Overclocking CPU/GPU/Memory Stability Testing Guidelines

[DilipVasan]

I've seen people claiming that overclocking caused them huge repair bills is it so?

 

[Markfw]

I've seen people claiming that overclocking caused them huge repair bills is it so?

If you do it recklessly, sure. Anyone can wreck a car, computer, anything if they do it badly enough. You need to qualify that.

IF you use good recommended voltage limits, and have good cooling, its almost impossible to kill a CPU. That voltage limitation is highly speculative, and there are a lot of opinions. But if you stay below what 90% of people say, then its really hard to damage one.

 

[hasu]

This thread is a work in progress. The OP will be updated/amended to reflect information as made available by thread contributors.



HCI memtest is superior to LinX, Prime (large or small), and memtest86+ for determining memory stability.

• Launch one instance per thread supported by CPU (8 instances of HCI memtest for 2600K)
• Set each HCI instance to use an appropriate fraction of the memory...16GB on a 2600K means each HCI instance (there will be 8 instances) are to use 2048 MB.
• Stability Criterion: Must pass 200% coverage minimum, passing 1000% coverage is preferred (considered gold standard)

Looks like HCI memtest is a windows program, so in what ways is it superior to memtest86+, that comes with Linux Live CDs?

 

[Idontcare]

Looks like HCI memtest is a windows program, so in what ways is it superior to memtest86+, that comes with Linux Live CDs?

Not sure, I just have experience where memtest86+ fails to detect memory errors that HCI memtest does detect. But have never had HCI memtest fail to detect errors that memtest86+ detected.

Others have reported the same. So in light of the situation, better to rely on HCI memtest if you can.

 

[nsafreak]

The deluxe edition of HCI memtest is available as a bootable ISO image that runs on Linux but it looks like that's the only way to get it to run on Linux which unfortunately does involve payment.

 

[hasu]

Not sure, I just have experience where memtest86+ fails to detect memory errors that HCI memtest does detect. But have never had HCI memtest fail to detect errors that memtest86+ detected.

Others have reported the same. So in light of the situation, better to rely on HCI memtest if you can.

I see... I was always under the impression that memtest86+ was the best memory tester.
Are there any better and free memory tester that is Linux based or with its own boot cd?

 

[Idontcare]

I see... I was always under the impression that memtest86+ was the best memory tester.
Are there any better and free memory tester that is Linux based or with its own boot cd?

None that I know of 🙁 Maybe contact the developers of HCI and see if they are interested in compiling their program for Linux?

 

[PhIlLy ChEeSe]

Whats with all this spam?

 

[PhIlLy ChEeSe]

Deleted spam links.
Markfw900

World class overclockers there, we got a few here 🙄

 

[Techhog]

So, I learned that OCCT and Furmark are actually pretty much useless for a GTX 970. OCCT couldn't even max it out, and then in full screen it wasn't moving, yet reported 60FPS. Maxwell is weird.

 

[PhIlLy ChEeSe]

Deleted spam links.
Markfw900

World class overclockers there, we got a few here 🙄

Those weren't spam links, not even close to what I seen on this site! I'm an active member here, course this ain't the last site on the net.......

Oh spam like Anandtechs front page lolz! Now that is spam!!!!

 

[ArisVer]

What can we run instead of the OCCT GPU test? It will not run on a server os (the free edition).

 

[PhIlLy ChEeSe]

What can we run instead of the OCCT GPU test? It will not run on a server os (the free edition).

Why would you want to run a video stress test on a server GPU? Try 3Dmark 2006,
https://www.futuremark.com/benchmarks/legacy test DX9

https://www.futuremark.com/benchmarks/3dmark-vantage

Vantage tests DX10 and 3Dmark11 tests DX11

https://www.futuremark.com/benchmarks/3dmark11

Also a good down load place with many tests
http://www.techpowerup.com/downloads/Benchmarking/

 

[ArisVer]

It's my home system, checking stability.

One more thing. I runned 4 instances of HCI at the maximum memory setting of 2048 MB (I have 4 cores, no HT). From the 12 GB RAM I have, there was still 1.7GB available. What you do in this case? Run a fifth test allocating the 1.7 left? Also at this test the first test from the last test had an hour difference (65 hours vs 66 hours and ten minutes). This means the cores are not equal?

 

[Idontcare]

It's my home system, checking stability.

One more thing. I runned 4 instances of HCI at the maximum memory setting of 2048 MB (I have 4 cores, no HT). From the 12 GB RAM I have, there was still 1.7GB available. What you do in this case? Run a fifth test allocating the 1.7 left? Also at this test the first test from the last test had an hour difference (65 hours vs 66 hours and ten minutes). This means the cores are not equal?

Yes, no harm in running more instances of HCI, they all just run that much slower is all.

 

[lazybedone]

fantastic news ... very creative

 

[Chriskey2015]

thanks for sharing.. helped me a lot

 

[CrazyElf]

@Idontcare

A few questions:

What are your thoughts about AIDA64? I believe Intel recommends it.

On a separate note, for Intel Burn Test, would you recommend maximum for lots of RAM? Say someone had 64 GB of RAM or even 128 GB (apparently some of the newer kits have 8 x16GB DIMMs)? The larger RAM means more complex problems? Would 16GB be adequate? Or would you need all 128GB to pass 20 cycles of Linpak?





The only other thing I think might be worth putting on the post is a note indicating the voltage increases with AVX (it can use about 0.1V with AVX enabled).

 

[Idontcare]

Hi CrazyElf, thanks for the suggestions!

For IBT, you want to use maximum ram because your CPU needs to be stable with however much ram you have in the system. Only have 4GB? Then the OC only need be stable with 4GB (can likely have a higher stable OC). Have 128GB? Then the OC needs to be stable even when all 128GB of ram are in play. An unstable OC does you no good if you have a need for 128GB system ram, stability is going to be important.

That said, the OP of this guide is horribly outdated IMO and is in need of someone with modern hardware to step up and draft a new thread on the same topic so we can retire this outdated one. I have no experience with modern hardware as my most recent CPU is an Ivy Bridge. We need people with Haswells and Skylakes to step up and craft a new OC'ing guide in my opinion.

 

[CrazyElf]

Thanks Idontcare.

There don't need to be many changes I think.


For Intel Burn Test
All RAM, for x number of hours (versus x number of cycles).

As indicated, there are large kits now available (G.Skill and I think Corsair now offer 8x 16GB kits), for Haswell-E.

Maybe 2 hours?

For Prime95
Probably could be unchanged.

12 hours again.

For HCI MemTest
I think it may be worth mentioning that the premium version isn't too expensive ($5 USD, or about ~NT$165 at current exchange rates). It offers error logging as well.

For $14 USD, you can get a boot CD as well.


For OCCT
Probably unchanged for the GPU - still the most stressful GPU test, more so than Furmark.

I have seen some people use Folding@Home as a "real world" test, but that isn't as stressful.








As far as warnings go
Intel Burn Test and Prime 95 will lead to CPU temperatures of about 20-30C warmer than most loads (compared to say, x264 encoding or file compression or similar loads a typical end user will do).

OCCT GPU burns will run perhaps 20-30C warmer as well for GPUs, although the difference is smaller when compared to Folding or Coin Mining (now taken over by ASICs).


IBT and the later versions of Prime will enable AVX, which can use up to +0.1V more than other loads.



Maybe a safe voltages list?

For 22nm CPUs:

• Say 1.35V VCore on 22nm CPUs
• Probably advisable to have a system agent (VCCSA), VTT, voltages as well.
• Haswell should also have a maximum VRIN/VCCIN voltage (say 2.1V for 24-7)
• I'd recommend no more than 1.7V VDIMM for DDR3
• As far as general guidelines for the non-E Intel CPUs, delidding will net perhaps 15-30C load by reducing the distance between the core and the die (the large range is because Devil's Canyon generally sees smaller gains from delidding than Ivy Bridge/Haswell original); Haswell-E and AMD CPUs are of course soldered, so this is not needed

For Skylake

• Nobody yet has established a "safe" Vcore voltage, but I'd assume that 14nm would be lower than 22nm, so perhaps 1.25V is a good baseline? I have noticed many reviewers benchmarking at 1.4V - I would guess that is not something you want 24-7.
• Right now not all of the motherboards have even been released including many flagships (Ex: Asus Z170 Maximus Extreme, MSI Z170A Xpower, Gigabyte Z170 SOC Force, etc), so it's really early to tell, but day to day looks like 4.5-4.7 GHz, with a few "golden" 4.8 GHz chips.
• DDR4 maximum for 24-7 should be 1.4V VDIMM
• As far as delidding - Add a warning for those delidding (from Ian Cutress' article) on the thin PCB: http://www.anandtech.com/show/9505/skylake-cpu-package-analysis
  

These are just my figures off the top of my head and my ideas. Final guide could be different.

 

[AdamK47]

Really, SuperPi? I don't believe it.

 

[DanielDi]

Those are some pretty stupendous numbers. I'm very tempted de-lid at some point, but my build is still new - so I think I'll hold off for now.

 

[Brian Stirling]

Much of this sticky goes back to 2011 with a few more recent posts and I'm wondering if there are new or better testing programs out there and if so what they are.

I am just beginning to build a new box with Asus x99-Pro/USB 3.1 motherboard, G.Skill Ripjaws V series 32GB (8GBx4) DDR4-3200 with 14-14-14-34 timings, i7-5820K CPU with a Corsair H100i GTX water cooler, EVGA GeForce GTX 980 Ti Hydro GPU with water cooler, and Samsung M.2 PCIe SSD 950 Pro.

I'm not really that much of a gamer and the main purpose of this box is image and 4K video editing using Adobe CC suite (Photoshop, Ligthroom, Premiere Pro and AfterEffects). My case is the Thermaltake V71 with three 200mm fans (2 intake, one exhaust) and in addition the two 120mm H100i fans and one 120mm 980 Ti fan with one more 120mm intake fan at the bottom.

I plan to OC a bit but not push it. Probably the best point would be just a bit more than the max GHz/W point which I'd guess would be about 4.1GHz so perhaps 4.2GHz or there about's. I figure a similar conservative OCing of the GPU and RAM. The goal of OCing is to improve performance without risking reliability or upping the noise by very much. A large video render could tax the system for a few hours though with a system this powerful that's not that likely. Still, I want it to be stable and not too loud while rendering.

So that brings me to the best CURRENT stress tests and benchmarks. It looks like Prime95 and IntelBurn Test will stress the CPU pretty well but I don't think I want to really push the tests way beyond real world stress levels. For GPU I was thinking UNiGiNE would be a good test with Cinebench for benchmark. For RAM ... MemTest.

I guess what I'm asking is what, in 2016, are the best tests for stress and benchmark programs for a box that's aimed at image and 4K video editing?


Brian

 

[LionelHardesty]

[FONT=&quot]Oh Yeah, Your posting is really helpful to me as these program outside of Prime95 and memtest86+ but I didn’t know otherwise I thing else,[FONT=&quot] i [FONT=&quot]appreciate[/FONT] you[FONT=&quot]. [/FONT] [/FONT]
[/FONT]

 

[cnadshc]

This thread is a work in progress. The OP will be updated/amended to reflect information as made available by thread contributors.



LinX (Intel burntest) is superior to Prime95 small FFT for determining CPU core logic stability.

• Must run with the IBT thread count set equal to the physical core count of the CPU.
• HT slows it down and reduces the ability to determine stability. Set to 4 threads on a 2600K.
• Set memory to "All".
• Stability Criterion: Must pass 5 cycles minimum, passing 20 cycles is preferred (considered gold standard)

Prime95 large FFT is superior to LinX for determining L3$/IMC stability.

• Must use large FFT, blend is insufficient. <- reports indicate this is false for AMD stability tests, see post #4
• HT is ok for this test.
• Stability Criterion: Must pass 2 hours minimum, passing 12 hours is preferred (considered gold standard)

HCI memtest is superior to LinX, Prime (large or small), and memtest86+ for determining memory stability.

• Launch one instance per thread supported by CPU (8 instances of HCI memtest for 2600K)
• Set each HCI instance to use an appropriate fraction of the memory...16GB on a 2600K means each HCI instance (there will be 8 instances) are to use 2048 MB.
• Stability Criterion: Must pass 200&#37; coverage minimum, passing 1000% coverage is preferred (considered gold standard)

OCCT GPU test w/error checking enabled is superior to Kombustor for determining GPU stability. (updated link to OCCT 4.0.0, thanks NoobyDoo!)

• Error checking MUST be enabled by the user (check the box), otherwise you are leaving it up to your eyes to detect visual artifacts which renders the test entirely subjective.
• Stability Criterion: Must pass 20 minutes minimum, passing 1 hour is preferred (considered gold standard)

any good stress tester for testing multiple GPUS at once with SLI/xfire?