Overclock.net banner

1 - 7 of 7 Posts

·
Registered
Joined
·
12 Posts
Discussion Starter · #1 ·
Picked up a Z97 rig after not being able to OC x5670 Xeon in my Gigabyte X58 board. Just doing some testing and the stock temperatures on the 4790k that I bought seem awfully high. Batch number X437B357. I first used a spare air cooler that I had laying around, Zelman ENPS10X Optima, at it was hitting 80+ in Prime FFT test so I went out and bought a Nepton 140XL thinking it was the cooler but the temps are just as bad as before. I checked the Arctic Silver 5 application and it looked OK (see photo). The second photo is where I tried to use less of the thermal compound but it didn't change the temps any.... if anything, they may have gone up by a degree or so. The CPU head spreader seems to sit nicely on the cooling block as I was not able to rock it when I laid the CPU on top of it (not very scientific I know).

Motherboard: ASRock Z97 Extreme6
Case: Fractal Design R4






 

·
Premium Member
Joined
·
10,755 Posts
Quote:
Probably fine. You're using ~1.2vcore with a synthetic FPU test.

Yknow what happens when you run those on Haswell architecture?
tongue.gif


366c4dd8_z.png


x264 test there = 100% CPU load on all cores video encoding - every nonsynthetic is 57c or cooler

Linpack (100c and throttling) = synthetic FPU test

If you assume 20c case temperature, that means temps are 37c over ambient while encoding and 80c over ambient during synthetic FPU test

power consumption roughly doubles when going from 100% load on all cores video encoding vs 100% load on all cores with synthetic FPU test and very fast RAM

etcetcetc

the problem is not with the chip or the TIM, unless you consider "runs certain types of workloads unusually fast" to be a problem. Just don't test in those ways if you want to overclock in the ways that most people do.

If you want probably the hardest test to pass for vcore, run prime95 with custom test, fft size 1344-1344. Watch it fail with low temps - there are tests out there ~30-40c hotter that need less vcore

If you want a more real test, read the OCN guide (you should probably do that anyway before overclocking - http://www.overclock.net/t/1411077/haswell-overclocking-guide-with-statistics - grab the x264 package from here - https://mega.co.nz/#!3tAGnAqZ!QbCz2r1fG0WjM8DgGYeExngGypaHftAzPUgTSn2kAdk - and run the 64 bit batch file with 16 threads. You'll also need to adjust stuff like input voltage while OCing higher, and you should manually set your uncore/cache frequency and voltage - try locked 33x @1.15v until you have the core overclock on the highest level that you want, then adjust it after everything else is set
 

·
Registered
Joined
·
12 Posts
Discussion Starter · #4 ·
Thank you for the quick and very informative reply. I will definitely read through the OC guides you linked to before I start trying to overclock. Right now I just want to test everything at stock and make sure that I don't have any bad components before the return/exchange window expires. I was distraught with the temps I saw because that's only at 1.182V so assumed that OCing it any higher would be a definite no-go due to the additional heat that will be generated. I also saw a lot of posts where people claim much lower temperatures under load... but it was rarely made clear what their loading conditions were.

Do you have the source of the graphic that you posted, or if not, know what cooling solution was being utilized? I would like to run some of the other benchmarks listed and compare my temps to those.
 

·
Premium Member
Joined
·
10,755 Posts
Quote:
Thanks for the quick reply... nice graph, do you know what type of cooling was being used?
It was a 4670k at ~1.25v with an nh-d14 IIRC. It's from the Haswell overclocking guide with statistics here on OCN, by Darkwizzie - THE Haswell thread, ignore most of the devil's canyon stuff because it's generally not as high quality as the original thread or discussion there. These guys have often been around since week 1 of Haswell (like me) and some regulars just sat around and read 15k posts in the thread so there are more people there who actually know what they are talking about~

I linked that thread in my first post - it's in one of the spoilers in the OP, stressing/temperature discussion

compare temps with x264, it's pretty good, not as harsh as some other tests but easily usable to set the most important voltages while overclocking. Vcore, Input voltage at least - if you can overnight x264 and then add a tiny bit extra to those, you can run an OC for years and not see any signs of instability

IIRC, i loaded CPU at 1.3vcore with silver arrow and ~17c ambients and my load temps hit about 73 in x264. That was without hyperthreading, but it was also with a 4770k, not a 4790k so the DC temperature improvements would help. You should be able to approach 1.4vcore without HT and comfortably use 1.3 with HT, AFAIK.

By 1.25vcore any any cooler or clc is expected to temperature throttle @100c with synthetic FPU stress (even though they wouldn't hit 60c video encoding at 100% load on all cores) - Haswell quad cores are able to break 210gflops at 4ghz with linpack, while first and second gen core were only able to hit about 40-60 (without avx) and ~120-130 (with avx1), IIRC, which is where the huge temperature gap comes from. If you're one of the 5 people on the planet using that type of workload, just run an OC around 1.0vcore and be happy with it being by far the fastest CPU architecture available for them
biggrin.gif
 

·
Premium Member
Joined
·
10,755 Posts
Quote:
Originally Posted by PontiacGTX View Post

Just dont run stress test based on avx2 and you will have regular temps
You can run avx2 tests fine, x264 uses avx2 and gets a performance increase from it, that's part of why Haswell shows a 20% IPC gain over Sandy Bridge for video encoding with it - but the synthetic tests that focus entirely on FPU and run 1.5-2x faster because of avx2 existing and not just avx1 (or no avx at all) are a problem
 
1 - 7 of 7 Posts
Top