Announcement

Collapse
No announcement yet.

AMD Publishes Open-Source Linux HSA Kernel Driver

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

  • ObiWan
    replied
    The upper (and higher) are turbo clocks,
    the lower ones are the standard non turbo clock.

    Leave a comment:


  • bridgman
    replied
    There are some published specs - click on "specs" tab at :



    Doesn't seem to say whether clock freqs are turbo or not, but from the numbers I imagine they are. The A10-7800 and A8-7600 are both there, guessing that fabbed parts yielded more A10-7800s than A8-7600s.

    Leave a comment:


  • edgar_wibeau
    replied
    When Kaveri (Desktop) was first presented in January, A8-7600 was amongst the models presented, though only the two flavors of A10 came to market until now. But they prsented Specs for A8-7600 when running as default 65W config, as well as for the 45W configuralble TDP, see the first slide here.

    The numbers are for 65 / 45 Watts:
    Default CPU freq: 3.3 / 3.1 GHz
    Max Turbo Core: 3.8 / 3.3 GHz
    GPU Frequency: 720 / 720 MHz
    CPU Cores: 4 / 4
    GPU Cores: 6 / 6 (384 shaders)

    So there were defined frequencies for the 45W cTDP published, for A10-7800 though, they didn't publish these numbers. But at least they (the frequencies) can be expected to be a little higher than those of A8-7600. GPU and CPU cores are the same as for 65W config (4 CPU, 8 GPU aka 512 shaders) of course. The nice part is that the GPU doesn't seem to need downclocking though in paractice/average it might run al little slower then in 65W mode.

    Leave a comment:


  • bridgman
    replied
    Originally posted by molecule-eye View Post
    You're thinking of the A8-7600 which is available NOWHERE at the moment. We should see it by the end of the year, meaning it was a total paper release. Sad.
    At first glance it appears that we might have been able to meet the original A8-7600 power levels without having to fuse off any GPU cores and so decided to sell the part as an A10-7800 instead of A8-7600. That is not an official statement, just my impression of what probably happened.

    Leave a comment:


  • dungeon
    replied
    As i see they have it in hands in Japan and even advertised cTDP of 45W .


    Leave a comment:


  • dungeon
    replied
    Originally posted by molecule-eye View Post
    You're thinking of the A8-7600 which is available NOWHERE at the moment. We should see it by the end of the year, meaning it was a total paper release. Sad.
    No i am thinking of what i ask . After reading this announce 9 days ago:

    The AMD A10-7800 APU will be available for purchase in Japan starting today, with worldwide availability at the end of July.

    Leave a comment:


  • molecule-eye
    replied
    Originally posted by dungeon View Post
    Is that new A10-7800 APU on the picture - 65W/45W part? Like lower TDP on that one, 512 shaders with 45W sounds cool .
    You're thinking of the A8-7600 which is available NOWHERE at the moment. We should see it by the end of the year, meaning it was a total paper release. Sad.

    Leave a comment:


  • agd5f
    replied
    Originally posted by chithanh View Post
    Last I checked, AMD no longer lists Kabini as HSA compatible[1]. So I would avoid buying Kabini for HSA. The Kabini/Temash successors Beema/Mullins however seem to support HSA in some form.
    FUll HSA support requires an APU/CPU/chipset with an iommu that supports ats/pasid. Kabini does not have a compatible iommu.

    Leave a comment:


  • liam
    replied
    Originally posted by bridgman View Post
    Yeah, the nice thing is that you don't have to transfer betwen them at all -- the GPU (or DSP etc..) runs in the same demand-paged virtual address space as the CPU so they can share data structures without much in the way of special programming.
    Can the GPU be used as an io accelerator? That'd be a nice, cheap alternative to the pricey alternatives offered by the IBM/oracle that let you do io offloading?
    This is a bit of a detail but do you know if the HSA patches make extensive use of lockless memory access? I can just imagine much of the memory pool being held by, say, mutexes from the huge numbers of additional threads the GPU introduces😀

    Leave a comment:


  • curaga
    replied
    Originally posted by Bucic View Post
    As a general rule I disregard less than 3 y.o. technologies that require application developers input. For obvious practical reasons. So I'm interested only in benchmarks using non-HSA enabled software. Or software using lower level components that benefit from HSA >today<.
    The jpeg code is (well not merged yet I believe) in libjpeg-turbo. That means all jpeg-using applications gain without recompiles.

    Leave a comment:

Working...
X