AVX(2) and Z97-E (4790K)

[blppt]

Got a really odd situation here....my Z-97E, even under full load with the AVX heavy Prime95, will never get into the 90s with a CM Hyper-T4 cooler, and wattage indicated in CoreTemp never goes into the 100s.

Now, with the same model cpu (4790K), with the same cooler, in a Gigabyte Z97X-Gaming 7 mobo, when running the AVX heavy Prime95, will hit 150W (according to CoreTemp) and surpass CPU throttling temps very quickly.

Is there something on ASUS mobos that disables or mitigates the standard AVX voltage bump in Haswell CPUS? It would seem that the Gigabyte board lacks such a feature. And yes, I've reseated and re-pasted the cooler several times, and verified that the stock voltage settings are normal under load (according to CoreTemp, which I'm sure does not indicate the internal haswell AVX voltage boost).

Both motherboard boost all cores to 4.4 with XMP enabled (non standard turbo, but I verified the temps are similar with XMP disabled on the Gigabyte, so that isnt it).

Any clue as to why the ASUS board doesnt boost wattage to insane levels when using AVX extensions in P95? Could it be that a new Gigabyte Bios reflashed some CPU microcode as well?

[Korth]

Asus and Gigabyte use entirely different VRMs on their motherboards, different hardware components, different logic controllers, different firmcode and different BIOS programs which make it all run. Different motherboards will often produce results on hard overclocking. Your tests appear to indicate the Z97-E has finer voltage and current regulation than the Z97X-G7 under stress loads.

You've reinstalled and repasted the cooler (and processor? and DIMMs?) several times, which largely rules out mounting variances. But performance at the extremes can vary significantly from install to install, the timings and voltages are so tight that trivial quirks all add up into noticeable variations at maximal performance thresholds.

[blppt]

Yeah, but a difference of 60W (150W vs ~90) under AVX2/FMA3 load (for a 90-100W TDP CPU) seems like a rather odd difference, no?

I'm starting to wonder if maybe the AVX/FMA instructions arent actually in use during the P95 test on the ASUS, because it is a common Haswell issue, even people with custom loops cant get SmallFFT P95 (the versions i think, post 27.x, included the latest release) with 4790K under 100C, and I'm on air.

It says "FMA3" in the P95 window tho. Just really perplexing to me. Unless theres a setting in the BIOS (or hardcoded by ASUS to keep people from melting their motherboard/cpu) that counters the Intel AVX2 spec that we know bumps voltage under such a load.

Thanks for the reply!

[Korth]

Can you confirm your results with other testing software? I don't trust software, software is full of bugs, software lies, lol.

[blppt]

Well, I'm going to make sure I have the latest P95 for both systems, but I'm at least positive both clients are past the "Danger Zone" of P95 versions for Haswell (27.x+). Something is amiss here....

[blppt]

NM, figured it out----apparently the non insane wattage ASUS system was running a slightly older build of P95, v27.7, which I *thought* still was the Haswell killer, but when I dl'd the latest P95 available, it hit 150W like the Gigabyte one.

Well, mystery solved. Thanks anyways!