwise though, except typical ASRock things:

- missing BIOS options

- BIOS options not working

- other massive design flaws, like

- wifi antenna connectors too close to fit custom antennas

- extremely bad chipset cooling

- putting the front USB-C connector in line with primary PCIe slot, so your only option were slow mid tier GPUs to use this connector at time of release (thanks Nvidia for 30series reference PCBs)...

 

I hope AMD will replace another chip though as I got contacted by an engineer the last time and he kind of excluded the board as the 3950X still runs fine. Will try to replace the board anyway, but definitely no ASRock one anymore.

 

Would be interesting to know, if anyone had fried a 5950X as well or has these random reboots, as Google search is full of 5950X reboot issues.

[iCoNrant]

5 hours ago, Krilega said:

Yep it's odd, since the 5950X should overall have less power requirements as the 3950X (~20W less PPT), including the SoC.

VCore  is typically near 1.5V at very low loads and in line with VID. VSoC runs at 1.2V since AGESA 1.2.0.0, before 1.1V (seems AMD brute forced FCLK stability again)

 

Since the board is rather old (july 2019) and the chipset cooling has been modified by me to fit a proper Noctua fan, there is nothing to RMA here...

 

Board did work fine otherwise though, except typical ASRock things:

- missing BIOS options

- BIOS options not working

- other massive design flaws, like

- wifi antenna connectors too close to fit custom antennas

- extremely bad chipset cooling

- putting the front USB-C connector in line with primary PCIe slot, so your only option were slow mid tier GPUs to use this connector at time of release (thanks Nvidia for 30series reference PCBs)...

 

I hope AMD will replace another chip though as I got contacted by an engineer the last time and he kind of excluded the board as the 3950X still runs fine. Will try to replace the board anyway, but definitely no ASRock one anymore.

 

Would be interesting to know, if anyone had fried a 5950X as well or has these random reboots, as Google search is full of 5950X reboot issues.

IMO Ryzen 7 5800x also had this issue. Is just that amd mass produce chip and ryzen 5000 series have couple issue about some chip are not create equally as others. Just matter of luck, you get a great chip or bad chip. 

[jwwagner25]

i would try a b550 board at this point. the likelihood that they have sent you three different CPU's that all happen to be faulty seems very improbable. I would rather count on winning the lottery. 

[Guest]

Quote

Memory: Corsair 4x8Gb SR 3600MHz CL16 @1.35V and G.Skill 2x16Gb DR 3600MHz CL14 @1.45V

 

Oh. This would create problems, running mix matched all sticks and XMP together.

 

I'd gather the IMC just hates this configuration. Can it break the IMC? 

Yes. It only supports up to 3200mt/s to start with and its never been common place to mix memory.

 

 

[ZeroLine]

33 minutes ago, ShrimpBrime said:

 

Oh. This would create problems, running mix matched all sticks and XMP together.

 

I'd gather the IMC just hates this configuration. Can it break the IMC? 

Yes. It only supports up to 3200mt/s to start with and its never been common place to mix memory.

 

 

I would rule out memory issues as he ran 2133mhz on the third RMA.

[Guest]

51 minutes ago, ZeroLine said:

I would rule out memory issues as he ran 2133mhz on the third RMA.

That means nothing to me.

I'd pull the corsair memory and run only the G.Skill B-Die by itself. Probably have less or 0 issues. 

[Krilega]

3 hours ago, ShrimpBrime said:

 

Oh. This would create problems, running mix matched all sticks and XMP together.

 

I'd gather the IMC just hates this configuration. Can it break the IMC? 

Yes. It only supports up to 3200mt/s to start with and its never been common place to mix memory.

 

 

I just tested two different kits here and did not mix them. Sorry for the confusion here.

The Corsair kit was my first DDR4 memory kit since I swapped from AMD Phenom to Intel Haswell and they are still working fine at the rated speeds. They are Samsung B-die as well, just relatively old and rated for only 2133 MHz SPD. It is a 4-Stick kit, so I went with the ASRock board initially because it has a T-Topology DRAM layout.

I swapped to the G.Skill kit just because they were actually listed as a QVL kit by ASRock for this board, are Samsung B-die and rated for 2666 MHz SPD.

 

Both kits are working fine with XMP and 1:1 FCLK ratio with my 3950X as well.

 

I ordered a MSI MAG X570 Tomahawk WIFI board today as a replacement. It seems to have a superior designed layout and every feature/connectivity I need. It is only missing some convenience features, but hey: Nobody is perfect ^^

B550 unfortunately is a no go for me as the connectivity speeds may be impacted by the PCIe 3.0 chipset interface.

 

[Guest]

29 minutes ago, Krilega said:

I just tested two different kits here and did not mix them. Sorry for the confusion here.

The Corsair kit was my first DDR4 memory kit since I swapped from AMD Phenom to Intel Haswell and they are still working fine at the rated speeds. They are Samsung B-die as well, just relatively old and rated for only 2133 MHz SPD. It is a 4-Stick kit, so I went with the ASRock board initially because it has a T-Topology DRAM layout.

I swapped to the G.Skill kit just because they were actually listed as a QVL kit by ASRock for this board, are Samsung B-die and rated for 2666 MHz SPD.

 

Both kits are working fine with XMP and 1:1 FCLK ratio with my 3950X as well.

 

I ordered a MSI MAG X570 Tomahawk WIFI board today as a replacement. It seems to have a superior designed layout and every feature/connectivity I need. It is only missing some convenience features, but hey: Nobody is perfect ^^

B550 unfortunately is a no go for me as the connectivity speeds may be impacted by the PCIe 3.0 chipset interface.

 

Thanks for that. Wasnt sure. 

 

Fingers crossed on the new board. I'll keep an eye on this thread and hope for a good resolve.

[iCoNrant]

Interesting will wait for your updates,

[Krilega]

Here is my update to the topic:

As I mentioned, I replaced the motherboard with the MSI MAG X570 Tomahawk WiFi and attempted a while of testing the new combo for occurring issues.

 

The supposedly dead 5950X ran immediately without any shutdowns, reboots or BSODs on the new board. AMD is understandingly not willing to replace the current chip, ofc.

What I noticed was, that the MSI board basically stops at 5.0 GHz single threaded boost clock and doesn't go any MHz further out of the box, while on the ASRock X570 Phantom Gaming X board all the chips did clock ~50-100 MHz higher on basically every workflow (5.1 GHz max. on Cinebench single thread), incl. the overall 0.05 V higher average VCore and 0.1V higher SoC/VDDP/VDDG. It seems that the ASRock board did ran the chips on their max. clock and voltage capabilities or did apply some sort of overclock, until the silicon degraded to the point of instability.

 

But overall I would say, the MSI board is superior to the ASRock board and possibly the ASRock X570 Taichi (same board design) as well, despite being quite a bit cheaper in "missing" functionalities like power/reset buttons, debug lcd and a second intel ethernet controller:

- VRM runs ~10°C cooler despite having a smaller heat sink

- more and proper located on board connectivity

- included WiFi antennas have superior signal strength and can be easily changed with aftermarket antennas

- heat sink fan has 0-rpm mode out of the box (you need a BIOS update for the ASRock board)

- chipset runs ~10-20°C cooler because of the simplified, but way more efficient chipset cooler (ASRock cooler requires ~2.5mm thermal pad to make slight contact with the chip and the cooler has a massive air flow restrictor on top)

 

The only "issues" I have with the MSI board is missing basic RGB controls in the BIOS and the HP Reverb G2 not working properly on any of the rear USB ports, but on one of the front 2 USB 3.0 headers.

[Mister Woof]

ASRock has been a crapshoot these days. At least on Intel side, their motherboards have been abject failures. 

I don't know what's up with them.