Um, no? Most server workloads do not benefit much from AVX(512) instructions. With HPC, there are indeed many users, but e.g. those who insist on reproducibility of their floating point computations will avoid contractions like FMA.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Anyone regret getting threadripper after seeing 7980xe leaks?
- Thread starter tangoseal
- Start date
ManofGod
[H]F Junkie
- Joined
- Oct 4, 2007
- Messages
- 12,863
This thread makes me desire a Threadripper build. 
That said, I cannot justify spending the money, although I do have the money to do so.
That said, I cannot justify spending the money, although I do have the money to do so.JustReason
razor1 is my Lover
- Joined
- Oct 31, 2015
- Messages
- 2,483
Same except I Don't have the money.This thread makes me desire a Threadripper build.
I said it was an standard, I didn't say anything about the amount of users. Yes, there are more users on HPC by two reasons: (i) there are more stuff that can be parallelized, and (ii) 512bit has been available in the Phi line for years. 512bit has only been added to the ordinary Xeon line with Skylake. Despite being novel in the server space, customers are already using it in several real-life cases.
But Juan you cant keep comparing Threadripper and Intel for something like avx512 and then use that as a justfication to purport intels superiority over AMD. Its akin to saying sure a Corvette z07 is a damn quick car but a Formula 1 open wheel is far superior. Yes but two completely different usage scenarios and design purposes.
A home HEDT is N E V E R going to make use much less care about AVX performance except in the most hyper rare circumstances. Its a total moot argument.
A home HEDT is N E V E R going to make use much less care about AVX performance except in the most hyper rare circumstances. Its a total moot argument.
No. I only mentioned AVX512 as a bonus, because most reviews ignore this. AVX512 is not needed to "purport intels superiority over AMD." On ordinary x86 code, Intel is faster and more efficient.
Little I agree. Anyways I went to Microcenter to see if they had any 8700 in the back, I have two friends that work there, they said nope but they have a shit ton of boards. I was going to build an 8700k but probably sometime next week.
OrangeKhrush
[H]ard|Gawd
- Joined
- Dec 15, 2016
- Messages
- 1,673
yea yea efficiency and faster jazz, lemme fetch my Sax.
Monkey God
Mangina Full of Sand
- Joined
- May 7, 2007
- Messages
- 6,723
I wouldn't call machine learning hype rare. People who do that kind of stuff are more likely to have a HEDT than ordinary PC hardware.
You should be lucky they do because the power use for that is insane.
Precisely SIMD is a standard technique to increase efficiency of a core.
OrangeKhrush
[H]ard|Gawd
- Joined
- Dec 15, 2016
- Messages
- 1,673
Per Intels very own worksheet on SIMD array coding Intel supports all proprietary standards of AVX while their competitors and this was highlighted only get baseline support. I don't think AMD CPU's are allowed to go beyond 128bit Array on intels own cooked proprietary standard.
SIMD array in general computing, yay. To be honest, if the 7890XE couldn't win at most things it would be very sad, most of these are not even done on clock vs clock basis anymore, Intel clock policy pushes high clocks to boost scores, since mhz > core count, again during Jayz live side by side his 1700X pushed max 3.6ghz all core turbo while the 8700K was intermittently hitting 4.7ghz over the stated 4.3ghz all core and still only matched the 1700X in stock form.
AMD may not be the fastest, they never needed to be but they have certainly pissed on a few batteries along the way, that with less financial resources and less CPU execution resources, kudos and bravo, and they gave it to the public at a very good price with no compromises on overclocking or not, kudos and bravo.
SIMD array in general computing, yay. To be honest, if the 7890XE couldn't win at most things it would be very sad, most of these are not even done on clock vs clock basis anymore, Intel clock policy pushes high clocks to boost scores, since mhz > core count, again during Jayz live side by side his 1700X pushed max 3.6ghz all core turbo while the 8700K was intermittently hitting 4.7ghz over the stated 4.3ghz all core and still only matched the 1700X in stock form.
AMD may not be the fastest, they never needed to be but they have certainly pissed on a few batteries along the way, that with less financial resources and less CPU execution resources, kudos and bravo, and they gave it to the public at a very good price with no compromises on overclocking or not, kudos and bravo.
Speaking of power, has anyone tested the consumption of the 7980XE on VX512 load while overclocked?
Does the 7980XE exhibit the weird consumption and performance variations as the new 8700X parts?
Most sites have stayed away from doing that. I would love to see it.
What weird variations? Some sites tested 8700k with MCE enabled and others tested with MCE disabled. There is no MCE on Skylake-K parts.