AMD ships B3 Barcelona samples to customers

During its fourth-quarter financial results conference call in January, AMD executives stated that they expected samples of B3 quad-core Opterons to ship in late January or early February, with production chips to follow in the late first quarter or early second quarter. News.com now reports that B3 samples have finally started rolling out, and that AMD’s partners are gearing up for the launch of their Barcelona servers.

The B3 revision of AMD’s quad-core “Barcelona” Opteron CPU will do away with the infamous translation lookaside buffer (TLB) erratum, the BIOS fix for which induces severe performance hits. AMD officially launched B2 quad-core Opterons with the TLB erratum in September, but we later heard that it issued a “stop ship” order and suspended shipments to server vendors and the distribution channel because of the erratum. So far, News.com says, AMD has been selling B2 chips only to high-performance computing customers. Server makers like Dell, HP, and IBM have been waiting until the B3 revision to release quad-core Opteron servers.

With B3 chips now sampling and production chips due out soon, an HP spokesman told News.com, “We expect to start shipping systems in early Q2.” Quad-core Opterons will likely be headed to HP’s ProLiant DL585 server, as documents (PDF) on HP’s website suggest. News.com also got confirmation from Dell that it, too, plans to ship “Barcelona” servers in the second quarter of this year. A spokesman for Dell added, “Look for us to expand our portfolio (of systems) too.”

Comments closed
    • herothezero
    • 12 years ago

    q[< AMD dont need to catch up Intel to compete. Staying close is enough to make a good competition and wait for a proper time to do the overtake. Patience is the key. ^_<<]q When you're financially bleeding to death, exactly how patient should shareholders be? AMD needs to leapfrog Intel again, /[

      • poulpy
      • 12 years ago

      I’m pretty sure he didn’t mean: /[<"wait for a proper time to do the overtake whilst bleeding cash to death"<]/ -which would be a silly strategy- now what do you think? He is right in that AMD doesn't need to overtake the top end of Intel's offering to make money and be a positive competition to Intel for us consumers. Once the swallowing of ATi is complete (which seems almost there now) and the rough times of the 2xxx series on the GPU front (done, the 3xxx are doing quite well AFAIK) and Phenom B2 on the CPU one (almost done, B3 is round the corner and will allow them to sell higher priced parts to the masses), AMD will be profitable again. Some investors may be losing patience but AMD still has -IMO- a huge potential for the years to come, especially as they are in way more than just the PC CPU market. That being said if/when they leapfrog them again this time they should be able to really reap the financial benefits from day one, as last time around people stuck by Intel for ages whilst AMD had the upper hand.

      • LoneWolf15
      • 12 years ago

      /[

    • pogsnet
    • 12 years ago
      • Tarx
      • 12 years ago

      But then Intel can just play the price game and squeeze AMD to sell only low cost CPUs – low/no profit means less R&D and keeps AMD behind the 8 ball…

    • rivieracadman
    • 12 years ago

    #2 Don’t quote me, but as I remember it, the whole reason we didn’t see higher clock speeds on the previous release was, because the TLB error would have a higher chance of occuring above 2.3Ghz. We should see higher clock speeds and some minor tweaks. Since the TLB fix should improve performance overall, along with other fixes, we should see a little performance boost. However, as far as I can tell, other then the 2.4 and 2.6Ghz models that will be comming out, there are no major performance improvements to the core until C1. Which will be 45nm, and be mostly be just a shrink to get costs down, and some changes to the memory controlers. I heard that it’s already in the mix from a couple diffrent sources, so we can expect those sometime around the end of the year.

    I bet the AMD boys have been dreaming about fully programable cores since Sept. Nvidia has a really nice setup in this arena. I bet they’d be willing to lend them for a few bucks on the side too.

    • pikaporeon
    • 12 years ago

    Cant wait.
    While Intel is largely a better value propisition right now, there are still people who’d rather have the extra two cores for as cheap as possible, and getting rid of that performance eater is great news.

    • Fighterpilot
    • 12 years ago

    If the retail Yorkfield quads perform like expected,it’s sayonara for Phenom in the high end user market.
    Ironic given the amount of times”Wait for K8L / K10″ was thrown around here……
    Without 3GHz clockspeeds they stand no chance against 45 nano Penryn Quads.

      • tfp
      • 12 years ago

      I would say that statement is incorrect. For applications that use huge bandwidth there will still be a market for these chips. Also if they are priced right that will help move them as well.

        • Flying Fox
        • 12 years ago

        He basically meant “high end *[

          • MattMojo
          • 12 years ago

          I use a dual proc w/ dual core system (opteron 2220) setup as my high end desktop computer and will most likely be taking my next board from the next lineup of Asus workstation boards…. nothing runs vista x64 like 4 cores and a shitton of ram!

          Ultra high-end desktop PCs use workstation boards and Intel just now (really) enter that market — but it cost an amazing amount to take that path vs. going the AMD path.

          Mojo

      • Krogoth
      • 12 years ago

      Who cares, the older, affordable Kentsfield units already did most of the damage to K10 family.

      Yorksfield is just icing on the cake for Intel.

        • calyth
        • 12 years ago

        Consider looking up prices on FB-DIMMs before you think that Intel’s have an ironclad hold on the high-end desktop market.

        For the same amount of money to get 2GB FB-DIMMS at the cheapest local shop (happens to be NCIX), I could get 4GB DDR2 with change.

        Intel’s FSB is still a major problem for them – there is a very good reason why Intel started the giant L2 cache trend. It’s not just that having lots of L2 is a good thing; eventually the increase of L2 doesn’t decrease the L2 miss rate enough to justify the cost and the power usage. Current Intel chip needs the L2 just to hide the memory latency.

        Phenom/Barcelona is probably not going to perform clock to clock until they’ve got to the C1, 45nm spin, and even that is iffy (disclosure: I’m an AMD fan). But HyperTransport will still make multi-socket that much easier compared to Intel’s offering. They tend to perform better with well-designed multi-threaded apps, which most games are not well multi-threaded. I can’t remember where the benchmark was, but UT3 did well, and Mainconcept H.264 is a benchmark that Phenom has performed well consistently (IIRC performs between Intel Q and QX series).

        Don’t be too quick to write off AMD just because they don’t do well on your favourite game benchmark. They are still quite useful in the server market, and you better hope that they have a comeback. I’m sick of Intel’s price grades – e.g. Q6600 at $279CDN and Q6700 is at $629 for a miniscule frequency difference.

          • Flying Fox
          • 12 years ago

          Krogoth was talking Kentsfield, so what’s all this about FB-DIMMs?

          • Krogoth
          • 12 years ago

          Kentsfield = Single-socket LGA 775, 65nm, quad-core CPU. 😉

          You are referring to Cloverton (dual-socket capable, LGA 771, 65nm quad-core CPU). They are still kinda pricey though (especially the higher-end models).

      • UberGerbil
      • 12 years ago

      Opteron will still own some advantages in the 2S server/workstation market and especially in the 4S server market. No FB-DIMMs means lower total system cost and power consumption; glueless 4S (and potential bandwidth advantages) makes Opteron an especially good price/performance proposition in that market. It’s a niche, of course, but a profitable one (especially for the OEMs).

    • dermutti
    • 12 years ago

    Does B3 offer any other improvements aside from the TLB fix? Or is this revision the exact same as B2 except that one detail? Basically, will this spin be faster/cooler?

      • poulpy
      • 12 years ago

      AFAIK it’s only a TLB bug fix, maybe with some tweaks to allow higher frequencies and/or lower power consumption but no IPC enhancement.

    • Hattig
    • 12 years ago

    Good news for AMD, they’ve had so many issues with this so far. I imagine they’ve lost quite a bit of income because of this. They really should have done the “hacky” dual-die quad-core method back in 2006.

    Wonder if they’ve fixed any other speed issues, or if they’ll still be “trundling” (yeah, it’s still pretty fast!) along at under 2.5GHz.

      • greeny
      • 12 years ago

      if they did the hacky way you would have needed seperate memory for each pair of cores cos of the on-die memory controller

        • UberGerbil
        • 12 years ago

        No. There were 2 socket Opteron boards where all the memory was hanging off one socket. The CPU on that socket acted as the “northbridge” for the other socket. In a hypothetical two-die MCM the two chips would use ccHT to talk to one another in the same way, and the IMC in one of them would connected to memory. It’s a little inelegant, and you get a slight latency hit for the cores that aren’t directly connected, but it’s not a huge deal.

        • lolento
        • 12 years ago

        Right, and they cannot attain the same pin count on this method.

        If AMD can do it this way, I’m sure they would. Instead, they spent a year hyping up monolithic quads…which turned out to be a dud.

          • UberGerbil
          • 12 years ago

          g[

Pin It on Pinterest

Share This