Strix Z390-E 9700k, ANY AVX offset causes crash

[Duckysaysquack]

Hi all,

I recently built a new system and have a perplexing problem that I can't seem to solve. I've tried pretty much every option and setting so I'm reaching out for help. I have a Strix Z390-E and 9700k. I've managed to get it stable to 5.0ghz at 1.330v with llc level 6. I've run 1 hour of Prime with AVX instructions disabled and stress loads are at 1.332v and temps are pretty good. One or two cores get up to 90+ degrees for a few seconds but there's no thermal throttling. However the moment I try prime with AVX instructions, it crashes . My solution was to introduce AVX offset but it seems like ANY AVX offset creates a quick hard crash in PRIME WITH AVX DISABLED.

I think this is very odd because I am using Prime with AVX instructions DISABLED like before so it should be running the same test as before. Prime WITH AVX enabled crashes but that's not surprising. I've tried increasing voltage, made much bigger AVX offsets, different LLC, etc, it seems like every time ANY AVX offset causes a hard crash. I've tried lower LLC and that seems to last a little longer before crash, but llc 5 and lower requires a MUCH higher voltage. Ie I need 1.365v at llc5 to maintain 1.332v on load and still crashes.

TL DR:
5.0ghz 1.33v llc6 AVX offset 0 - prime AVX disabled success, prime AVX enabled crash
5.0ghz 1.33v llc6 AVX offset 1 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.33v llc6 AVX offset 2 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.33v llc6 AVX offset 3 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.33v llc6 AVX offset 4 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.33v llc6 AVX offset 8 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.395v llc6 AVX offset 2 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.395v llc5 AVX offset 2 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.395v llc4 AVX offset 2 - prime AVX disabled crash, prime AVX enabled crash

I've tried many of the other settings, increase frequency from 140% to 170%, etc, nothing seems to work.

[mdzcpa]

When you set your AVX Offset, due you see your CPU speed just before crash? Is it limiting your CPU speed to 4800 with Offset of 2 (as example)?

[Falkentyne]

Duckysaysquack wrote:
Hi all,

I recently built a new system and have a perplexing problem that I can't seem to solve. I've tried pretty much every option and setting so I'm reaching out for help. I have a Strix Z390-E and 9700k. I've managed to get it stable to 5.0ghz at 1.330v with llc level 6. I've run 1 hour of Prime with AVX instructions disabled and stress loads are at 1.332v and temps are pretty good. One or two cores get up to 90+ degrees for a few seconds but there's no thermal throttling. However the moment I try prime with AVX instructions, it crashes . My solution was to introduce AVX offset but it seems like ANY AVX offset creates a quick hard crash in PRIME WITH AVX DISABLED.

I think this is very odd because I am using Prime with AVX instructions DISABLED like before so it should be running the same test as before. Prime WITH AVX enabled crashes but that's not surprising. I've tried increasing voltage, made much bigger AVX offsets, different LLC, etc, it seems like every time ANY AVX offset causes a hard crash. I've tried lower LLC and that seems to last a little longer before crash, but llc 5 and lower requires a MUCH higher voltage. Ie I need 1.365v at llc5 to maintain 1.332v on load and still crashes.

TL DR:
5.0ghz 1.33v llc6 AVX offset 0 - prime AVX disabled success, prime AVX enabled crash
5.0ghz 1.33v llc6 AVX offset 1 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.33v llc6 AVX offset 2 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.33v llc6 AVX offset 3 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.33v llc6 AVX offset 4 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.33v llc6 AVX offset 8 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.395v llc6 AVX offset 2 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.395v llc5 AVX offset 2 - prime AVX disabled crash, prime AVX enabled crash
5.0ghz 1.395v llc4 AVX offset 2 - prime AVX disabled crash, prime AVX enabled crash

I've tried many of the other settings, increase frequency from 140% to 170%, etc, nothing seems to work.

This is explained here.
https://rog.asus.com/forum/showthread.php?106375-MCE-explanations-and-others

It's a combination of guardband voltage increasing due to the AVX offset, combined with flatter loadline calibration and worse transient response.
To address this issue, raise your bios voltage and drop the LLC down a level until you get the same absolute load voltage as before. Lower LLC aids transient response greatly. Maxing PWM phase control and switching frequency will also help slightly if you do the other things along with it.

[Duckysaysquack]

Falkentyne wrote:
This is explained here.
https://rog.asus.com/forum/showthread.php?106375-MCE-explanations-and-others

It's a combination of guardband voltage increasing due to the AVX offset, combined with flatter loadline calibration and worse transient response.
To address this issue, raise your bios voltage and drop the LLC down a level until you get the same absolute load voltage as before. Lower LLC aids transient response greatly. Maxing PWM phase control and switching frequency will also help slightly if you do the other things along with it.

Thanks for your input. If you check my list of trials, my baseline is 5.0 at 1.330v and llc6. I've tried LLC5 and LLC4 at 1.395v. I did notice that LLC 4 and 5 lasted longer than LLC6 under prime non AVX. LLC6 causes a crash within minutes and LLC7 is a near instantaneous crash. I don't know if I can go LLC3 as the Vdrop will be too high and I will need a much larger vcore. However I think 1.4v+ voltage for a 5.0 OC may be a bit high in terms of temps? I will try maxing out PWM Phase control and switching frequency and report back.

[Falkentyne]

Duckysaysquack wrote:
Thanks for your input. If you check my list of trials, my baseline is 5.0 at 1.330v and llc6. I've tried LLC5 and LLC4 at 1.395v. I did notice that LLC 4 and 5 lasted longer than LLC6 under prime non AVX. LLC6 causes a crash within minutes and LLC7 is a near instantaneous crash. I don't know if I can go LLC3 as the Vdrop will be too high and I will need a much larger vcore. However I think 1.4v+ voltage for a 5.0 OC may be a bit high in terms of temps? I will try maxing out PWM Phase control and switching frequency and report back.

My question is why even use an Avx offset in the first place?
If it's causing you problems, just disable the offset.
However you also did not say which exact prime95 test (FFT size) you were testing.
This matters. Because I don't believe small FFT AVX stable is worth anything except power virus bragging rights.

You do -not- under any circumstances whatsoever need to stress test prime95 AVX small FFT's ---ever. Unless you number crunch for a living and enjoy 100C temps and 200A power draw.

I am aware that the "1344K" prime in place fixed FFT AVX enabled test is outdated now, but my own tests have shown personally that you can pass Realbench 2.56 for 2 hours (uses AVX) and still fail 1344K AVX in place fixed FFT's after 15min-1 hour. So if you can manage to pass 1344K AVX enabled in place fixed FFT's for 2 hours, Realbench 2.56 should run all day and night, and you should have no windows games or applications that will bring your system down unless temps get unmanageable.

So my advice for you is to ditch the AVX offset, run prime95 29.6 beta 2 (beta 3 has a bug where you can't even start the stress test), and do a custom 1344K in place fixed AVX test with NO offset (disable AVX2 please, FMA3 is only for people that enjoy self torture) and run it for 2 hours. Raise voltage (don't use a higher LLC than LLC6!!) until you can pass this for 2 hours, then run realbench 2.56 and let it loop overnight, which should pass easily.

Then you can relax and be assured of a stable system.

Note: there is a certain RAM tertiary (or subtiming) that GREATLY affects AVX stability and AVX speed, but you have to do your own research on that. The only thing I can find is this quote here, but this mentions HEAT rather than stability:

Fun fact about tRDWR: these timings directly impact AVX. The tighter they are, the hotter AVX is. The looser they are, the cooler AVX is. Those of you that fear AVX, you might be able to use this to your advantage, and make those stress tests easier on yourself. I promise not to judge you.

https://www.overclock.net/forum/18051-memory/1630388-comprehensive-memory-overclocking-guide.html

I do know some RAM timings can really mess up hyperthreaded AVX stability with no change in temps, but this is beyond my knowledge.