It's not a surprise, but nevertheless: AMD's next APU will feature a GPU branded the AMD Radeon HD 7000 series, as indicated by a GlobalFoundries slide shown in a presentation today.

The slide specifically references the upcoming Trinity APUs, 32nm parts that are slated to replace current Llano A-series APUs when they're released in 2012. AMD says that Trinity APUs should outperform Llano APUs by about 50 percent (at least, when measured in teraflops). 

Like the existing Radeon HD 6000 series, the 7000 series will eventually include both on-board APU graphics and more traditional dedicated graphics cards - these dedicated GPUs are currently codenamed Southern Islands, and are expected to be manufactured on a 28nm process.

Source: Fudzilla



View All Comments

  • Soulkeeper - Tuesday, August 30, 2011 - link

    Any word on if these will be FM1 compatible ?

    or will they be transitioning to ddr4 for these in a new platform ?
  • stmok - Tuesday, August 30, 2011 - link

    They will use Socket FM2. Both Komodo and Trinity will use the same CPU socket format. Their differences will be the chipsets and memory controllers for their specific markets. (Performance/Enthusiast and Mainstream respectively).

    Trinity will re-use Llano's (current A-series) chipsets with minor tweaks and new features.

    DDR4 is a few years away from being mass produced and affordable. DDR3 will still be used in AMD's 2012 lines.

    The 50% increase in gigaflop performance; they're talking about GPGPU performance.
  • JarredWalton - Tuesday, August 30, 2011 - link

    So that would make the next-generation Fusion GPU approximately at the level of HD 6700 I guess. That's a lot faster than Intel's current HD 3000, but I'm still waiting to get at least 800 shader cores in an APU. Reply
  • DanNeely - Wednesday, August 31, 2011 - link

    I suspect memory bottlenecking would make that many shaders futile. The current 400 shader parts fell behind their dedicated card equivalents. The 4850 has ~2x the bandwidth of dual channel DDR3-2133, the 5770/6770 2.5x, and the 4870 4x. Reply
  • iamezza - Tuesday, August 30, 2011 - link

    The current gen is already memory bandwidth limited so if it is 50% faster it will need 50% more bandwidth, which means tri-channel DDR3 1600.
    AMD could probably put something on there 4x faster if they wanted but where would the memory bandwidth come from?
  • MonkeyPaw - Tuesday, August 30, 2011 - link

    I'm thinking quad channel memory. Even then, it is a 256bit memory bus sharing time with the CPU. If the cache is shared in the APU, or we get a ring bus, then I think it will be quite potent. Reply
  • DanNeely - Wednesday, August 31, 2011 - link

    Nope, too expensive to implement in hardware due to the extra space needed on the mobo; it's almost certainly still going to be a dual channel design. The 50% GPGpu performance gain is almost certainly limited to apps that have low memory demands; graphics performance gains will be smaller because the bottle neck will bite harder. Reply
  • Arnulf - Wednesday, August 31, 2011 - link

    Or, this could infact be a result of more efficient GPU data caching (along the lines of what Intel did in SB), requiring less memory bandwidth. Reply
  • rocketbuddha - Tuesday, August 30, 2011 - link

    Per the Slide
    The CPU part of Trinity is "Piledriver"
    The GPU part of Trinity is "Radeon HD7000"

    The APU is still TRINITY only.
  • jfelano - Wednesday, August 31, 2011 - link

    Wow, this is news, so your saying after the HD6000 series comes the HD7000 series???

    OMG, you guys delivered again Anandtech, how do you guys find out such reclusive information??


Log in

Don't have an account? Sign up now