![]() Yeah, there's a problem with explanation - cache quantity per thread affects the FFT size thresholds at which cache and memory bottlenecking kick in. And AVX or FMA3 enabled makes this even worse. People simply don't realize that no program you can normally run will cause as much vdroop as small FFT prime95. If you monitor this when doing small FFT and 1344K in place fixed FFT (Checkbox checked), you will see the difference in vdroop, and in amps draw in HWinfo64. If you have an Asus Maximus XI board, or access to the "VR VOUT" sensor on a Z390 board in HWInfo64 (most Gigabyte and MSI boards do), this is the CPU on-die sense, meaning the true cpu vcore. That's the issue here.ġ344K functions like most real world heavy programs while keeping the CPU loaded, under realistic conditions, which is why it is a good vcore test. Eventually - testing Prime95 takes ages and is tedious AF )ġ344K is a good CPU Vcore test because it is large enough to test the entire CPU subsystem without repeatedly hammering the chip like small FFT, causing power virus loads which can push a stable chip to being unstable *just* because of 1) high heat (heat can push a borderline stable chip to unstable because heat degrades clock signal integrity) from constantly hitting the cores and cache without RAM access, 2) the much higher current load of small FFT causing higher CPU core 'vdroop' (you need a very accurate sensor, or CPU on-die sense measurement to measure this!) pushing the CPU voltage even lower than before, past the point of stability. If I can get some more info on how this was all worked out, that would be awesome for the whole community. What methodology was used to prove this works and come up with this list of FFT/voltage pairings? I'd love to redo this testing as part of my investigation and see if it's still applicable post-Skylake.Ĭan anyone point me in the right direction? I've already come up with enough data to replace Prime95's presets with something a lot better, it would be awesome to be able to check whether "1344k tests ur CPU" is actually still true as it's still regularly quoted by a lot of folks (included myself until recently) and the quality of guides to Prime95 available right now are pretty poor and dated. Thing is, this is all old information that applies to Devil's Canyon chips. At least, not in English - I found a link in the comments section of Der8auer's 2014 article about Prime95 that points to a community forum post in German (link below) that contains a fairly comprehensive list of FTT sizes and the voltages they test: ![]() I'm trying to find out what methodology was used to come up with this idea and the size lists I've seen, but I've not been able to find any info about it online. ![]() One of the things I want to look into is the claims that specific FFT sizes can be used to test whether specific voltages are set correctly. Some of you have probably noticed I'm doing a deep dive into Prime95 right now.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |