Interesting find, 5900x not stable without curve optimizer, is my CPU faulty?

[ffletchs]

Just found out something interesting, wondering if the CPU is faulty or if its a bios issue....

I've been having a hard time getting my CPU/memory/bios/motherbord to be stable since I got 5900x

relevant specs are
AMD Ryzen | Nvidia GTX 3080 | Asus ROG Crosshair VIII Hero | 2x16GB DDR4 Samsung B-Die DDR 3600 CL16

Basically I've been having issue where computer would be stable for awhile, then a random boot durning idle, and then no matter what settings in bios (not even ddr 2133/2400 with timings on auto) would the machine would not boot again to windows. Had to clear cmos or flash again to get machine back up. This was at any memory settings, all auto, DOCP only or any tweaking.

Also had to change a lot of bios settings to get close to stability, never got it 100% stable, I was setting LLC to 3/4, idle current to typical, VSOC 1.15V at least, IOD VDDG around 1 to 1.05 and CCD around 0.98 to 1.05, still would crash at idle occasionally and then back to clear CMOS/reflash


With bios 3202 I decided to try undervolting with curve optimizer, enabled all cores, negative value -20, no boot to windows, -10 no boot to windows, basically no matter what I tried, no boot to windows. Just for fun I tried per core cause I've read some cores undervolt more than others.

To make a long story short I found out that 2 cores, cores (3&4 counting from zero) are unstable at any negative value, all other cores run fine at -20 (haven't tried higher). Infact I had to set those 2 cores to positive +5 to get the machine stable.

And with these Curve Optimizer settings, all cores to -20 but 3&4 +5, machine is completly stable!? I don't need to set any volts for SOC, CCD, VDDG or LLC, all is on auto now.

And I'm getting some great memory overclocks, currently at DDR3800 @ 14-14-14-28 and tweaked all subtimings. AIDA64 giving me 59.057MB read and 57ns Latency

Disabling Curve optimizer and I'm back to previous troubles

Thoughts? bios issue or is my CPU faulty?

[xeizo]

ffletchs wrote:
Just found out something interesting, wondering if the CPU is faulty or if its a bios issue....

I've been having a hard time getting my CPU/memory/bios/motherbord to be stable since I got 5900x

relevant specs are
AMD Ryzen | Nvidia GTX 3080 | Asus ROG Crosshair VIII Hero | 2x16GB DDR4 Samsung B-Die DDR 3600 CL16

Basically I've been having issue where computer would be stable for awhile, then a random boot durning idle, and then no matter what settings in bios (not even ddr 2133/2400 with timings on auto) would the machine would not boot again to windows. Had to clear cmos or flash again to get machine back up. This was at any memory settings, all auto, DOCP only or any tweaking.

Also had to change a lot of bios settings to get close to stability, never got it 100% stable, I was setting LLC to 3/4, idle current to typical, VSOC 1.15V at least, IOD VDDG around 1 to 1.05 and CCD around 0.98 to 1.05, still would crash at idle occasionally and then back to clear CMOS/reflash


With bios 3202 I decided to try undervolting with curve optimizer, enabled all cores, negative value -20, no boot to windows, -10 no boot to windows, basically no matter what I tried, no boot to windows. Just for fun I tried per core cause I've read some cores undervolt more than others.

To make a long story short I found out that 2 cores, cores (3&4 counting from zero) are unstable at any negative value, all other cores run fine at -20 (haven't tried higher). Infact I had to set those 2 cores to positive +5 to get the machine stable.

And with these Curve Optimizer settings, all cores to -20 but 3&4 +5, machine is completly stable!? I don't need to set any volts for SOC, CCD, VDDG or LLC, all is on auto now.

And I'm getting some great memory overclocks, currently at DDR3800 @ 14-14-14-28 and tweaked all subtimings. AIDA64 giving me 59.057MB read and 57ns Latency

Disabling Curve optimizer and I'm back to previous troubles

Thoughts? bios issue or is my CPU faulty?

Glad you could fix it, but yes, basically if two cores needs more volt than default to even boot - your CPU is faulty. Bad QC from AMD. It may overclock bad, that would be ok, but not running default is not ok. With that said, running like you do is exactly what CO is made for to make possible.

[xeizo]

On further comment, this was a great question asked, it inspired me to find my "bad" core/cores. Turns out Core 1 on my 5900X is a lot worse than all the other cores, it easily spits out WHEA errors if not fed enough juice. Curve Optimizer is the savior in this situation, I can isolate Core 1 with a conservative setting and tweak the other cores to my hearts liking without any instability 🙂

[ffletchs]

xeizo wrote:
On further comment, this was a great question asked, it inspired me to find my "bad" core/cores. Turns out Core 1 on my 5900X is a lot worse than all the other cores, it easily spits out WHEA errors if not fed enough juice. Curve Optimizer is the savior in this situation, I can isolate Core 1 with a conservative setting and tweak the other cores to my hearts liking without any instability 🙂

Cool, yeah its easy to spot the "weak/bad" cores by undervolting them, my computer is finally completely stable and thats with very aggresive memory timings :cool:

[BillBittel]

This is really interesting. I have read MANY posts about Ryzen 5000 series CPUs, especially 5900X and 5950X are throwing WHEA errors and BSODs while under light load. People are sending them back because of this. In OPs case, I think even though he can work around it, he should consider RMAing it because as xeizo said, if it won't boot and run stable at defaults, that is not OK. The down side is having to go through the RMS process with AMD and waiting for a replacement. This sounds like it may be what a lot of the people having these problems should do. They can decide whether an RMS is worth the hassle based on how bad their weak cores are. Great info!

[MikeS3000]

So same discovery a few weeks ago on my "faulty" 5900x. My CPU fails prime95 and occt single thread tests on my best gold star #1 core. I have to use +5 curve optimizer on that core just to pass stress tests. The other cores can undervolt a little bit. I should RMA but the computer is quite stable and I only loose about 75 to 100 mhz on the best core and the rest of the cores boost nicely. Plus, the RMA process would take 2 months to complete with as little supply as there is available.