Yesterday AMD announced that it had begun revenue shipments of the first Bulldozer processors for servers. More specifically AMD is referring to its Interlagos CPU which features two Bulldozer die on a single multi-chip-module package. A single Interlagos CPU features 16 integer cores (8 FP cores) and will work in existing AMD Opteron 6100 series platforms. AMD mentioned that the production version of Interlagos was ready in August, the past weeks have simply been used for testing and validation. 

What about the Socket-AM3+ version of Bulldozer? AMD mentioned that it will be shipping Zambezi processors before the end of September and is "tracking to an expected Q4 launch and availability for those parts". Why the lag between server and desktop parts? A look at Bulldozer's architecture reveals that it's clearly aimed at servers - addressing that market first makes the most sense. Secondly, AMD's Opteron parts have traditionally run at much lower frequencies than their desktop counterparts. It's always easier to push out a lower clocked part than to push the envelope of the design and process. Finally, it looks like the desktop part went through one more stepping (perhaps related to clock speed?) than Interlagos which pushed back its launch.

With AMD saying that Zambezi will ship in September but launch in Q4 it's safe to say that rumors of a September launch date aren't true. Q4 encompasses October, November and December. My money is on an early-to-mid Q4 launch.

Source: AMD

Comments Locked

22 Comments

View All Comments

  • yankeeDDL - Thursday, September 8, 2011 - link

    Well said. Anyone knows when the first benchmarks will come out? I guess Anand and others already have them in house but are bound by NDA?
  • fic2 - Thursday, September 8, 2011 - link

    Agree. I am pretty close to saying f*ck it and buying a Sandy Bridge. AMD just keeps delaying and delaying. Seems like the next announcement out of AMD will be a Q1 2012 release, then a Q2, then... By the time Bulldozer is out Ivy Bridge will be either out or just around the corner.

    Some benchies might alleviate the feeling that I should have just bought an SB months ago.
  • BaronMatrix - Thursday, September 8, 2011 - link

    If you consider them to be 256bit there are 8, but 95% of code will be 128bit FP.
  • cfaalm - Thursday, September 8, 2011 - link

    +1 on that. So many people overlook that feature. It's one of the reasons the cores are built like this. To be able to combine 2 x 128 in to 1 x 256 and be AVX capable.
  • BSMonitor - Thursday, September 8, 2011 - link

    BaronMatrix is alive? Still waiting on the TRUE quad-core from AMD to wipe the floor with the Intel dual die quads ehh?

    Still hoping for that told-ya-so??

    If these things were any good, we'd have benchmarks flying around everywhere.

    Instead, crickets..
  • WeaselITB - Thursday, September 8, 2011 - link

    Or, maybe, because they're all still under NDA?

    Wait ... logic has no place in an irresponsible flame war. My mistake. Carry on.

    -Weasel
  • BSMonitor - Friday, September 9, 2011 - link

    Intel had no problem showing off Conroe, Penryn, Nehelem, Sandy Bridge... If one wanted to generate buzz and excitement about a new product, why hide its performance??
  • silverblue - Thursday, September 8, 2011 - link

    http://en.wikipedia.org/wiki/Troll_%28Internet%29
  • silverblue - Thursday, September 8, 2011 - link

    It's confusing watching John Fruehe talk about Interlagos in comparison with "8 core Sandy Bridge" but he's got a point; Sandy Bridge's FPUs can only do 1x128 OR 1x256 in the same cycle, which is a sizeable disadvantage with today's workloads. Bravo to Intel for being very forward thinking with AVX but for the first time in a while, they stand to be behind in instruction set support at the very least. As for performance, that's not something most of us know anything about yet, though Sandy Bridge will definitely be ahead in some areas, you can count on that.
  • yankeeDDL - Friday, September 9, 2011 - link

    Indeed.
    The fundamental difference in architecture will make it difficult to run proper benchmarks and to do fair comparisons. Synthetic results could come up completely skewed.

Log in

Don't have an account? Sign up now