Overclock.net banner

161 - 180 of 180 Posts

·
Premium Member
Joined
·
20,111 Posts
Indeed it does. And, yes, it does not support AVX 512 or AVX3. Some intel motherboards have the ability to adjust these things in order to pass new version of Prime to lower thermals.

Realbench tests for these things, too, but only the makers of Prime know why their tests are tougher. I bet regular apps that use AVX won't exact same load. I am not saying the programmers of Prime screwed something up but as a result of the addition of AVX tests, adjustments may be necessary to be made in the BIOS.
 

Attachments

·
Registered
Joined
·
43 Posts
I am not saying the programmers of Prime screwed something up but as a result of the addition of AVX tests, adjustments may be necessary to be made in the BIOS.
AMD does not have AVX offset setting in BIOS like Intel.

Anyway, I did contacted ASRock and AMD support regarding this issue, got pretty much nothing. I did some more tests and the funny thing is that prime95 fails only on 16K and 20K long FFTs. It can pass hours of 21-64 or 1-15 or even 18 (only FFT between 16 and 20), yet it fails within few minutes of starting 16 or 20K. 16K and 20K also somehow are run with lower CPU clock than everything else I tried (~3650, when 18K runs 3750). All the fails are only observed on threads 17,18, 21 and 22, so pretty much 2 cores only (AFAIK prime always binds same worker threads to same cores). Temperature is not a problem - it stays around 75 C.

I can work around this issue with PBO, increasing TDC by few amps (still testing for exact amount) seems to solve the issue. Still it seems that my CPU is bit dodgy, I am still considering RMAing it. To me it looks like those CPUs are binned really tight, sometimes (like in my case) even too tight....
 

·
Registered
Joined
·
951 Posts
AMD does not have AVX offset setting in BIOS like Intel.

Anyway, I did contacted ASRock and AMD support regarding this issue, got pretty much nothing. I did some more tests and the funny thing is that prime95 fails only on 16K and 20K long FFTs. It can pass hours of 21-64 or 1-15 or even 18 (only FFT between 16 and 20), yet it fails within few minutes of starting 16 or 20K. 16K and 20K also somehow are run with lower CPU clock than everything else I tried (~3650, when 18K runs 3750). All the fails are only observed on threads 17,18, 21 and 22, so pretty much 2 cores only (AFAIK prime always binds same worker threads to same cores). Temperature is not a problem - it stays around 75 C.

I can work around this issue with PBO, increasing TDC by few amps (still testing for exact amount) seems to solve the issue. Still it seems that my CPU is bit dodgy, I am still considering RMAing it. To me it looks like those CPUs are binned really tight, sometimes (like in my case) even too tight....
If I may ask, you mention here 16 and 20K and the only ones I saw while running Prime95 small FFT for about 30 minutes or so is 8K, 10K, 12K, 16K 20k, 24K and 28K, but you mentioned some higher and some lower as well. Look I did not let it run longer than 33 minutes, actually did not see the point in doing it. This was with a manual overclock of 4.2ghz using 1.1875vcore set in bios, dropping to 1.1625vcore or there about. Below is while running and when done. I think it is pretty stable for me and temps not too bad.
 

Attachments

·
Registered
Joined
·
43 Posts
If I may ask, you mention here 16 and 20K and the only ones I saw while running Prime95 small FFT for about 30 minutes or so is 8K, 10K, 12K, 16K 20k, 24K and 28K, but you mentioned some higher and some lower as well.
I run custom mode instead of small FFT. You can specify ranges of FFTs to run there :D, I checked all available FFTs up to 64K this way.
 

·
Registered
Joined
·
951 Posts
I run custom mode instead of small FFT. You can specify ranges of FFTs to run there :D, I checked all available FFTs up to 64K this way.
Thank you kindly my good sir. Just wanted to understand. At present I am happy with my setup as is. I can push the processor higher, maybe even 4.3Ghz but at 1.25vcore I am somewhat worried about the temps it would emit.
 

·
Premium Member
Joined
·
20,111 Posts
AMD does not have AVX offset setting in BIOS like Intel.

Anyway, I did contacted ASRock and AMD support regarding this issue, got pretty much nothing. I did some more tests and the funny thing is that prime95 fails only on 16K and 20K long FFTs. It can pass hours of 21-64 or 1-15 or even 18 (only FFT between 16 and 20), yet it fails within few minutes of starting 16 or 20K. 16K and 20K also somehow are run with lower CPU clock than everything else I tried (~3650, when 18K runs 3750). All the fails are only observed on threads 17,18, 21 and 22, so pretty much 2 cores only (AFAIK prime always binds same worker threads to same cores). Temperature is not a problem - it stays around 75 C.

I can work around this issue with PBO, increasing TDC by few amps (still testing for exact amount) seems to solve the issue. Still it seems that my CPU is bit dodgy, I am still considering RMAing it. To me it looks like those CPUs are binned really tight, sometimes (like in my case) even too tight....
Im aware there isn't any. Got 4 ryzen systems and none of them have any. Prime can be edited, though, unless of course it isn't the AVX test that is causing the issue. Wonder if using an old version of Prime without AVX will still cause any issue with these chips.

I read in the past The Stilt recommended a minimum of 1.28v prior to using the newer version of Prime on Ryzen systems. Can't find the info anymore or i totally misquote him or something.

Observe your EDC and other readings when running Prime if they are spiking too high.
 

·
Registered
Joined
·
43 Posts
Im aware there isn't any. Got 4 ryzen systems and none of them have any. Prime can be edited, though, unless of course it isn't the AVX test that is causing the issue. Wonder if using an old version of Prime without AVX will still cause any issue with these chips.
With AVX2 disabled prime passes without any problem, but it does not change anything. I want my CPU to pass any stress test at DEFAULT settings, even if test is using a specific supported instruction set.
Observe your EDC and other readings when running Prime if they are spiking too high.
There is nothing spiking. During prime95 16k my TDC is capped at 100% (95A) and that is it.
 

·
Premium Member
Joined
·
20,111 Posts
With AVX2 disabled prime passes without any problem, but it does not change anything. I want my CPU to pass any stress test at DEFAULT settings, even if test is using a specific supported instruction set.

There is nothing spiking. During prime95 16k my TDC is capped at 100% (95A) and that is it.
But, what if the programmers did screw up and here owners of this chip are banging their heads to pass the test?
 

·
Premium Member
Joined
·
2,711 Posts
16K FFT size was mentioned, any other sizes which seem to be more prone to produce errors than the others?
 

·
Registered
Joined
·
43 Posts
But, what if the programmers did screw up and here owners of this chip are banging their heads to pass the test?
That is also possible of course, but I do no think this is the case here. First of all - in such case upping voltage should not help with the stability, yet it does help most people, myself included. Secondly last stable prime95 (29.8 build 6) was released over a year ago, I think such bug would be found and fixed by now.

16K FFT size was mentioned, any other sizes which seem to be more prone to produce errors than the others?
I only saw errors on 16K and 20K on my CPU, both fail within few minutes. I can run custom 16K-16K, in-place and it usually fails within first loop (6 mins), same for 20K. Fails always happen on one of the 4 workers -#17,#18,#21,#22 (cores 9 and 11, right?).I tested all FFT sizes up to 64K and never fail with sizes other than 16K and 20K.

I realized I never posted screenshot of my settings here, so here it is, everything default, prime95 fails after 2 minutes. Maybe you guys can spot something I am setting incorrectly.
 

Attachments

·
Registered
Joined
·
885 Posts
That is also possible of course, but I do no think this is the case here. First of all - in such case upping voltage should not help with the stability, yet it does help most people, myself included. Secondly last stable prime95 (29.8 build 6) was released over a year ago, I think such bug would be found and fixed by now.


I only saw errors on 16K and 20K on my CPU, both fail within few minutes. I can run custom 16K-16K, in-place and it usually fails within first loop (6 mins), same for 20K. Fails always happen on one of the 4 workers -#17,#18,#21,#22 (cores 9 and 11, right?).I tested all FFT sizes up to 64K and never fail with sizes other than 16K and 20K.

I realized I never posted screenshot of my settings here, so here it is, everything default, prime95 fails after 2 minutes. Maybe you guys can spot something I am setting incorrectly.
Guru3d have a 30.3 build 4 version!

Have absolutely no idea where they got this from.

OK found it

https://www.mersenneforum.org/showthread.php?t=25823
 

·
OG AMD
Joined
·
8,803 Posts
Thanks for the link. I tried this version but results were the same. I think I will try to RMA the CPU, already talked to the retailer about this.
How do you get CL15-15-15 without overclocking? How do you get a 59c motherboard temp? In your case 700rpm case fans.....on a 3950X. LOL

Change those timings to CL16-16-16 and the rest auto an turn some fans on in your case and the errors go away. :)
 

·
Registered
Joined
·
951 Posts
Guru3d have a 30.3 build 4 version!

Have absolutely no idea where they got this from.

OK found it

Prime95 v30.3 - mersenneforum.org
Interestingly, with the previous Prime 95 running small FFT I am stable with running my 3950X at 4.2ghz using 1.1875vcore and temps just about reaching 70C. So I tried this 30.3 and interestingly my temps within seconds jumped up to 80c and soon after my system restart. Just to make sure my setup can run it I decided to go back to stock (no PBO) and see where the speed it is at while running and low and behold it dropped from 3.875ghz to 3.4 to 3.5ghz with temps not even entering 60C while running.
 

·
Registered
Joined
·
43 Posts
How do you get CL15-15-15 without overclocking?
By loading BIOS defaults. RAM is clocked at 1066 MHz. I am pretty sure it can even run 11-11-11 at those speeds.

How do you get a 59c motherboard temp?
That is a screwed up readout by HWInfo, it is some other sensor near chipset. ASRock utility/BIOS shows different MB temp.

In your case 700rpm case fans.....on a 3950X. LOL
It is 3900X and one case fan is even at 450 rpm, you missed that one, CPU is under 70 C, its fine.

Change those timings to CL16-16-16 and the rest auto an turn some fans on in your case and the errors go away. :)
Surprisingly, loading XMP and settings fans to max speed did not solve anything (yes I actually tested).

Interestingly, with the previous Prime 95 running small FFT I am stable with running my 3950X at 4.2ghz using 1.1875vcore and temps just about reaching 70C. So I tried this 30.3 and interestingly my temps within seconds jumped up to 80c and soon after my system restart. Just to make sure my setup can run it I decided to go back to stock (no PBO) and see where the speed it is at while running and low and behold it dropped from 3.875ghz to 3.4 to 3.5ghz with temps not even entering 60C while running.
I checked again your prime95 screens you put here mate. Your prime worker windows said: "using Pentium-4 type-1 FFT" (should be "using FMA3 FFT"). I assume you disabled FMA3 by putting "CpuSupportsFMA3=0" in your prime.txt in your old install of prime. Best to remove prime.txt in your old folder (will reset settings to defaults) and try running it again. Seems like your OC was not fully stable in the end :-(.
 

·
Registered
Joined
·
951 Posts
I checked again your prime95 screens you put here mate. Your prime worker windows said: "using Pentium-4 type-1 FFT" (should be "using FMA3 FFT"). I assume you disabled FMA3 by putting "CpuSupportsFMA3=0" in your prime.txt in your old install of prime. Best to remove prime.txt in your old folder (will reset settings to defaults) and try running it again. Seems like your OC was not fully stable in the end :-(.
Thank you kindly. If you can explain to me where you see "using Pentium-4 type-1 FFT" I'd really appreciate it. I did delete the prime.txt file and retested it. Temps went higher as in 75c or so in about 3 to 4 minutes then stabilised for a bit but I am sure it would have gone higher. I checked inside the prime.txt file for both version and no "CpuSupportsFMA3=" or anything of that sort. Numbers in the newer one was a little higher and a bit more lines in the file.

I know that my processor is perfectly stable running 4.2ghz at 1.1875vcore as I have never had a blue screen or restart or any issue related to instability.
 

·
Registered
Joined
·
951 Posts
I checked again your prime95 screens you put here mate. Your prime worker windows said: "using Pentium-4 type-1 FFT" (should be "using FMA3 FFT"). I assume you disabled FMA3 by putting "CpuSupportsFMA3=0" in your prime.txt in your old install of prime. Best to remove prime.txt in your old folder (will reset settings to defaults) and try running it again. Seems like your OC was not fully stable in the end :-(.
My appology, I saw now the Pentium 4 thingy
 

·
Registered
Joined
·
951 Posts
I checked again your prime95 screens you put here mate. Your prime worker windows said: "using Pentium-4 type-1 FFT" (should be "using FMA3 FFT"). I assume you disabled FMA3 by putting "CpuSupportsFMA3=0" in your prime.txt in your old install of prime. Best to remove prime.txt in your old folder (will reset settings to defaults) and try running it again. Seems like your OC was not fully stable in the end :-(.
Okay, so I've tried the new Prime95 just now with my processor back to stock with PBO set and -0.1000 vcore offset. The vcore while running states 1.044 lowest and 1.069 highest with processor speed at between 3.925 and 3.95ghz, temps climbed up to 77.8c in about 5 minutes or so. But the test reads "using FMA3 FFT length........." so I'd say it is fine. Test is running okay. I will try it fully stock without PBO to see the temps, last time was just about 50c.
 

·
OG AMD
Joined
·
8,803 Posts
By loading BIOS defaults. RAM is clocked at 1066 MHz. I am pretty sure it can even run 11-11-11 at those speeds.


That is a screwed up readout by HWInfo, it is some other sensor near chipset. ASRock utility/BIOS shows different MB temp.


It is 3900X and one case fan is even at 450 rpm, you missed that one, CPU is under 70 C, its fine.


Surprisingly, loading XMP and settings fans to max speed did not solve anything (yes I actually tested).


I checked again your prime95 screens you put here mate. Your prime worker windows said: "using Pentium-4 type-1 FFT" (should be "using FMA3 FFT"). I assume you disabled FMA3 by putting "CpuSupportsFMA3=0" in your prime.txt in your old install of prime. Best to remove prime.txt in your old folder (will reset settings to defaults) and try running it again. Seems like your OC was not fully stable in the end :-(.

That doesn't sound right with geardown being disabled by auto
 
161 - 180 of 180 Posts
Top