Overclock.net banner

141 - 160 of 180 Posts

·
Registered
Joined
·
148 Posts
Discussion Starter #141
Interestingly I've ran Prime95 this morning just to check my cpu temps again and was shocked to see that it quickly climbed up to 86C with my cpu clocked at 4.2ghz using 1.1875vcore. Now this is not what I usually saw when testing with this cpu speed. Found a new bios for my MSI MEG X570 Ace and flashed it but thought about the temps and decided to just check the Corsair IQUE software to check the pump speed. Interestingly it was reset to quiet and the pump was barely running 1000 rpm. I set it to extreme and ran Prime95 again for 31 minutes or so and was happy to see temps was just reaching 70c on one CCD. I'm pretty happy with the setup as is.
Do you know the batch number of your 3950X?
 

·
Registered
Joined
·
1,539 Posts
Does the retailer’s return policy still allow you time to exchange for an Intel CPU?
 

·
Registered
Joined
·
148 Posts
Discussion Starter #143
Does the retailer’s return policy still allow you time to exchange for an Intel CPU?
In my case, for the first 3950X I'm eligible for a refund/replacement of my choice but only after it goes through the whole RMA process. As for the second one (just arrived), I have 15 days to request a refund/replacement no questions asked. After that period the same rules for the first one apply.
But yeah, I could request an Intel CPU replacement either way, but I'm not sure I would do it though. Both of these vendors are out of 10th gen i9 CPUs, and I'd have to get a new motherboard as well...
 

·
Registered
Joined
·
3 Posts
Try enabling XMP at low ram speed, like 3000Mhz or less (so we know the RAM is fully stable) + without any other app open. Run it without any monitoring app.

If this still fails in 10 seconds. Try to run it with +50mV CPU offset. Just to be sure is not CPU requiring more voltage..

My first 3700X was failing at stock voltage in P95. Was from the first batches. Was working fine with +50mv voltage offset. So this could had meant the main issue was on the CPU

My second 3700X works much better. But sometimes I opened P95 (all stock UEFI Settings) with with other apps opened and failed as well. However the newer is "stable" i run P95 max power with: 3733Mhz ram OC + PBO with scalar 1X and 3800X TDP settings and -50mv undervolt and P95 is stable. So I can run it at stock and UV a little bit.

Anyway...my opinion is that AMD PB voltage curve is messed and not 100% stable for all the samples. all my P95 issues were always at low frequency and voltage (like 1V 1.1V)
 

·
Registered
Joined
·
148 Posts
Discussion Starter #145 (Edited)
Try enabling XMP at low ram speed, like 3000Mhz or less (so we know the RAM is fully stable) + without any other app open. Run it without any monitoring app.

If this still fails in 10 seconds. Try to run it with +50mV CPU offset. Just to be sure is not CPU requiring more voltage..

My first 3700X was failing at stock voltage in P95. Was from the first batches. Was working fine with +50mv voltage offset. So this could had meant the main issue was on the CPU

My second 3700X works much better. But sometimes I opened P95 (all stock UEFI Settings) with with other apps opened and failed as well. However the newer is "stable" i run P95 max power with: 3733Mhz ram OC + PBO with scalar 1X and 3800X TDP settings and -50mv undervolt and P95 is stable. So I can run it at stock and UV a little bit.

Anyway...my opinion is that AMD PB voltage curve is messed and not 100% stable for all the samples. all my P95 issues were always at low frequency and voltage (like 1V 1.1V)
It failed even without XMP, that was one of the first things I checked. And +50mV voltage offset worked as well, until a couple of days ago when everything started crashing randomly (I used Low LLC for ~5 months as p95 didn't crash with it).


Anyhow, I replaced the CPU and ran some benchmarks and compiles. So far so good, it looks like the previous one was definitely defective. Prime95 didn't crash on start on stock settings or with XMP enabled.
Clocks with p95 small ffts are around ~3750MHz which looks good. I can't remember the clocks previous one had on stock, but with PBO they were around ~3850MHz. Previous CPU batch number was 1946PGT btw.

EDIT: Just remembered that I uploaded a screenshot of Ryzen Master in the original post. Previous CPU was clocked at ~3580MHz when runing p95 small ffts. Not sure what to think about that, I'm confused why the new one would clock almost 200MHz higher by default (not that I'm complaining though).
 

·
OG AMD
Joined
·
8,803 Posts
What does it have to do with the PSU power rating?
Unless the PSU is complete garbage, 650W is more than sufficient for the described setup, unless something moronic is done with the RTX.

Come on dude. You know at stock settings it will. Add 4 HDD's, 6 RGB Fans and full RAM plus other stuff and you could easily draw more than 650w from the wall. I would bet the 3950x with 4 sticks of RAM and the 2080ti can peek at 600w alone? 2080Ti is 758mm die, can really suck up the juice.
 

·
Turtle Lives Matter
Joined
·
4,787 Posts
Agreed. There's plenty of PSUs with 50mV of ripple that would break a lot of clocks. ATX spec allows for a huge amount of ripple on the 12V rail, something like 120mV going like ~~~~~~ going to trash some overclocks.

Keep in mind that those capacitors used to filter ripple degrade over time as the dielectric evaporates, and they do so much faster under higher temperatures.
 

·
Registered
Joined
·
148 Posts
Discussion Starter #148
While that may be the case in some other workloads I fail to see what it's got to do with running prime95 small ffts that mainly targets the CPU and cache. So, yeah, you should have a good enough PSU to power both 3950X and RTX 2080Ti when both are actually used and under load, which simply isn't true for prime95.
 

·
Registered
Joined
·
951 Posts
Do you know the batch number of your 3950X?
Hey @webstar I cannot remember the exact and I don't really want to take off my cooling right now, but what I can say is that it was from somewhere week 40 to 45 or so in 2019. I stand to be corrected on it, but it was one of the first available here in South Africa.
 

·
Registered
Joined
·
148 Posts
Discussion Starter #150
Hey @webstar I cannot remember the exact and I don't really want to take off my cooling right now, but what I can say is that it was from somewhere week 40 to 45 or so in 2019. I stand to be corrected on it, but it was one of the first available here in South Africa.
No worries. That seems close to my first 3950X. Do you remember the stock clocks when running prime95 small ffts?
 

·
Registered
Joined
·
951 Posts
No worries. That seems close to my first 3950X. Do you remember the stock clocks when running prime95 small ffts?
Hi, give me a couple of minutes then I'll check at stock no PBO setting to see what the clocks are while running Prime95 small fft.
 

·
Registered
Joined
·
951 Posts
No worries. That seems close to my first 3950X. Do you remember the stock clocks when running prime95 small ffts?
Okay I've quickly done a test with cpu at stock only adjustment is a -0.05Vcore and all seem good. Cpu speed as per the screen grab below was between 3850 and 3875 in the beginning I did also see a 3900 but that was only briefly. Temps were hovering in the low 50c range.
 

Attachments

·
Registered
Joined
·
148 Posts
Discussion Starter #153
Okay I've quickly done a test with cpu at stock only adjustment is a -0.05Vcore and all seem good. Cpu speed as per the screen grab below was between 3850 and 3875 in the beginning I did also see a 3900 but that was only briefly. Temps were hovering in the low 50c range.
Thanks dude. Your 3950X seems like a really good chip.
 

·
Registered
Joined
·
36 Posts
Very sad to hear that. maybe you need return money for aorus extreme and buy motherboard from different brand or tests your cpu on PC friend.
By the way, I tested prime95 small fft's 2.5h without errors and custom size 4-36k 2 hours, stable
Unfortunately, I ran into problems with bsods at idle time and only in idle time. All tests pass under load (prime95 small and occt small data with avx2). Solution that helps to disable it Global C-State Control. It looks like the processor is still defective. When the new 4000 series processors are released I will change under warranty. Bios latest 1.16 with AGESA 1.0.8.0.

idle voltages (hwinfo)
vcore
all auto (windows power plan high perfomance, freq 3600) - min 0.350v - bsod
all auto (windows power plan high perfomance, freq 3600), Global C-state Control Disabled - min 1.024-1.040v - normal
all auto (windows power plan high perfomance, freq 3600), Power Supply Idle Control Low - min 0.350v - bsod
all auto (windows power plan high perfomance, freq 3600), Power Supply Idle Control Typical - min 0.976v - bsod
 

·
Registered
Joined
·
951 Posts
Thanks dude. Your 3950X seems like a really good chip.
It is fine, just a little hot with manual overclock when going close to 1.3vcore which is why I run it 4.2ghz at 1.1875vcore.
 

·
OG AMD
Joined
·
8,803 Posts
Unfortunately, I ran into problems with bsods at idle time and only in idle time. All tests pass under load (prime95 small and occt small data with avx2). Solution that helps to disable it Global C-State Control. It looks like the processor is still defective. When the new 4000 series processors are released I will change under warranty. Bios latest 1.16 with AGESA 1.0.8.0.

idle voltages (hwinfo)
vcore
all auto (windows power plan high perfomance, freq 3600) - min 0.350v - bsod
all auto (windows power plan high perfomance, freq 3600), Global C-state Control Disabled - min 1.024-1.040v - normal
all auto (windows power plan high perfomance, freq 3600), Power Supply Idle Control Low - min 0.350v - bsod
all auto (windows power plan high perfomance, freq 3600), Power Supply Idle Control Typical - min 0.976v - bsod
Well that sure seems like a bios/ageasa bug. Why return cpu?
 

·
Registered
Joined
·
43 Posts
Did anyone encounter similar issue with 3900X? Cause it seems that mine 3900X is affected as well.

My CPU is pretty early batch (I got it last year in August IIRC). On all default settings it fails within few minutes of starting prime95 with FFT length 16K with AVX2 enabled (16K is the one where it fails the most, but I also saw a fail while going through 20k). So far only cure I found is to use offset of +50mV, but I am still investigating this issue. I didn't play with LLC yet, for now I use Auto, which according to BIOS = LLC3. I tried 3 latest BIOSes for my mobo (3.00, 3.20, 3.40), issue occurs on all of them. Seems that disabling AVX2 makes prime95 pass. Also custom run of prime95 with FFTs between 2 and 15K passed over 11 hours.

I am not sure if this is some kind of CPU degradation, cause last time I tried small FFTs with prime95 was just after I got this rig (last year) and back then Taichi BIOSes were completely broken when it comes to power usage reporting and CPU was going wild in prime95, CPU was thermal throttling at 95 C and it is not really a fair comparison to current state of MOBO/BIOS. Only recently I run smallFFT to check something else and I found out that 16K fails.

I am trying to figure out if I should bother with RMA, as I am mildly bothered with the fact that my CPU does not work correctly at default....

Rest of the rig:
- X570 AsRock Taichi;
- GSkill F4-3200C14D-32GTZKW (2x16GB B-Die);
- Corsair AX860 (unit from end of 2017, so quite new);
- Palit GTX 1080;
 

·
Premium Member
Joined
·
20,111 Posts
Did anyone encounter similar issue with 3900X? Cause it seems that mine 3900X is affected as well.

My CPU is pretty early batch (I got it last year in August IIRC). On all default settings it fails within few minutes of starting prime95 with FFT length 16K with AVX2 enabled (16K is the one where it fails the most, but I also saw a fail while going through 20k). So far only cure I found is to use offset of +50mV, but I am still investigating this issue. I didn't play with LLC yet, for now I use Auto, which according to BIOS = LLC3. I tried 3 latest BIOSes for my mobo (3.00, 3.20, 3.40), issue occurs on all of them. Seems that disabling AVX2 makes prime95 pass. Also custom run of prime95 with FFTs between 2 and 15K passed over 11 hours.

I am not sure if this is some kind of CPU degradation, cause last time I tried small FFTs with prime95 was just after I got this rig (last year) and back then Taichi BIOSes were completely broken when it comes to power usage reporting and CPU was going wild in prime95, CPU was thermal throttling at 95 C and it is not really a fair comparison to current state of MOBO/BIOS. Only recently I run smallFFT to check something else and I found out that 16K fails.

I am trying to figure out if I should bother with RMA, as I am mildly bothered with the fact that my CPU does not work correctly at default....

Rest of the rig:
- X570 AsRock Taichi;
- GSkill F4-3200C14D-32GTZKW (2x16GB B-Die);
- Corsair AX860 (unit from end of 2017, so quite new);
- Palit GTX 1080;
The newer versions of Prime include AVX. If you don't use any apps that utilize AVX, disabling it makes Prime a better tool for stability - IMO.

EDIT: If Prime test for AVX 512, which Ryzen does not support - will that cause for a failed test?
I know for a fact that some intel motherboards have AVX settings in BIOS and may need to be adjusted to pass tests like Prime95.
 

·
Registered
Joined
·
36 Posts
Well that sure seems like a bios/ageasa bug. Why return cpu?
I'm not sure if this is a BIOS bug because I tried all available versions 1.00, 1.10, 1.10A, 1.15A, 1.16, 1.30. Found something interesting, if I turn on the PBO and increase the limits by 15% (PPT: 163, TDC: 109, EDC: 161), then I can't get pass prime95 small fft (cpu not overheat), even with global c-state disable. I think the processor is defective and I will change it again.
 
141 - 160 of 180 Posts
Top