7900x OC stability issues.

Soldato
Joined
27 Feb 2015
Posts
12,636
The "plan to use AVX" I feel is a bit of a false question. Since I concluded you use AVX anyway in everyday computing.

When I was deciding on my avx offset on my 8600k, I was reading OC forums, and there was claims that AVX is some kind of specialised function not commonly used.

However I discovered mainstream apps like chrome and most a/v software, and even windows update would trigger the AVX offset, I concluded that indeed most software uses AVX instructions, and I then decided to use an avx offset of 0.

So in short to me, a 5ghz CPU is not a 5ghz cpu if it requires an AVX offset of anything other than 0. So if I was buying pre binned hardware, I would refuse to buy a product not binned with an offset of 0. I am probably a stubborn person in this respect, but it is my point of view on AVX and overclocking.

With that said, I do feel pumping a AVX stress tester is far from realistic.

So you might be able to find a middle ground here.

Try this.

Set offset back to 0
Test a non AVX workload and see if it is stable at the target clocks.
Set a TDP limit in the bios that you know will force the cpu clocks to TDP throttle during a AVX bench/test. But not throttle things like AVX gaming such as gta5.
Then see if that is stable.
 
Associate
Joined
28 Aug 2012
Posts
2,480
Also undervolting the cache when already experiencing instability is questionable advice. Leave both cache ratio and voltage in auto till the problem has been resolved.
however, while trying to get the core stable id suggest sticking it at 30x auto volts for now.

yeesh... i was referring to his comments about temperatures, something to do when hes stable, as you can see. only thing questionable is your eyesight ;)

if it really is a vcore issue id be asking for clarification on the overclock bin from the seller. no use if it doesnt work.


It's pretty much always dpc_watchdog_violation

Think there's only been 1x irql_not_less_or_equal with some different settings and the old ram

yeah this sounds like vcore.

i've seen some games crash faster than dedicated stress tests, probably due to the spiky loads. if you want a decent quick test, cinebench r15, run that 10 times, if it makes it you are fairly (not necessarily completely) stable, its a quick method.

also download hwinfo, it will tell you accurate core temps as well as the package, ive seen some apps are a bit unreliable at temp reporting.
 
Last edited:
Soldato
Joined
5 Sep 2011
Posts
12,838
Location
Surrey
yeesh... i was referring to his comments about temperatures, something to do when hes stable, as you can see. only thing questionable is your eyesight ;)



Not really. He needs to get stable firstly, which is what the thread is about. Best advice would have been to leave both in Auto which you did not stipulate, don't really think there's any room for interpritation there.


if it really is a vcore issue id be asking for clarification on the overclock bin from the seller. no use if it doesnt work.

Likely won't make any difference depending on what was used to estabilish stability. If the conditions the overclock were obtained in were much different to what he's tried already, it's fairly obvious this CPUs stability at 4.7 is conditional at best.
 
Soldato
Joined
5 Sep 2011
Posts
12,838
Location
Surrey
The "plan to use AVX" I feel is a bit of a false question. Since I concluded you use AVX anyway in everyday computing.

Assuming everyones workload is an interesting point of view. Certainly not a valid one. Also, the platform you've used by example is a poor one, as the current draw differences are huge. Not all AVX workloads will trigger the offset.
 
Soldato
Joined
27 Feb 2015
Posts
12,636
regardless, I was finding the offset been triggered really frequently. Nothing like the rarity that has been claimed by various overclockers.
 
Associate
Joined
28 Aug 2012
Posts
2,480
Not really. He needs to get stable firstly, which is what the thread is about. Best advice would have been to leave both in Auto which you did not stipulate, don't really think there's any room for interpritation there.
30/auto is pretty exactly what i test core overclocks with because its completely stable and i'm sure it is on his too, its so overvolted on auto the chances that it crashes because of this = none.



Likely won't make any difference depending on what was used to estabilish stability. If the conditions the overclock were obtained in were much different to what he's tried already, it's fairly obvious this CPUs stability at 4.7 is conditional at best.
exactly so it shouldn't have been binned as a 4.7. if you cant plug in the numbers they give you and have it work (crashing in games is bad) , then he should send it back. (he has a top of the range motherboard too so cant blame that)
 
Associate
OP
Joined
26 Apr 2018
Posts
17
Apologies, busy weekend. Went to post from the first page didn't see everyone else was chiming in!

Tried what you said regarding a bigger AVX offset, went down to -12 and -14 and still the same issue, BSOD around the 11 minute mark even XTU is sometimes fine for 10-20 minutes and sometimes failing inbetween.

Also tried going down to 4.6 (not to sound like a tool if it worked still wouldn't be happy as paid for 4.7, even though it's negligible it's still what was paid for) to no avail, exactly the same BSOD's :/

Still at 1.25v but it's significantly hotter than testing at 1.235v.

Motherboard has been RMA'd and installed 3 times (this is a ballache with a custom loop btw), splashed out on new ram + ran memtest for hours, tested the psu and cooling is more than adequate so ruled out everything there.

Will tryout the next recommended settings and get back to you guys don't want to give up and really appreciate the help from everyone so far but it's heading that way.
 
Last edited:
Associate
OP
Joined
26 Apr 2018
Posts
17
The "plan to use AVX" I feel is a bit of a false question. Since I concluded you use AVX anyway in everyday computing.

When I was deciding on my avx offset on my 8600k, I was reading OC forums, and there was claims that AVX is some kind of specialised function not commonly used.

However I discovered mainstream apps like chrome and most a/v software, and even windows update would trigger the AVX offset, I concluded that indeed most software uses AVX instructions, and I then decided to use an avx offset of 0.

So in short to me, a 5ghz CPU is not a 5ghz cpu if it requires an AVX offset of anything other than 0. So if I was buying pre binned hardware, I would refuse to buy a product not binned with an offset of 0. I am probably a stubborn person in this respect, but it is my point of view on AVX and overclocking.

With that said, I do feel pumping a AVX stress tester is far from realistic.

So you might be able to find a middle ground here.

Try this.

Set offset back to 0
Test a non AVX workload and see if it is stable at the target clocks.
Set a TDP limit in the bios that you know will force the cpu clocks to TDP throttle during a AVX bench/test. But not throttle things like AVX gaming such as gta5.
Then see if that is stable.

Nooby question... I cannot find TDP anywhere in the BIOS, never had to change this on previous builds. Is that Long/Short Duration Packet Power Limit? Seen someone mention setting them both to 4095.

Also, XTU only reports Package TDP as 1-2W when stress testing whereas seen others with around 179w?
 
Associate
OP
Joined
26 Apr 2018
Posts
17
if its not even working at 4.6 id be returning it now. unless its memory related which i dont think it is based on your bsod codes.

Damn, hopefully the RMA process isn't a wait around for it to be tested lost so much time with this already would just like to buy a new one asap.

The only setting I got btw was the 1.235 vcore, not sure if anyone else got more but I called them a week after having issues, got told they're tested by employees who don't write down all the settings and "You should know how to overclock"... eh fair enough.
 
Associate
Joined
28 Aug 2012
Posts
2,480
well fair enough, but if they are going to sell binned chips for a lot more money then they need some validation, i doubt they have time to stress test them for a long time so i would imagine that some fall through the cracks.
 
OcUK Staff
Joined
20 Feb 2012
Posts
10,163
Location
John Smiths Stadium
OK this seems a messy thread...
Easy set xmp, set avx offset to 6 and avx 512 to 7,all cores 47... Min cache 27 max cache 27 Dram speed 3000, digi power cpu current cap 240%, cpu vcore 1.25-1.275v cache auto 0.92v memory 1.375, Vccio 0.92v... Test...

Now mono block in my opinion after a lot of testing are not good on this platform. Normal cpu block is better by far and this is what I advise all customers buying my bundles now. Really a good 10-15c worse on mono...

Test this by using your pc and report back.
 
Soldato
Joined
27 Feb 2015
Posts
12,636
Nooby question... I cannot find TDP anywhere in the BIOS, never had to change this on previous builds. Is that Long/Short Duration Packet Power Limit? Seen someone mention setting them both to 4095.

Also, XTU only reports Package TDP as 1-2W when stress testing whereas seen others with around 179w?

yep its the long/short duration power limit.

4095 would be a value to set if you care not one jot about power consumption.
 
Associate
OP
Joined
26 Apr 2018
Posts
17
OK this seems a messy thread...
Easy set xmp, set avx offset to 6 and avx 512 to 7,all cores 47... Min cache 27 max cache 27 Dram speed 3000, digi power cpu current cap 240%, cpu vcore 1.25-1.275v cache auto 0.92v memory 1.375, Vccio 0.92v... Test...

Now mono block in my opinion after a lot of testing are not good on this platform. Normal cpu block is better by far and this is what I advise all customers buying my bundles now. Really a good 10-15c worse on mono...

Test this by using your pc and report back.

Didn't know that about the monoblock! Can't send it back now though which sucks so kinda stuck with it but worth knowing for the future.

Ok, cleared cmos and set all what you mentioned and finally thought it was stable with a 15 minute test... BSOD on 14 minutes.

At 1.25vcore and above it starts to get way too hot hitting 95oc+ (have reseated block several times) which I didn't expect with 2x 480mm rads, delid and ambient 18-20oc.
 
Associate
OP
Joined
26 Apr 2018
Posts
17
these blocks are not good at all!!Dont stress test just use...

Seen people hitting 80c non-delid at stock speeds with 1 rad, 1.235v hits around 87oc getting close to Tj so might be worth investing if it can drop 10oc... poor bank account.

Only stress tested because playing and using OBS it can take anywhere between half hour to two hours before crashing which mid stream looks awful, same with video/picture editing and everything gets lost.

Will play around a bit and run OBS w/ non-live stream and see what happens, otherwise not sure whether to cut losses and replace.

Really appreciate everyone's time and help with this, will give it awhile and report back.
 

RSR

RSR

Soldato
Joined
17 Aug 2006
Posts
9,627
I've just been reading through this and its interesting to see the differences.

How are you getting on now?
 
Associate
OP
Joined
26 Apr 2018
Posts
17
I've just been reading through this and its interesting to see the differences.

How are you getting on now?

No luck which is disappointing as everyone's taken the time to try and help, it's running ok with a lower clock on some things but after awhile of using OBS or editing software it BSOD's... only been given one choice to send it back for testing. Initially just wanted to order another and send this one back for a refund but apparently that's not a choice.
 
Back
Top Bottom