• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Non-K OC attempt

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

mackerel

Member
Joined
Mar 7, 2008
Since everyone else is playing with Kaby Lake, and I'm too tight to get one while waiting for Ryzen, I just decided to have a go at non-K OC.

Test system:
i3-6100
MSI Z170A Gaming Pro, BIOS from http://overclocking.guide/download/msi-z170-pro-non-k-oc-bios-2/
2x4GB Crucial 2666 something or other
Deepcool Gammaxx 200T cooler

I can change bits if needed later on. Above is just what happens to be in the box right now.

Before I installed the bios, I tried seeing how far I can get. The answer is: nowhere. 103 refused to boot at all. 102 booted, but once in Windows the reported clock was still 100. Clearly a waste of time so I moved onto the bios flash, which has just been done. Now to start tinkering!
 
Yeah, bclk overclocking won't work with conventional bioses because the PCI-e bus and the BCLK are locked together and the PCI-e bus just won't tolerate much change at all from stock frequency. As you know, there are some modded bioses for some motherboards that will allow you to do this on Skylake anyway and some motherboard manufacturers (e.g. ASRock) offered some socket 1151 boards with a clock gen chip that allowed you to get around this.

So how is the modded bios working out for overclocking this CPU?
 
Last edited:
Skylake chipset doesn't have the bus limitation, but Intel still had it implemented in the system as a whole to block it from non-K CPUs. I'm running the mod bios at the link in OP now.

CPU is currently testing at 125x37 = 4625 MHz at 1.35v set. Temps are <60C running Prime95 small FFT, but the expected performance tank is there from not having functional AVX. I find it odd, the AVX and AVX2 is still reported as present, and Prime95 says it is using FMA3 (which came with AVX2). How bad is the performance tank? About 3x, which fits with my previous observations that Haswell AVX2 was 1.5x IPC over Sandy Bridge, and on limited info Sandy Bridge AVX seemed to give about 2x IPC over Nehalem?

What's a good combination of stress tests that don't use AVX/AVX2? It's obvious Prime95 is not stress enough in this case. I have Aida64 running with CPU, FPU and cache settings at the moment.

I've not stopped clocking at this point by any means, but I'm doing some other things so might as well see if this is stable as a reference point.

I should have a added, individual core temp readings aren't working but it still reports the CPU overall temp. I need to drop out of non-K OC mode at some point and check how these match up.
 
Nearly all Z170 popular motherboards have BIOS which you can use to unlock non-K processors. It's only matter of microcode. Some features won't work but for most users it's not an issue.
 
Realbench? Quite though, without breaking the temps, bu quick to find instability (Sillicon Lottery uses it 1 hour to test their OC).
 
Celerons and Pentiums don't have AVX anyway, do they? How important is AVX and what does it do?
 
Celerons and Pentiums don't have AVX anyway, do they? How important is AVX and what does it do?

from wiki:
Software

Blender uses AVX2 in the render engine cycles.
Prime95/MPrime, the software used for GIMPS, started using the AVX instructions since version 27.x.
dnetc, the software used by distributed.net, has an AVX2 core available for its RC5 project and will soon release one for its OGR-28 project.
Einstein@Home uses AVX in some of their distributed applications that search for Gravitational Waves.


https://en.wikipedia.org/wiki/Advanced_Vector_Extensions
 
Found this on Wikipedia. Looks like it only makes much difference in some niche applications.
 

Attachments

  • AVX.PNG
    AVX.PNG
    33.2 KB · Views: 1,132
Still testing 4625 at 1.375v, averaging 52C with Aida64. Seems like P95 even without AVX is running hotter than that. It's now passed the previous BSOD runtime. How long to give it? An hour? Then I'll get realbench out.

AVX and AVX2 are part of the single instruction, multiple data category. If something uses it, it can provide a huge benefit over not. In applications similar to Prime95, Haswell is 3x higher IPC than Nehalem. If an application doesn't use it... it does nothing! The apps I use which can make use of it are LLR, PFGW and genefer. All are prime number testing software, and for practical hardware purposes, can be compared to running Prime95. That's not to say other software doesn't use it, but those are ones I know that do. For gaming as a target, the vast majority don't use it.
 
Cleared 1h20m of Aida64 with CPU, FPU, cache selected. Installing realbench now.
 
Realbench BSOD in under 20 minutes, so I guess it is more stressful than Aida64. Dropped clock to 4.5GHz now at same voltage.
 
The prime finding applications I mentioned earlier.

The realbench run at 4500 must be close to an hour now, which is how long I set it for. I'm debating if it is worth running graphical loads, but would they be more GPU limited than CPU? I'll throw firestrike at it as a minimum (280X in that system), but can't do timespy as Win7. I have GTA5 and Doom already on that system also so they could give a real world gaming stress.

Edit: it just finished. I note the logged temperatures reached 81C peak, 71C average. That's a lot hotter than Aida64 got, and is enough to make me think about upgrading the cooling... if that might be a contributing factor at 4625 as I suspect CPUs can be more stable at a given clock/voltage if temperature is lower (opposite case to to thermal runaway).
 
Last edited:
Now this is interesting... I observed the power consumption of the system while Aida64, Realbench, and Prime95 were running (not at the same time!). Aida was drawing around 93W, P95 around 100W regardless of FFT size, Realbench varying from 145W average, with short peaks to 170W. I understand it is a blend of several different tasks, so on my to do list is to work out which of those tasks is the thirsty one.

I also had a thought on P95, and the weirdness that it is running FMA without working AVX. What if I explicitly disabled AVX so it used another instruction? I added the line in config, and... it did worse than expected there. Performance dropped by a few %, and power consumption increased similarly.
 
Realbench uses AVX for x264 encoding, hence the BSOD. Aida64 and P95 26.6 as far as i know are the only ones that stick to SSE (non-AVX). Or add the line CpuSupportsAVX=0 to "local.txt" in your Prime95 directory if you still want to use 28.10.

- - - Updated - - -

What applications use FMA3?

x265 encoding seems to use AVX2/FMA3.
 
Realbench uses AVX for x264 encoding, hence the BSOD.
Could you clarify what you mean there?

To sum up what I've seen so far, AVX does seem functional, but very slowly like it wasn't there. My early run BSOD, but a later run with more voltage was fine. So although AVX was being used, it wasn't stressing like it would be if it were operating correctly. In looking up Realbench's own page, it seems one of the tests it does can use GPU acceleration which might explain some of the power increase, but still, the CPU temps I saw running it were higher than anything else.

I've decided to expand this as my test system going ahead. It is bed time now, but I've dug out my spare H110i GTX 280mm AIO. I will install that tomorrow and see if I can push much further. I don't have a case that it'll fit in either, which is why it is currently unused. I've ordered a Fractal Core 3300 as the cheapest case that would fit it. Replacing the old cheap case it is currently in has been a wish list item for a long time. Hmm... anyone analysing this build would think I'm crazy :) Almost as crazy as buying the i3-7350k...
 
I saw a few threads like this on LTT, overclocking a non-k CPU seems to break AVX somehow, it's there, it's recognised but doesn't function as it should. One of the tests for Realbench is x264 encoding which uses AVX 1 (p95 27.7, IBT). If it isn't working properly it's normal that it would crash/freeze/fail.


http://www.pcauthority.com.au/Featu...nce-from-non-k-series-intel-6th-gen-cpus.aspx

"When overclocked, Non-K CPUs automatically have their integrated GPU disabled, so you’re also going to need a dedicated graphics card. The temperature monitoring will also be buggered, with the chip reading 100C or other crazy temps, so you’ll need to disable any features on your mobo that shut off the chip if the temperature goes too high. Most power-saving features, such as C-states, will also be disabled, so the overclocked CPU will always be running at top speed using full power. Finally, and this is a biggie for those who do a lot of media transcoding, scientific or financial applications, the Intel Advanced Vector Extensions (AVX) performance plummets. We’re not exactly sure why this is, but gamers and general desktop users shouldn’t be fussed about this, as these kinds of scenarios don’t use the AVX instructions."
 
Last edited:
You're also replying to a thread I posted on LTT :)

I just cancelled my order for the Fractal Core 3300 and swapped it for the Cooler Master Masterbox 5. That will also take a 280 rad, looks nicer, and is slightly cheaper although not in a significant way.

Getting the coffee in now before revisiting the system.
 
Yep, i remember the "fishy name" comment :rofl: Is this CPU worth the hassle though ? i would've traded it for a K and get it over with. You get broken AVX 1 and AVX 2, temp readings, power saving features yadda yadda yadda. Fair enough nothing that really hurts a gamer but...
 
Back