4 ms·

My 3970X on the ASRock Taichi (with default settings, generally) does not seem to reproduce this issue at this time - the system remains operational despite the

My 3970X on the ASRock Taichi (with default settings, generally) does not seem to reproduce this issue at this time - the system remains operational despite the FMA3 path being used (I'm assuming this is behind the AVX2 flag? disabling FMA3 leads to a plain AVX path) while running an all-core test with 16K FFTs in Prime95. Either a slight background workload (Windows seems to be trying to use half a core for an OS update) resolves this, or this board does not have a broken power design?

franzb 7y ago
Thanks for sharing your experience. What version of Prime95 are you using? Make sure to use the latest one (v29.8 build 6). The only CPU options I see in the torture test settings are: (1) Disable AVX-512 (grayed out since unsupported on this CPU), (2) Disable AVX2, (3) Disable AVX. There's nothing about FMA3.
- ntauthority 7y ago
  Yeah - I explicitly updated to the latest one; the FMA3 setting is one that existed in prior builds in local.txt so I toggled it off there just to be sure I was hitting an AVX2 code path (in case it didn't mean the UI saying FMA3 in each worker window), but it seems to interpret AVX2 as being synonymous to FMA3 I guess.
  - hrgiger 7y ago
    same here on 1950x, I built from source, when toggled avx2 it shows "using type-1" I dont know what it is: https://pastebin.com/tPuYzYC0 https://pastebin.com/tPuYzYC0
- c2h5oh 7y ago
  Consider trying a couple different versions/builds of prime95 If you look at just the list of issues fixed in v29.x series there is more than one that is close to the code path that is problematic here: https://www.mersenneforum.org/showpost.php?p=508842&postcount=2 https://www.mersenneforum.org/showpost.php?p=508842&postcoun...
leeter 7y ago
I'm curious too because looking at pictures of the VRM it looks like it has less output filtering. My guess is that the board firmware tells the CPU to be less aggressive until it can deliver the power because it has a less overbuilt VRM (still 90A power stages though...). That probably means more of the phases are always active which in this case is probably an advantage as it means there is no long ramp to turn on a phase. I'm going to guess the fact that the phases are doubled also means the board is less likely to turn off phases as it would lose two not one.
KingMachiavelli 7y ago
I'm having the same results. If I disable AVX (AVX1) and FMA3 then I just get no usage of AVX or anything. If I just disable FMA3 then it just uses AVX. I'm on a 3900X.

franzb 7y ago

Thanks for sharing your experience. What version of Prime95 are you using? Make sure to use the latest one (v29.8 build 6). The only CPU options I see in the torture test settings are: (1) Disable AVX-512 (grayed out since unsupported on this CPU), (2) Disable AVX2, (3) Disable AVX. There's nothing about FMA3.

ntauthority 7y ago
Yeah - I explicitly updated to the latest one; the FMA3 setting is one that existed in prior builds in local.txt so I toggled it off there just to be sure I was hitting an AVX2 code path (in case it didn't mean the UI saying FMA3 in each worker window), but it seems to interpret AVX2 as being synonymous to FMA3 I guess.
- hrgiger 7y ago
  same here on 1950x, I built from source, when toggled avx2 it shows "using type-1" I dont know what it is: https://pastebin.com/tPuYzYC0 https://pastebin.com/tPuYzYC0
c2h5oh 7y ago
Consider trying a couple different versions/builds of prime95 If you look at just the list of issues fixed in v29.x series there is more than one that is close to the code path that is problematic here: https://www.mersenneforum.org/showpost.php?p=508842&postcount=2 https://www.mersenneforum.org/showpost.php?p=508842&postcoun...

leeter 7y ago

I'm curious too because looking at pictures of the VRM it looks like it has less output filtering. My guess is that the board firmware tells the CPU to be less aggressive until it can deliver the power because it has a less overbuilt VRM (still 90A power stages though...). That probably means more of the phases are always active which in this case is probably an advantage as it means there is no long ramp to turn on a phase. I'm going to guess the fact that the phases are doubled also means the board is less likely to turn off phases as it would lose two not one.

KingMachiavelli 7y ago

I'm having the same results. If I disable AVX (AVX1) and FMA3 then I just get no usage of AVX or anything. If I just disable FMA3 then it just uses AVX. I'm on a 3900X.