Announcement

Collapse
No announcement yet.

AMD Announces Radeon Pro W5700 RDNA Workstation Graphics Card

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • aht0
    replied
    If wishes were horses, beggars would ride

    Leave a comment:


  • evilryry
    replied
    I really wish AMD would roll out a low power GPU for workstation use or put graphics on their high end desktop CPUs. I'd like lots of cores for dev work but I don't want a card that is expensive, beefy, old, or doesn't have solid free drivers.

    Leave a comment:


  • ddriver
    replied
    The memory errors issue is only exacerbated as the process of chip manufacturing shrinks down.

    The numbers that circulate around date back to a time when process was much less susceptible to random bit flips.

    Smaller cells and higher cell density will in all likelihood make things worse, not only in the frequency but also the severity of memory errors.

    Leave a comment:


  • cb88
    replied
    Something like a couple a day, but when you are doing scientific compute work and the error is cumulative on your multi day job...well you get the idea. It was a huge deal and even made the news awhile back or at least EE and tech news.

    Leave a comment:


  • tildearrow
    replied
    Originally posted by cb88 View Post
    some "workstation ish" cards like the Titan cards usually lack ECC, which bit a bunch of HPC clusters that bought them as they couldn't even complete any work without errors...
    Wait, what's the frequency of a bit error on a non-ECC graphics card?
    To be honest I didn't think it was that frequent...

    Leave a comment:


  • cb88
    replied
    Originally posted by wizard69 View Post

    It doesn’t say one way or the way other on their web site. It just says GDDR6 which could implement EEC.
    Yes it does. It says no ECC. Also it says this for many other of AMD's WX series cards, if you want a card with ECC you need to pony up for it and get one of the cards that lists support.

    Frankly for 90% of workstation use cases ECC does not matter. And for the 10% where it might they are already getting a high end card.

    The situation is the same on NVidia worksation cards you don't get ECC up untill around the P5000 card, and even some "workstation ish" cards like the Titan cards usually lack ECC, which bit a bunch of HPC clusters that bought them as they couldn't even complete any work without errors...

    Leave a comment:


  • wizard69
    replied
    Originally posted by tildearrow View Post
    A workstation card...
    ....without ECC?
    It doesn’t say one way or the way other on their web site. It just says GDDR6 which could implement EEC.

    Leave a comment:


  • tildearrow
    replied
    A workstation card...
    ....without ECC?

    Leave a comment:


  • L_A_G
    replied
    Originally posted by cb88 View Post
    RDNA has lower peak compute efficiency... but higher efficiency for graphics. Ofthen the hybrid graphics/compute pipeline of Vega would go underutilized Navi mitigates the overhead of these bubbles in the pipeline.
    I will admit that AMD has muddled things quite a bit with versions of their actual architecture being called different "architectures" and not really explaining what RDNA really is.

    AMD has been using their GCN architecture for quite a while as the basis for their graphics "architectures" (Vega, Polaris, etc.) and RNDA is actually a new architecture, but Navi isn't even a full implementation of it. It's actually a halfwaypoint between GCN and RDNA as it uses a bunch of carried over IP blocks from Vega. A full RDNA "architecture" until AMD's yet-to-be-named one coming at some point next year.

    The reason why Navi is more efficient in graphics tasks is in part down to being manufactured on a newer manufacturing node and in part with the raster-geometry-GPGPU-setup being more optimal for graphics tasks. With fewer pipeline stalls caused by the raster hardware simply not being able to keep up you can either perform your job faster or run at a lower, more efficient, frequency. The thing about efficiency and frequency is that as you get closer and closer to the maximum allowed by your manufacturing node, the worse your efficiency will get.

    Leave a comment:


  • cb88
    replied
    Originally posted by L_A_G View Post

    These kinds of GPUs are so general purpose there is no such thing as something designed from scratch for a specific use. Only thing you can do to specialize is very the amount of dedicated raster and geometry hardware in relation to the general purpose GPU cores, but that's it. You're not completely on the wrong track as Navi supposedly has more dedicated raster hardware in relation to the general purpose GPU cores than Vega. However rDNA is the name of the overall architecture, including the general purpose GPU cores, which are the exact same ones across the entire range.
    RDNA has lower peak compute efficiency... but higher efficiency for graphics. Ofthen the hybrid graphics/compute pipeline of Vega would go underutilized Navi mitigates the overhead of these bubbles in the pipeline.

    Leave a comment:

Working...
X