The Bulldozer architecture within in AMD's current FX series CPUs, has generated no end of controversy (see pp. 1-123) since its release late 2011. For one sister site Atomic's own lab testing saw the top model FX-8150 come up short in a large number of areas. How then, will Bulldozer's successors fare?
In the first update for AM3+ systems, second generation processors (codenamed Piledriver) are very much inbound for desktops this year, and a few tweaks bring a welcome 10% performance bump and much lower power draw. In that vein we should see desktop A-Series branded (CPU + graphics parts) over the next few weeks and FX-Series (enthusiast CPU only processors) later in 2012.
Even more interesting is just how much next-year's further evolved 'Steamroller' processor (expected to also be compatible with AM3+) will shake things up. Bulldozer had the disctinction of being the first completely new x86 (PC) chip design since the VIA Cyrix III circa 2000 and chipmaking is a hugely complex task. Five year gestation or not, this first of a breed ended up rushed and has significant potential for optimisation - AMD have given us their story on 'how' and 'how much faster' Steamroller will be.
Where the second-gen Piledriver concentrates primarily on adding 'Resonant Clock Mesh' technology to its Bulldozer predecessor, the third gen Steamroller will be a major reworking of the underlying design that debuted in Bulldozer. Previous AMD roadmaps targeted a 10% per generation improvement in performance (which Piledriver met) however with the design being more fleshed out, AMD now imply a 30% performance increase. This is a pretty significant boost if true and more than a few sources suggest it is - then again 'more than a few' sources suggested Bulldozer would be good.
Chief among the design improvements is doubling the number of pre-processing 'decode' modules, bringing this part of the chip back to the traditional one-per-core level. Given the amount of time Bulldozer and Piledriver spend with inactive processing areas (indicating a design flaw in the pre-processing stage) this is a significant and straightforward win. The processing centres (Floating Point and to a lesser extent Integer units) also receive significant changes, as do the Prediction unit and Cache:
Steamroller was the subject of the recent keynote speech by AMD Chief Technology Officer Mark Papermaster, along with a fair bit of marketing speak on a 'Surround Computing Era'. Nonetheless along with a move to a slightly more advanced 28nm 'Bulk
' (from 32nm 'SOI
') manufacturing process, the significant redesigns outlined in Steamroller do hold the potential to yield equally significant performance gains. This in turn may close the gap with Intel, whose fourth-generation 'Haswell' architecture (due May-June 2013) is set to give roughly a 10-20% over its third generation 'Ivy Bridge' designs. We should see more concrete information from Intel on Haswell over the next week at its flagship IDF conference, which starts September 11th.
All in all it does seem certain AMD will deliver a significant performance jump with Steamroller, exactly how significant this will turn out to be and when the chips will reach shelves is not as clear yet. We'll keep you up to date as the details come into focus.