Overclock.net banner

1 - 20 of 27 Posts

·
Registered
Joined
·
18 Posts
Discussion Starter #1
So I've got the following components in a build:

- Ryzen 7 1700
- Asrock Taichi X370 (BIOS v3.20)
- Noctua NH-D15 SE-AM4
- G.SKILL Flare X Series 16GB F4-3200C14D-16GFX (Samsung B-die)
- GTX 1050 Ti
- EVGA Supernova 550 G3

I already built this exact same system a few days ago for myself and I had tremendous success with overclocking the CPU. This duplicate build is for a friend at work that's paying me to build it for him.

The problem is I can't get this duplicate to overclock at all.
mad.gif
It's really ticking me off big time. I guess to be more accurate I can get 3.8GHz overclock @ 1.3V to pass every stress test application _except_ for the OCCT Large Data Set.

For example I ran the following tests for 8hrs each in the order shown and they all pass:
- Memtest86
- Aida64
- Prime95 Small FFT
- Prime95 Large FFT
- Prime95 Blend
- OCCT: Linpack
- OCCT: Small Data Set

But the OCCT Large Data Set fails in less than 2 minutes every time I run it! What is OCCT Large Data Set doing that's so special that it keeps locking up the system? I thought Prime95 was supposed to be the big kahuna in stress testing, but OCCT Large Data Set seems to find instability when prime95 doesn't - and very quickly too!

The OCCT Large Data Set test will only pass when the system is not overclocked. I successfully ran it for 10hrs in the non-overclocked base configuration. But if I overclock the CPU _at_all_, even just 3.1GHz it crashes the system.

Help? Ideas?

How can all 3 prime95 tests pass for 8hr runs but OCCT Large Data Set fails in <2min?? Either OCCT is a crazy awesome stress test tool or there is something wrong with it... Although I don't think we can say there is something wrong with it since it runs without issue in the non-overclocked base configuration for 10hrs.
sadsmiley.gif
 

·
Registered
Joined
·
1,361 Posts
I don't know what exactly OCCT does but I've long since considered it the best stress tester for AMD CPU's back to my 8150. I could get my overclocks to pass 12+ hours of Prime95 but only 20 minutes of OCCT. I've found that any overclock that passes 4 hours of OCCT has been 24/7 stable for my applications (gaming, internet, light photo work). As crappy as it may sound, if that CPU won't pass OCCT at 3.8 then back it down to 3.75 and see what happens. Not all 1700's will hit 3.9 or even 3.8 and you might have one that won't.

I'd work on finding the highest OCCT stable overclock first then work up from there.
 

·
Registered
Joined
·
18 Posts
Discussion Starter #3
Quote:
Originally Posted by MacLeod View Post

I'd work on finding the highest OCCT stable overclock first then work up from there.
That's the problem though. As I stated, even a slight overclock to 3.1GHz causes OCCT LDS to fail when all other stress tests run just fine for many hours. So I think something weird must be going on.

Do you typically run the medium/large data sets? Or just the small and LINPACK tests?
 

·
Registered
Joined
·
1,361 Posts
I run the large sets. Linpak is basically Intel Burn Test and I could get ridiculous over clocks with my 8150 and 8350 to pass that but crash almost instantly under OCCT and P95.

That does sound weird tho. Maybe uninstall then reinstall OCCT, maybe it's buggy or something. If it's still crashing even at 3.1 but passes other tests with no problem then I'd chalk it up to an anomaly and not worry about it.
 

·
Registered
Joined
·
18 Posts
Discussion Starter #5
Quote:
Originally Posted by MacLeod View Post

If it's still crashing even at 3.1 but passes other tests with no problem then I'd chalk it up to an anomaly and not worry about it.
Yeah, that's my current plan. Just sucks though... :/

May I ask which version of OCCT you are currently using? I'm using v4.5.1. Maybe I should try a different version? I will also try uninstalling and reinstalling the current version.

I've now run all three P95 tests (blend/large/small) for 16+hrs _each_ without a single failure. I don't get why the heck OCCT LDS keeps failing. It's so bizarre. I just wish I knew more about what was going on under the hood of OCCT LDS. Maybe it would clue me in to what to look for.
 

·
Registered
Joined
·
18 Posts
Discussion Starter #7
Quote:
Originally Posted by AlphaC View Post

Check if it's failing in AVX.
Sorry, but can you explain how to do that? I'm not even sure what AVX means exactly - this is my first overclocking experience so sorry for my ignorance.
Quote:
Originally Posted by AlphaC View Post

If it maybe you aren't using the right Prime95 version
Prime95 About box says "Windows64 v29.5 build 5"
 

·
⤷ αC
Joined
·
11,239 Posts
AvX is a type of instructionset like SSE is.

Prime 29.5 should have AVX and AVX2.

Basically it heats up your CPU more and is more likely to fail due to more work (vectors) being done at the same clockspeed.

----

Also check your RAM timings and motherboard BIOs version. Some CPUs have better memory controllers than others. Perhaps you lucked out on yours but he received a bad sample.
 

·
Registered
Joined
·
18 Posts
Discussion Starter #9
Quote:
Originally Posted by AlphaC View Post

Also check your RAM timings and motherboard BIOs version. Some CPUs have better memory controllers than others. Perhaps you lucked out on yours but he received a bad sample.
Well the interesting thing is that the OCCT LDS test fails even when the RAM is not overclocked at all.
frown.gif
But at the same time I ran memtest86+ from flash drive for 24.5 hours and it had 0 errors. So I know the memory is not faulty at least.
 

·
Newb to Overclock.net
Joined
·
4,526 Posts

·
Registered
Joined
·
18 Posts
Discussion Starter #11
Quote:
Originally Posted by mouacyk View Post

I bet you have a CPU that is affected with the hardware segfault bug:
http://www.overclock.net/t/1635749/phoronix-segmentation-faults-on-zen-cpus-under-heavy-workloads

Contrary to original reports, this is not a marginalized Linux issue.
That's definitely interesting. However, I am running windows 10. From reading the thread this issue only seems to affect linux. Is that true?

Also, is there a way to tell (hopefully without removing my heatsink and thermal paste) which manufacturing week my CPU is from?
 

·
Registered
Joined
·
18 Posts
Discussion Starter #12
Quote:
Originally Posted by mouacyk View Post

I bet you have a CPU that is affected with the hardware segfault bug:
http://www.overclock.net/t/1635749/phoronix-segmentation-faults-on-zen-cpus-under-heavy-workloads

Contrary to original reports, this is not a marginalized Linux issue.
Well I researched this some more and it does seem like it is possible for this to happen on Windows. And my friend does plan on using WSL on Windows 10 so I tried running some stress tests on Ubuntu on WSL with phoronix. I ran the following command a couple times:

PTS_CONCURRENT_TEST_RUNS=4 TOTAL_LOOP_TIME=480 phoronix-test-suite stress-run build-linux-kernel build-php build-apache build-llvm build-gcc

I ran the above test while simultaneously running various combinations of Aida64, OCCT Linpack, Pime95, Cineblend, etc.. Unfortunately I could not get phoronix to cause a segfault. So I guess that's not it either. :/

This is very irritating...
 

·
Newb to Overclock.net
Joined
·
4,526 Posts
May want to look at your manufacturing date now then. Also, people who replicated the bug quickly also had opcache and SMT both enabled.
 

·
Registered
Joined
·
18 Posts
Discussion Starter #14
Quote:
Originally Posted by mouacyk View Post

May want to look at your manufacturing date now then. Also, people who replicated the bug quickly also had opcache and SMT both enabled.
I checked my BIOS and both SMT and opcache are both _enabled_. I never changed them from their defaults so they were turned on when I got the motherboard.

...Maybe a good test would be to try disabling them and see if OCCT LDS passes...?
 

·
Registered
Joined
·
18 Posts
Discussion Starter #15
Quote:
Originally Posted by MacLeod View Post

...OCCT...
Quote:
Originally Posted by AlphaC View Post

...AVX...
Quote:
Originally Posted by mouacyk View Post

...segfault...
I just noticed something incredibly disturbing/concerning. Essentially _all_ of the posts that I can find across numerous websites where people are claiming that their Ryzen overclock successfully runs OCCT Large Data Set are _all_ dated _prior_ to the release of OCCT v4.5.1 (which was on 14 August 2017). This is very significant because the OCCT v4.5.1 release note states "Added Ryzen and Skylake-X support". So it would seem that all of these people were testing with a version of OCCT that didn't stress Ryzen specific instructions (quite possibly AVX).
 

·
Registered
Joined
·
1,361 Posts
No, mine was installed just a couple weeks ago when I built this rig so mine would be the most recent version and that's the one I tested my overclock on.
 

·
Registered
Joined
·
18 Posts
Discussion Starter #17
I think I'm gonna call it quits on trying to get OCCT LDS to run. I've been running all of the other stress test tools I mentioned in my OP (including lengthy compilations with phoronix on WSL, realbench, and GPU stressors) all simultaneously for the last 72hrs and the machine hasn't crashed a single time. I also played Witcher 3 for about 7 hours straight and no problems.

Whatever is going on with OCCT LDS I don't think it's telling me anything useful. Even if there really is some instability somewhere that it just happens to be uniquely finding I'm not convinced at all that this machine will ever experience it during normal daily use.

Time to call it good and put this machine to use.

Thanks everyone for the help/suggestions. Much appreciated.
 

·
⤷ αC
Joined
·
11,239 Posts
Quote:
Originally Posted by jwdonal View Post

I just noticed something incredibly disturbing/concerning. Essentially _all_ of the posts that I can find across numerous websites where people are claiming that their Ryzen overclock successfully runs OCCT Large Data Set are _all_ dated _prior_ to the release of OCCT v4.5.1 (which was on 14 August 2017). This is very significant because the OCCT v4.5.1 release note states "Added Ryzen and Skylake-X support". So it would seem that all of these people were testing with a version of OCCT that didn't stress Ryzen specific instructions (quite possibly AVX).
Well AIDA64 , x265 or x264 encoding, and Prime95 newer than 26.6 all use AVX.

Unless your Ryzen CPU fails the kill-ryzen script in Linux with stock settings , then it likely is something else.

Maybe your RAM isn't fully stable.
 

·
Premium Member
Joined
·
2,749 Posts
Try with Prime95 28.9 >> Custom >> Min 128 / Max 128 >> Run FFTs in place.
If the CPU is unstable the test will fail in less than 30 mins. If not, the CPU itself is not the issue.
 

·
Registered
Joined
·
18 Posts
Discussion Starter #20
Quote:
Originally Posted by AlphaC View Post

Maybe your RAM isn't fully stable.
I'm certain that it's not an issue with the RAM itself. I ran memtest86+ for 3 days straight with both base and overclocked settings and didn't get a single bit error. Not to mention all the other tests that I ran that stress memory usage. Additionally, OCCT LDS hangs the system regardless of whether RAM is overclocked or not as I stated in a previous post.
 
1 - 20 of 27 Posts
Top