• Road to Southern Islands: AMD has working 28nm-silicon in house, Interlagos 35% faster than Magny Cours

    AMD Company LogoAMD has published their quarterly earnings for fiskal Q2/2011 on July 21st and industry as well as analyst uptake seems to be quite positive on their performance and expectations for the future. One of the contributing factors next to the pretty successful line of APUs seems AMD's early transistion to 28 nm production for their GPUs - probably marketed as Radeon HD 7000 series - at TSMC's foundries. But with new Bulldozer performance figures, this respective outlook is not as great as it seemed earlier this year for AMD's new micro architecture still on schedule to be shipping in the third quarter 2011.
    In seekingalpha's transcript of the quarterly earnings call transcript, AMD's interim CEO Thomas Seifert states the following about AMD and their progress on TMSC's 28 nm process technology:
    We also passed several critical milestones in the second quarter as we prepare our next-generation 28-nanometer graphics family. We have working silicon in-house and remain on track to deliver the first members of what we expect will be another industry-leading GPU family to market later this year. We expect to be at the forefront of the GPU industry's transition to 28-nanometer.
    What does not mention though, is whether or not those 28 nm samples belong to AMD's recently announced architecture called „Graphics Core Next” or GCN for short. To lessen risks commonly associated with both the transition to a new process technology as well as a new micro architecture, it would probably make sense to have a pipe-cleaner similar to what RV740 was for 40 nm process. Cayman-GPU's, initially planned for the cancelled 32 nm process at TSMC, do still have one ID left unused…

    On a slightly less positive notion, AMD's upcoming CPU architecture code-named Bulldozer being characterized as optimized for today's cloud datacenter and excelling at compute-intensive and HPC workloads by interim CEO Seifert, is seen by AMD as bringing a performance benefit of up to 35 percent to the table when comparing the Interlagos SKU with 2 x 4 Bulldozer-modules against the current 12-core Magny Cours offering from AMD. This, obviously, leaves room for speculation as to the relative performance of the Bulldozer architecture. Remember, each Bulldozer chip is marketed as having eight cores, which is true only for the Integer pipelines, as their heavy-lifting parts (FPU, SSE and AVX) are shared between each core inside a module.

    So, depending on your workload, this could mean comparing 16 Interlagos style and 12 Magny Cours style cores or just 8 Bulldozer FPU/SSE/AVX execution units against 12 equivalents in the current line up of processors. From only 35 percent performance increase for 33 percent more cores up to 135 % performance with only 66% of the core count everyhting's possible - theoretically. In quarterly earnings calls, you usually present an at least moderately optimistic view of your business prospects. AMDs John Fruehe, director of marketing for server/workstation products, however, was giving an estimated figure of up to 50 percent performance benefit earlier this year (via) - but maybe it was just the basis that has changed since then as the Opteron 6180 SE with 2,5 was released on february 14th. Something which John Fruehe of course would have known when he wrote his blog cited above on february 8th and which was probably the reason he did it exactly when he did it…