Poll: Will Bulldozer deliver?

In case you missed it yesterday, AMD revealed a great many details about the new Bulldozer architecture that will underpin the company’s next generation of desktop and server CPUs. This bold new design appears to have loads of potential, but then so did Barcelona, and we all know how that panned out. AMD still doesn’t have a desktop CPU fast enough to sell for more than $300.

Does AMD have a better chance with Bulldozer, especially with Intel’s Sandy Bridge due out in early 2011? That’s the subject of this week’s poll, which asks: Will Bulldozer deliver? You can cast your vote over on the right column on the front page or after clicking on the comment link below.

In last week’s poll, we inquired about just how much storage sits inside your primary PC. A lot of folks have yet to crack the terabyte threshold, with 16% packing less than 500GB and 22% between 500GB and just under 1TB. 19% have 1-1.49TB at their disposal, while 12% fill in the gap between 1.5 and 1.9TB. Of those who voted, 14% are in 2TB territory, and another 6% have three-point-something terabytes of storage capacity. Impressively, 11% have 4TB or more in their primary system.

Comments closed
    • swaaye
    • 9 years ago

    I just hope that Bulldozer actually works out in general. It’s the first really new thing that they’ve done since K7! This is risky stuff. The first generation of a new architecture is frequently not that great.

      • indeego
      • 9 years ago

      core2 blew everything awayg{<.<}g

        • swaaye
        • 9 years ago

        I knew somebody would say that.

        Core 2 was not exactly new, you know. We had Core Duo, Pentium M, etc. It was just evolutionary with lots of neat enhancements.

        Bulldozer isn’t just yet-another-K7-tweak. Phenom 1 blew up in their faces pretty bad even though it was just that.

        But I do hope it’s amazing so AMD can have their own $1000 consumer-level CPUs again.

          • RtFusion
          • 9 years ago

          I’d like for AMD to bring back the FX brand.

    • ronch
    • 9 years ago

    This poll is quite useless, and only serves to set the stage for academic discussions. It doesn’t matter what people think now. The die has been cast. AMD can’t redesign Bulldozer again if it does end up underperforming anyway. Next year, we’ll have all the answers.

    By the way, does anybody know where Jerry Sanders is now? I kinda miss the guy.

      • TaBoVilla
      • 9 years ago

      the 1990s called, said “leave jerry sanders alone”

    • Derfer
    • 9 years ago

    After seeing the poll results so far I have to either assume a large portion of people here are either new to the pc tech world or have short memories. This will match nehalem for less money and less power, but not beat it out right. Nothing more.

      • grantmeaname
      • 9 years ago

      What does that have to do with memory? Absolutely nothing.

    • dragmor
    • 9 years ago

    Dont really care about Bulldozer, AMDs short – mid term future profit (and my personal focus) is Bobcat.

    My 6 year old 3500+ is more than fast enough for what i do, a Bobcat APU will probably be an upgrade.

    • Anonymous Coward
    • 9 years ago

    In many ways, AMD is busy implementing the design features Intel has already implemented. Its hard for me to see how they would go beyond Intel in very many areas. This new super-hyper-threaded design will doubtless perform well at full utilization (lots of threads) but I can’t see that it will give them an edge vs Nehalem on single threads. I expect the flexibility of hyperthreading on Nehalem to be an advantage over AMD’s partitioned approach unless there are a lot of threads in play.

    On a similar note, I doubt that Sandy Bridge will be too impressive compared to Nehalem either, until someone compiles a program for AVX.

    • bogbox
    • 9 years ago

    AMD said:
    2 integer cores = 80% real cores= -20% Performance
    Architecture improvements = 50% Performance
    Bottom line = 30% Performance gain.
    So best case matches,beats Nahalem.
    But will be price much better than SB or Nehalem .
    AMD will target small and efficient ,just like ATI ,and will probably put a 2 cpu in package to match much more powerful intel cpu .(like 8 cores vs 6 cores )
    If someone finds something wrong in this theory please tell.

      • Voldenuit
      • 9 years ago

      The 80% performance is compared to a dualcore. But it takes only 12% more space on the die, vs 100% more (neglecting cache) for a traditional dualcore CPU. So performance per square mm of Bulldozer is going to be a lot higher than it was for K10, even before we factor in the 30% improvement from the new architecture.

      I think Bulldozer is going to be very strong.

      EDIT: Further calculations. Per square mm, Bulldozer is going to have 1.78x the number of cores that a traditional dual-core design will have. With an 80% efficiency, that means that you have 1.43x the effective computing power. Multiply that by 30% architectural improvements and you get an 85% improvement per die. Now each bulldozer module is probably larger than an equivalent K10 module (size still unknown), so we’ll get less than a 1.85x increase (divide by however larger BD is vs K10), but the gains sound like they should be substantial. Add in improved Turbo Core for lightly threaded tasks, and this thing should slap K10 silly, and pose quite a challenge for Nehalem/SB.

        • TravelMug
        • 9 years ago

        “The 80% performance is compared to a dualcore.”

        This is getting confusing. Dualcore what? MC? What are we actually comparing? A dualcore current Opteron style would be 2×3 Int execution units. What BD will be 80% faster? The 1 module with 2×2 Int units or 2 modules with 2x2x2 units?

          • Anonymous Coward
          • 9 years ago

          I think they are saying that a new “one module dual core” is 80% as fast as an old “true dual core”, so two modules (four cores) of the new design would be perhaps 60% faster than two old cores, for a lot less than an 60% increase in die area and power use. Should work out pretty well.

          • Peldor
          • 9 years ago

          In case you can’t tell yet, there are a lot of buyers for pre-release hype, again. We’ve even got people throwing out % this, % that divided by % the other adds up to %awesome. Really?

          Personally, I voted No. Bulldozer was initiated as a Nehalem competitor, and obviously the project didn’t go all that well as it’s coming in 2011 instead of 2009. Projects don’t often get delayed for 1+ years and become even more awesome as a result.

      • Joel H.
      • 9 years ago

      The problem is that you assume AMD told the truth and gave accurate representations of performance/performance increases.

        • Voldenuit
        • 9 years ago

        Even if they were accurate, one would assume that AMD would quote “best case” scenarios.

        Real world performance will be a bit off from that.

        But I still think ‘dozer has potential, even (or especially) after Anand’s SB preview.

    • TravelMug
    • 9 years ago

    I wonder what the belief that it will match/surpass Sandy Bridge is based on. The current published integer workload results do not really suggest that.

    The top 12core/2socket/24thread Nehalem SPECInt_rate2006 result is 382/356 (peak/base). The top 24core/2socket/24thread MC result is 402/312.

    In total computation power counting the physical threads only MC has around 40% advantage. In scores it has 5% advantage in peak and 14% disadvantage in base. So the SMT implementation in nehalem gives it the required boost to match/surpass the 40% disadvantage it has compared to MC.

    • maroon1
    • 9 years ago

    Probably, only in multithreaded applications, and not by much.

    However, I expect higher end sandy-bridge (LGA 2011) to beat bulldozer in both single-threaded and multi-threaded tasks

    • maroon1
    • 9 years ago

    All those who voted to choice number 1 & 2 are dreaming.

    • NIKOLAS
    • 9 years ago

    I think AMD will at least match Sandy Bridge and probably slightly exceed it.

    However I expect Intel will move ahead when they move to 22nm(probably within 6 months of Bulldozer’s release), and isn’t Intel’s next big improvement in architecture meant to come with Haswell due in Q1 2012?

    So AMD’s moment in the sun may well be quite brief.

    • MadManOriginal
    • 9 years ago

    I am split between ‘No, just matches Nehalem’ and ‘Kinda, superior to Nehalem’ but voted for the former. Previews and technical possibilities are one thing but I’ve seen too much computer tech come and go that has great possibilities but doesn’t quite live up to them in real world use. Therefore I’m going to play it conservative although I can see it being better than Nehalem I don’t see it being better than SB from an absolute standpoint. It may be better or acceptable on a price/performance scale especially since we are in the ‘fast enough’ era of hardware versus software for the most part.

    What’s much more intriguing to me is how the true Fusion cores work out. One chip with a good IGP ondie with all the advantages that brings will be a definite step forward. Of course Sandy Bridge falls in to that area as well so it will be quite the showdown.

    • not@home
    • 9 years ago

    I voted it will beat sandy bridge and here is why:

    I read somewhere that Intel’s 32nm and AMD/IBM’s 32nm high K metal gate processes have extremely close to the same switching speed (looking for it and have not been able to find yet). AMD is very good at utilizing the processes that is available to the best that it can be used (look at TSMC 40nm ATI 5xxx vs Nvidia Fermi). IMHO Bulldozer will be a superior architecture as long as they get all the bugs out.

    • Voldenuit
    • 9 years ago

    Here’s my guess: Bulldozer will be architecturally superior to SB in IPC, but will not be able to match intel’s process technology and clockspeed.

    Net result is that the processors on the market will be of roughly equal performance, maybe a slight lead for intel.

    However, GPGPU/OpenCL/DirectCompute/Fusion will take off, and intel will be left in the dust by ATI/nvidia GPUs for high throughput/fp-intensive computing. Adobe Premiere Pro already supports CUDA for video encoding, hopefully they’ll include OpenCL as well.

    I expect Apple to lead the APU race on the software front – why do you think they’ve been stealthily equipping their entire range with discrete GPUs? Wintel machines with intel IGPs will be the big losers, and CULV laptops with intel inside will become computing lepers.

      • Corrado
      • 9 years ago

      Or it could be they aren’t a fan of the performance at higher resolutions of the on-chip Intel graphics, and if you’re using an i chip, you’re forced to use that if you don’t use a discrete card as there are no core logic sets with integrated graphics for those CPUs.

        • Voldenuit
        • 9 years ago

        True. Also, using intel IGP for any serious application is a pain because of the horrible, horrible drivers.

        I know firsthand from trying to use Catia V5 and Matlab OpenGL output on an intel IGP in college. Caved and bought a 4850, as it was incapable of rendering even the simplest geometric constructs without serious graphical errors.

    • blastdoor
    • 9 years ago

    I can see two possible paths to success for AMD:

    (1) Intel screws up because they let the MBAs trump the engineers. For example, they try to use their CPU business as a way to force people to buy their inferior GPUs. That would create a very obvious opening for AMD to dominate.

    (2) Intel doesn’t screw up at all, but AMD cleverly balances their chips to beat Intel for some types of workloads while losing in other types of workloads. If AMD has the better chip for 30% of workloads, then they gain marketshare. This is probably the most realistic success scenario for AMD and it seems like this is probably what they’re aiming for with Bulldozer.

    • HisDivineShadow
    • 9 years ago

    I think there is a great possibility of AMD surpassing Intel’s best of this generation and being close behind Intel’s latest and greatest next year, if they don’t have any showstopping bugs they release with. However, I also think there is a good chance they’ll ship with some bugs in the hardware that they’ll be patching around and it’ll give Intel time to leap ahead again.

    That said, they’ll be getting close to a point where the superior GPU in Bulldozer will be enough to make up the diff and make a somewhat slower CPU part unimportant.

    If AMD can get an Optimus technology in place that switches between built-in and dedicated, even if it is only AMD GPU’s, then they could really be on to something in laptops since theoretically they should be able to make the CPU/APU and GPU very compatible with one another. Given that they’re making both of them.

    Meanwhile, nVidia and Optimus has to rely on the good graces of Intel and AMD to let their GPU work together with a competitor’s CPU.

    • willmore
    • 9 years ago

    As Uber will quickly point out, my workload isn’t a typical one for a desktop, so I’m going to refrain from voting lest I skew the results.

    I judge processor speed by how well it can run Prime95 as that’s what I run 24*7. Assuming George can find a use for the new AVX instructions/functional units, it’s not very clear which processor will come out ahead. If Prime95 doesn’t get an AVX code path, then the Bulldozer cores won’t do very well. But, if Intel does something equivalent to sweeping both chips of a *dale processor into one die, but leaving the memory topology the same, Sandy Bridge will suffer due to memory latency issues.

    For normal workloads, I would be that a 4 module/8 thread AMD chip will beat a 4 core/8 thread Intel chip clock for clock. Having a dedicated integer pipe beats having to share one with SMT.

    And, yes, I am a computer engineer and I do understand processor design.

      • Deanjo
      • 9 years ago

      r[

        • willmore
        • 9 years ago

        The only meaningful way to judge the performance of a system is by how well they run the target application. Maybe you missed where I clearly said that I only run Prime95. So, that’s the only meaningful metric for me. YMMV.

        Prime95 doesn’t seem to avail itself to DSPs as the working set is too large and their memory BW is too low–better than you and I have tried. GPUs have shown some progress in becoming useful, but there is a lot more work to be done on that front. Even if the GPU becomes a larger contributer, not everyone will have one and even for those who do, the CPU still can help out.

        • Shining Arcanine
        • 9 years ago

        The EFF has a $150,000 prize for the first person to discover a prime number not less than 100,000,000 digits in length. Prime95 performance is important to anyone interested in the prize.

    • FuturePastNow
    • 9 years ago

    I voted “kinda” but the margin of error on that is plus or minus two places on the poll.

    • blitzy
    • 9 years ago

    historically they have usually been similar in performance, there or thereabouts.. i would be surprised if AMD had the edge, but it would be nice for competition

      • Deanjo
      • 9 years ago

      I guess we will just ignore the P3 – Core years then. During that time period AMD did have the superior solution with the K7 and K8 up until the Core 2’s debut.

        • blastdoor
        • 9 years ago

        Yes, indeed they did. But at that time Intel’s head was very far and very obviously up its a$$. Will Intel make the kind of blunders now that they made with the P4? Probably not, but you never know — it’s amazing how stupid smart people can sometimes be.

          • Deanjo
          • 9 years ago

          *cough* Larabee *cough*

          They are still capable of making huge ass mistakes.

        • Usacomp2k3
        • 9 years ago

        The Northwoods were very impressive if you consider the clockspeed.

          • Shining Arcanine
          • 9 years ago

          The Pentium 4 C revision Northwoods were awesome processors in their time. It was what Willamette should have been and it was amazing that Intel’s engineering department could do such a good job with such an awful architecture. The Pentium 4 C revision Northwood processor were a true successor to the Pentium III.

        • Shining Arcanine
        • 9 years ago

        The K7 was only slightly better than the Pentium III at the same clock speeds in the benchmarks I saw a decade ago.

          • Corrado
          • 9 years ago

          Slightly better per clock, clocked higher, and cost less money. They eventually got WAY better per clock, and overall at a sometimes 25% clock speed deficit. They were also still less money for a good long while.

            • Shining Arcanine
            • 9 years ago

            Are you including overclocking? I thought that the Pentium III went to 1.3GHz at stock clock speeds, which was competitive with AMD Athlon processors at the time.

            • Pettytheft
            • 9 years ago

            Overclocking a P3 past 1Ghz was extremely difficult. And to get 1.3Ghz would probably require some liquid nitrogen. Unless you are talking about the Tualatin processor but that was only placed on the desktop after notebooks were keeping pace with the P4’s.

            • Shining Arcanine
            • 9 years ago

            I am referring to Tualatin.

            • yuhong
            • 9 years ago

            Well, it was on notebooks, in fact it was the only option because P4 mobile would not be released until 2002.

      • ronch
      • 9 years ago

      Well, Intel has like, 10x more R&D budget and probably half of the world’s Ph.D.’s. I’d be surprised if AMD does end up faster this time. Then again, it’s happened in the past.

    • PcItalian
    • 9 years ago

    Will bulldozer deliver?
    Not likely its shovel is mainly used for moving/grading earth not so much delivering a payload. If it is fitted with a backhoe of some sort it may be able to deliver, but not a large amount.

    Best option would be a dump-truck, they deliver. Or the pizza man. mm pizza.

    • OneArmedScissor
    • 9 years ago

    Poll: How many of you think you are CPU engineers?

      • khands
      • 9 years ago

      I’ve changed CPU’s before, therefor I have engineered something with a CPU, making me a CPU engineer.

      • indeego
      • 9 years ago

      Wouldn’t a business analyst be a slightly better judge? (analysts rule the universe, especially from Gartner.) AMD can make pie if it wants but if the market doesn’t hear about it or want pie, there you gog{<.<}g

    • Meadows
    • 9 years ago

    Considering the gimping we’ve been hearing from the general direction of Sandy Bridge, I imagine intel have no idea what they’re doing this round (while AMD pretty much /[

    • jdaven
    • 9 years ago

    “AMD still doesn’t have a desktop CPU fast enough to sell for more than $300.”

    Aren’t there like only 3 other price points from Intel above $300: $580ish, $899 and $999. And aren’t these price points served to about 0.00005% of the industry. I don’t think AMD needs Bulldozer to get above $300 in the desktop space.

      • Krogoth
      • 9 years ago

      The $399-999 range makes little or no sense for desktop users, when mainstream applications barely take advantage of value quad-core offerings. This arena is really for prosumers where time is $$$$$.

        • OneArmedScissor
        • 9 years ago

        …who would likely be better off with a multi-socket system that can run several CPUs for less than the price of one of those.

          • stdRaichu
          • 9 years ago

          Who’s making affordable 2P boards these days? Last I looked “enthusiast” boards with two sockets like the old Tiger MPX had gone the way of the dodo, and only super-expensive workstation boards existed.

      • grantmeaname
      • 9 years ago

      Don’t questions end in periods.

    • anotherengineer
    • 9 years ago

    I hope its on par, be nice if it performs better, but I don’t think it will better.

    If they are both on par, it would probably be best for consumers anyway.

    I hope AMD and the mobo makers address the cpu voltage though, my gigabyte board defaults overvolting to my 955 BE to 1.420V, I manually set it to 1.200V and its perfectly stable. My cousins asus board with the same chip defaults to 1.345V, quite a difference from the gigabyte, maybe thats why their power consumption has been higher on all the amd mobo reviews.

      • OneArmedScissor
      • 9 years ago

      Yeah, it’s cute how CPU-Z screen shots on sites like Anandtech tend to show voltages far, far beyond what the 65nm Phenoms ran, even on the models with conservative clock speeds. Not all motherboards do that, but Gigabyte, particularly, seems bad about it.

      I have to wonder if it’s done because of AMD’s push with unlocked CPUs and making it easier to “overclock” them, but it’s a bunch of malarky, either way.

        • Deanjo
        • 9 years ago

        I have yet to see one board AMD/Intel/Nvidia or other wise that has accurate voltage reporting using the the terribly inaccurate sensor chips that are commonly used on boards. They are all way out of whack, even physical placement of those chips matter.

    • Krogoth
    • 9 years ago

    I think AMD will able to maintain a slight performance advantage over Sandy Bridge architecture, but Intel’s manufacturing prowess will still continue to give them the edge. IMO, they are designed for two different segments. Sandy Bridge is really just a Lynnfield replacement, while Bulldozer is going after Gulftown and its eventual replacements.

    Bobcat is really where it is at. It looks like a genuine “Atom Killer” as a platform. Netbooks and small form factors are the future in the mainstream market. The mainstream already has reached a plateau in terms of computing power, the problem is how to best utilize all of its potential. There just isn’t a killer app that demands more than what the current quad-core offerings can muster.

    • Konstantine
    • 9 years ago

    I think a quad-module Bulldozer will outperform a quad core Sandy Bridge, if the performance figures AMD’s provided are “genuine”.

    On another topic, why the hell is a 4 module Bulldozer branded as a octa core? A whole bulldozer module with two 2-piplelined integer clusters and a 128 bit FMAC will likely be as big or smaller than a SB core with:? two 256 bit wide AVX units and a quad-pipelined integer cluster.

    • 5150
    • 9 years ago

    CHEEEEEEEEEEESSSSSSE!

    • Ryhadar
    • 9 years ago

    I voted, “Mostly, Equal to Sandy Bridge.”

    I’m not really in the position to answer this poll (and I’d imagine most people aren’t either) but to me there’s really only two, logical answers:

    A) AMD did a good job on their competitive analysis and was able to effectively communicate that with their engineering department to get the most out of the new architecture. Thus, Bulldozer will rival Sandy Bridge.

    B) They eff’ed up.

      • derFunkenstein
      • 9 years ago

      Yeah, that’s kinda where I am with it, too. And I voted the same way. Eithe rthey match Sandy Bridge or they fail. There’s no in-between, and I prefer to think that they made it. Especially if they can keep Socket AM3 compatibility. 🙂

    • wira020
    • 9 years ago

    I think bulldozer could rival Sandy Bridge.. maybe it’s just wishful thinking tho.. i havent read any extensive review on sandy bridge architecture yet.. so.. i’m just comparing these two by my understanding that sandy bridge is some sort of an APU targeted at mainstream consumer…

    • ClickClick5
    • 9 years ago

    Superior? Maybe… It might actually be fairly destructive with its new design. Benches will prove.

    • OffBa1ance
    • 9 years ago

    Something is wrong with the comment on percentage share numbers for hard drives:
    16% + 22% = 38% below 1TB
    19% + 12% + 14% + 6% + 11% = 62% above 1TB
    Article states “Most folks have yet to crack the Terabyte Threshold”?

      • Dissonance
      • 9 years ago

      Doh, was referring to most being between 500 and just under 1TB, which had the highest percentage… but ended up rewriting that sentence to include less than 500GB and that jacked things up. Fixed.

      • wira020
      • 9 years ago

      You’re right.. or is the number wrong?

Pin It on Pinterest

Share This