AMD Radeon HD 8790M: Next-Gen Mobile Mainstream Graphics Preview

General Purpose Compute

One of the most attractive qualities of AMD's GCN architecture is its alacrity in compute-optimized applications, particularly now that we've seen Nvidia go the other direction with its Kepler design. The Mars- and Venus-based Radeon HD 8000M-based parts employ the same GCN pedigree, and AMD is adamant that they share similar advantages general-purpose applications able to exploit OpenCL.

So, our plan was to run a handful of synthetic tests, WinZip 17, our OpenCL-enabled Photoshop CS6 benchmark, and the Wireless Security Auditor. Testing hit a snare when we discovered certain applications returned OpenCL initialization errors. More than likely, this is a driver issue, and although AMD wasn't able to book the lab time to confirm, this is a preview, after all. Naturally, we expect these tests to run by the time Radeon HD 8000M-series hardware ships.

Although the real-world tests wouldn't complete for us, the synthetics did, strangely enough. In both LuxMark and CLBenchmark, our Core i5-2500K managed to turn back better benchmark results than the Radeon HD 7670M. But the 8790M blows them both out of the water. Particularly on mainstream 15" notebooks with much less potent CPUs than our 95 W Core i5, that could translate to a significant speed-up in OpenCL-optimized applications.

  • shikamaru31789
    Wow, can't say I expected to see any 8000 series/700 series benchmarks for at least a few months. Looks prettyimpressive for a mobile GPU, can't wait to see what the high end mobile and desktop cards can do.
    Reply
  • A Bad Day
    Sometimes I wonder why instead of CFing or SLIing two high end mobile GPUs, a desktop GPU would've achieved the same performance but without the stuttering or driver issues.

    And if thermal management is an issue, then the desktop GPUs could always be undervolted (but of course more expensive because of the extra step).
    Reply
  • A Bad Day
    Also, I'm interested to see if the MxM cards of the 8000s are available for retail purchase. I'd like to buy a 15" laptop that supports MxM and is on sale, and swap out the weak GPU.
    Reply
  • Novuake
    Second review I have seen that uses this module method. Interesting way of doing it. Thanks Toms and AMD.
    Reply
  • mayankleoboy1
    Nice preview, Aku and AMD.

    Though i suspect you and Chris already have large 'hints' about the HD8000 series performance, but under NDA.
    Reply
  • SteelCity1981
    looks like the value brand hit the end of the line with the 7000m series and id expect that with the APU's becoming the standard now for value graphics.
    Reply
  • acku
    mayankleoboy1Nice preview, Aku and AMD.Though i suspect you and Chris already have large 'hints' about the HD8000 series performance, but under NDA.
    In the words of Sgt. Schultz "I know nothing." =_=

    Cheers,
    Andrew Ku
    Tom's Hardware
    Reply
  • amuffin
    ackuIn the words of Sgt. Schultz "I know nothing." =_=Cheers,Andrew KuTom's Hardware
    That's what they all say! ;)
    Reply
  • Robert Pankiw
    A Bad DaySometimes I wonder why instead of CFing or SLIing two high end mobile GPUs, a desktop GPU would've achieved the same performance but without the stuttering or driver issues. And if thermal management is an issue, then the desktop GPUs could always be undervolted (but of course more expensive because of the extra step).
    If you look, there is only one chip (shown on this page) which means it is not being CrossFired. I agree that 8780M would be a better name than 8790M. Andrew Ku, maybe on the front page you can clarify this?

    About using desktop parts, it is my understanding that they sometimes do exactly that. Take the 7970M, which as far as I can tell, is an 78XX part (I forget which one) except the mobile chip has MUCH higher binning than the desktop 78XX.
    Reply
  • A Bad Day
    If the 7970M uses two 78XX, why not use a highly binned and undervolted 79XX? Again, better driver stability, more supports from games, more consistent performance scaling, and less micro-stuttering.
    Reply