TLDR: Does anyone with a rev 1.1 of Aorus Xtreme (or other Gigabyte TRX40 board) still have the ability to set AVX offset in the BIOS? And if so, what BIOS are you using? (the latest is FBh for rev 1.1)
I have rev 1.0 of Gigabyte Aorus Xtreme, 3970x, using nocuta’s 140mm air cooler (with extra fan). I ran the Prime95 test with 16k fft and everything passed (ran for over 20min). But interestingly my cpu clocks were at 3.5 GHz instead of the advertised 3.7. I know that sometimes (usually?) there is an offset for AVX loads so I went into the BIOS to check and it seems that the ability to set an AVX offset has been removed in the latest BIOS (F4p). Also, the PC Health page now only shows voltages; it used to include a lot of temperature readouts. From my understanding the switch from rev 1.0 to 1.1 was changing a DDR PWM from analog to digital. I have very limited electronics knowledge, but my research seemed to indicate that a digital controller would allow for programmatically set states, where as the analog one was bound to the hardware installed. I wonder if this has caused Gigabyte to remove control of setting AVX offset on the rev 1.0 boards. Does anyone with a rev 1.1 still have the ability to set AVX offset in the BIOS? Or is there another explanation of why my all-core frequency was 3.5 GHz instead of 3.7? (I could run Cinebench R20 again, but when I ran it before my all-core was wither 3.7 or 3.75 under load)
every time I’ve run into this, it was because thermals. whats hwinfo64 say about thermals during this period? tried it “side off” with a box fan or desk fan pointing inside your machine? you might be surprised…
I definitely don’t think it was thermals. My CPU temps stayed pretty constant at 77 deg C and max was 78 deg C. HWinfo also has the section regarding thermal throttling (3 lines) and all of them said it was not thermal throttling.
Got more data and the results are kind of interesting…
Ran Prime95 and Cinebench23 with side panel off (Fractal Define 7xl), close to same result for Prime95: temp was at about 75.3 deg C (compared to 77 deg C with panel on) and CPU Freq raised slightly to 3.55 (from 3.5 with panel on), ran for over 20min. Interestingly, according to task manager the cores never ran at 100% even when starting from 40 deg C temp; start at 96% (which correlated to CPU frequency of 3.6 and settle to 95%. For Cinebench processors always ran at 100%, steady state temps were appx. 72.5 deg C and CPU freq 3.8.
The one consistent data point between Prime95 run and Cinebench run was that HWInfro64 reported a CPU Package Power of 280W thru both tests (fluctuate only slightly around this #). I’m under the impression that Cinebench does not use AVX, so I’m wondering if this what is causing Prime95 to run at the slower speeds? That the limiting factor is the 280W power draw and for some reason this means lower core speeds when running AVX load?
Running now. I’m observing the start up conditions more closely because those shouldn’t be thermal throttled. Interestingly the cpu package power doesn’t hit 280W right away, its around 278 but the CPUs are still limited to 95% and about 3.55 GHz. the VRM MOS starts around 50 deg after a short while. I’m 5min into another Prime95 run and its (the VRM MOS temps) raised to 61 deg C
After about 13min it seems I’ve hit steady state temps, power draw. The CPU package power is fluctuating between 278.7 and 278.9 W roughly, the VRM MOS temps has maintained 68 deg C for awhile now (this is Prime95 with the 16k fft settings)
Thanks for the feedback @wendell! I tried Aida64 stress test (preferences indicated that it would use AVX work) and after running for 5 minutes the frequency was about 3.9GHz with a CPU temp of 84 deg C and package power of 280W, which all seems good, so you’re probably right about it being a Prime95 thing.
But, when I was doing some test with Prime95 on my system and mentioning it in another forum. Someone whom I believe to have more knowledge about this then myself, said, you need to let prime95 warme up first, and after 30min or so System depending you should see it start doing it’s actually calculations. That the first 30min is its testing itself phase.
I am currently in quarantine (negative - due to compassionate international flight) so I can’t test my point for you.
However run it for 60 min or so if not through av24-hour cycle if it is cool enough
