AMD's Ryzen CPUs (Ryzen/TR/Epyc) & Vega/Polaris/Navi GPUs

Discussion in 'Hardware Components and Aftermarket Upgrades' started by Rage Set, Dec 14, 2016.

  1. ajc9988

    ajc9988 Death by a thousand paper cuts

    Reputations:
    1,597
    Messages:
    5,815
    Likes Received:
    8,207
    Trophy Points:
    681
    Are you kidding? AMD has NOTHING competitive on the high end GPUs yet and from what I've seen Nvidia is still the go to for built in decoder/encoder engines. This is a mix in software support for the GPUs and what standards the GPUs support for hardware acceleration. It can do well in resolve, but had an issue of errors which may be resolved by now. But the performance of the 5700XT doesn't do as well as the 2080 Ti or Titan RTX in workloads.

    Hell, software support and less information about the encoder/decoder is why I picked up a 1660 cheap after the super dropped for a Plex hardware accelerator. Not worth waiting for the 5500XT.

    So until RDNA gen 2 comes out next year (which Ampere may be coming out in the spring or could be delayed, I've heard both rumors over the past couple months and pretty sure Nvidia is peddling misinformation to the rumor mill), there is no reason to grab AMD consumer hardware unless buying in at below the RTX 2070 Super price point.

    Speaking of price points, after seeing the 10940X overclocked in HU's review, it is dropped from my recommendation, making the 10980XE the only recommended chip in their lineup, and it being based on memory bandwidth and PCIe needs.
     
  2. TANWare

    TANWare Just This Side of Senile, I think. Super Moderator

    Reputations:
    2,530
    Messages:
    9,535
    Likes Received:
    4,939
    Trophy Points:
    431
    I found a MB for $499.95, but it does no one any good without a CPU to drive it.
     
    ajc9988 likes this.
  3. Deks

    Deks Notebook Prophet

    Reputations:
    1,151
    Messages:
    4,797
    Likes Received:
    1,925
    Trophy Points:
    231
    How to Bypass Matlab’s ‘Cripple AMD CPU’ Function
    https://www.extremetech.com/computi...ss-matlab-cripple-amd-ryzen-threadripper-cpus

    So, its not the issue of whether the software employs coding which discriminates against AMD cpu's... the issue is we are never told which software employs it and whether or not its switched on by default or not, and if it is, that it should have an option to disable 'Cripple AMD CPU' coding.

    That kind of coding shouldn't even exist.
    Software should automatically execute in the best possible manner on any CPU if the said hw has the coding which supports it... if AMD doesn't have AVX512, fine, but don't actively prevent the software from utilising AMD hw to the fullest.

    I wonder if this is happening to a similar degree in the gaming industry...
     
    jaybee83 and hmscott like this.
  4. ajc9988

    ajc9988 Death by a thousand paper cuts

    Reputations:
    1,597
    Messages:
    5,815
    Likes Received:
    8,207
    Trophy Points:
    681
    Couple points:

    1) Intel was sued and AMD won on a similar incident where the Intel compiler purposely gimped AMD performance. Ruled as antitrust violation.

    2) there is bad coding in the world. So not every time this is found is it purposeful.

    3) this needs investigated as to when the code was implemented that made it switched in by default. This can tell us if we are in situation 1, where Intel is getting the software companies to purposely gimp AMD performance, including looking into Intel proposed optimizations if Intel's software development division recommended a specific commit or not, or of we are in situation 2, where a code just forgot to add the logic switch for the hardware flags for AMD.

    Edit: after seeing it is the Intel MKL, Intel is doing shady **** and AMD should look deeper and possibly being legal action.
     
    Last edited: Nov 26, 2019
    hmscott likes this.
  5. TANWare

    TANWare Just This Side of Senile, I think. Super Moderator

    Reputations:
    2,530
    Messages:
    9,535
    Likes Received:
    4,939
    Trophy Points:
    431
    Well maybe a clue, from BH Photo web site, expected date 12/6/2019?
    3990x.jpg
     
    Last edited: Nov 26, 2019
    hmscott likes this.
  6. ole!!!

    ole!!! Notebook Prophet

    Reputations:
    2,385
    Messages:
    5,798
    Likes Received:
    3,716
    Trophy Points:
    431
    32 core cpu is using 8 core chiplets x4 right? so its almost as efficient as the 24 core cpu due to better silicon similar to 3950x?
     
    hmscott likes this.
  7. TANWare

    TANWare Just This Side of Senile, I think. Super Moderator

    Reputations:
    2,530
    Messages:
    9,535
    Likes Received:
    4,939
    Trophy Points:
    431
    3970 is slightly less efficient. The reason is both share the same L3 cache size so more cache per core for the 3960x.
     
    hmscott likes this.
  8. Deks

    Deks Notebook Prophet

    Reputations:
    1,151
    Messages:
    4,797
    Likes Received:
    1,925
    Trophy Points:
    231
    1.) Intel was sued yes, but the final result (aka the payment AMD got) was rather paltry/anemic in comparison to the damage Intel did (namely, AMD got a relatively small sum of money - and many people to this day haven't even heard of AMD (I just recently assembled an all AMD gaming machine for a friend's son, and she had no idea AMD as a company existed, or that it would be a better option for gaming - especially in regards to saving money and upgrades down the line).

    2.) That should be addressed at the source and standards need to be in place to prevent software from favouring one hw over the other. If a hw doesn't have instruction sets built in like AVX512, fine, but as I said (and continue to maintain), that shouldn't prevent hw from running at its full potential.
    Software in general should be open-source based as its a lot more flexible (while being just as capable) than closed-source and ensures it will work across all platforms.

    3.) Given how long Intel held the 'performance crown'/market dominance, the company had time to permeate virtually every facet of the market to implement this coding to one degree or another with software devs (even in professional software, developers primarily write code with Intel hw in mind, not AMD).

    Looking at games, we saw that when patches were released for Ryzen, performance jumped... suggesting that certain games may have been ported/coded with Intel CPU's in mind from the get go.
     
    Last edited: Nov 29, 2019
    hmscott likes this.
  9. rlk

    rlk Notebook Evangelist

    Reputations:
    132
    Messages:
    560
    Likes Received:
    292
    Trophy Points:
    76
    Toys :)

    I cut my computing teeth on the Connection Machine CM-2 (which was, depending upon how you looked at it, 64K single bit processors or or 2K 32/64 bit processors) and at one point probably had the fastest FFT, something like 13 GFlops single precision or 7 double. Of course, any halfway respectable phone blows that out of the water these days, but 3 decades will do that. More recently, when I was at Oracle, I worked on the M7-16. Think 16 Epyc 7742's in a box, except that instead of 2 hyperthreads/core there were 8. Then think about that little beastlet running a single system image.

    A TR would make a pretty nice CI/CD system for hosting a whole stack of projects. The weakness of the 3950X for that kind of work is the low memory ceiling; 64GB, even 128GB, is going to get tight for really big integration jobs; the CPU horsepower is way OTT (this is why I'd like to see a 3955X, with the HEDT memory and I/O capacity).

    More often than not in the industry, compute power has outstripped I/O. There have been a few exceptions; the VAX had pretty good I/O but not amazingly impressive compute power as I recall, and what really distinguished mainframes was the I/O capacity.
     
  10. rlk

    rlk Notebook Evangelist

    Reputations:
    132
    Messages:
    560
    Likes Received:
    292
    Trophy Points:
    76
    Think rendering farm in a box, not just single project. Or large scale software development. With that much horsepower, you could really ramp up testing, for instance.
     
Loading...

Share This Page