Leaked - AMD Barcelona versus Intel Clovertown and Tigerton

Leaked - AMD Barcelona versus Intel Clovertown and Tigerton

Summary: [UPDATE 9/10/2007 - My leaked extrapolated scores below are now officially confirmed on SPEC.org for Integer and Floating Point.

SHARE:
126

[UPDATE 9/10/2007 - My leaked extrapolated scores below are now officially confirmed on SPEC.org for Integer and Floating Point.  Since I only had two significant figures on percentage gain to work off of from the leaked slides, I was off by up to 0.3 which is to be expected.  AMD's official Barcelona launch announced newer scores submitted on 9/9/2007 which are now one point higher on SPECint_rate2006 moving up to a score of 89 which isn't too much of a change.  The SPECfp_rate2006 scores may have changed up half a point or so but it's hard to tell given the fact that AMD only gives two significant digits off their percentage gain.]

[Update 9/9/2007 7:30PM - IBM seems to have confirmed these leaked numbers with this PDF file labeled for release on 9/10/2007.  The scores are for an AMD Opteron model 2347 1.9 GHz quad-core and they're a little lower than the Barcelona 2.0 GHz scores I extrapolated from the leaked AMD slides for Monday's Barcelona launch.  It's still quite possible there are better scores for AMD's Barcelona but IBM's SPEC CPU 2006 scores seem to suggest that the numbes below are reasonable.

[Calculations on Intel advantage fixed at 7:30PM] IBM also gives us the single threaded results for SPEC CPU 2006 which allows us to see how fast each core can perform on its own.  This is very important for desktop applications (like games) and even server applications because there are some tasks that will always be single-threaded by nature.  From IBM's latest data, we can see that Intel still holds more than a 31.2% advantage on SPECint_2006 performance and more than an 18.2% advantage on SPECfp_2006 at 1.9 GHz.  This is calculated by looking at the Barcelona 1.9 scores of 11.3 and 11.2 on SPECint_2006 and SPECfp_2006 versus an Intel E5335 2.0 at 15.6 and 14 adjusted down by a ratio of 1.9/2 which is conservative performance for Intel at a theoretical 1.9 GHz.

[Estimates of Barcelona over K8 advantage added at 7:30PM] Over an existing Opteron "K8" 2.0 GHz processor that gets 10.7 on SPECint_2006 and 10.9 on SPECfp_2006 adjusted down by a ratio of 1.9/2, the Barcelona 1.9 GHz core boosted SPECint_2006 by ~11% and SPECfp_2006 by ~8%.  This pretty much confirms what I've been saying that Barcelona will not significantly improve single threaded IPC (Instructions Per Cycle) over AMD's K8 architecture.

Intel still holds a massive clock-for-clock lead when it comes to single threaded performance and this is primarily due to the 4-issue execution engine in Intel's Core Micro-architecture versus AMD's 3-issue execution engine.  What allows AMD to narrow the gap when we get to multiple CPU sockets is the Integrated Memory Controller per Processor and HyperTransport memory architecture.  But in order for AMD to beat Intel on performance, they will need to ramp their CPU close to 3 GHz but they have a long way to go at this point.]


It appears that some slides for AMD's Barcelona launch next Monday has come in to my possession and SPECint_rate2006 benchmarks were included.  AMD caught wind that I was going to be analyzing and posting the leaked data and they just called me explaining that they wanted to give me some updated slides.  I don't know exactly what it was but I know that some of the numbers posted for Intel on SPECfp_rate2006 were outdated.  I explained that I never signed (or verbally agreed to) an NDA on these slides and I didn't intend to sign an NDA at this late point in the game and the gentleman from AMD was fine with that.  But once I explained what I was going to do, AMD's rep asked me for 30 minutes to check on some numbers.When he called back, I never got any updated slides but I did get a strong hint that I might be posting inaccurate information on Barcelona performance and that maybe I wouldn't be journalistically accurate.  The gentleman strongly hinted that the numbers may change on the official launch next Monday compared to the numbers that I've extrapolated for now.  Now this sounded nothing like the first call and it sounded to me like AMD just didn't want the numbers posted.  Of course it's quite possible that newer SPEC benchmark numbers were submitted and the slides that I have are using older numbers.  Since the word "change" was used, that tells me that the numbers I have are legitimate but they may have been tweaked a bit since.Since the leaked slides indicated that the performance estimates were submitted to SPEC.org on 8/20/2007, that looks fairly new to me though it does not rule out the possibility of newer and better numbers for AMD.  I *cannot vouch* for the accuracy of the Barcelona leaked numbers and I cannot rule out the existence of newer Barcelona numbers.  But since the slides look authentic and the performance numbers are within the expected range, I've decided to post the performance numbers and let the reader decide if they're authentic or not. Barcelona SPEC CPU numbers allegedly submitted 8/20/2007[Update 11:00PM] Added September numbers for Intel.  Note that it is possible that AMD has numbers newer and better than the ones submitted on 8/20/2007 in the leaked slides so it's possible the number for AMD will go up a little. AMD Barcelona versus Intel ClovertownThe numbers here are as of August 2007. Barcelona SPEC CPU numbers allegedly submitted 8/20/2007The 160 score for the Barcelona 8350 2.0 fall within 4% of my "mid 150s" estimate.  Unless AMD has better numbers for Monday's Barcelona launch (they are hinting they do), it appears that even a Tigerton 1.86 GHz 4-socket Server beats a Barcelona 2.0 GHz 4-socket Server on SPEC CPU 2006 16-thread integer performance.  This latest picture on integer performance also explains why AMD priced Barcelona so low against Intel.[Update 9/8/2007 - Calculation fixed for Barcelona.  35% advantage should have been over August E5345 scores and not off September E5345 scores.] Here are the SPECfp_rate2006 peak numbers which are used for the niche HPC market.  The leaked Barcelona 2.0 numbers are based on AMD's slide that claims a 35% advantage over Intel's 5345 as of 8/28/2007 (this is the date that these slides were generated).  The numbers below show updated September numbers for Intel but it is possible that AMD has newer and better scores than the ones submitted on 8/20/2007. Barcelona versus Clovertown

Topics: Processors, Hardware, Intel

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.

Talkback

126 comments
Log in or register to join the discussion
  • "This latest picture on integer performance explains why AMD priced...

    George, when you are in the realm of speculation and dealing with "leaked slides" might I suggest that you steer clear of these black-and-white, declarative sentences that so often get proven wrong ? It would help your credibility.

    And why did you not get these slides directly from AMD if they are legit? Do you hear from the tech companies you cover?
    WarEagle
    • If they had given them to me, I would have been under NDA

      If they had given them to me, I would have been under NDA and I wouldn't have talked about these slides.

      As for being proven wrong "so often", I don't know what you're talking about. Even if AMD manages to push the numbers a little higher than these numbers submitted on 8/20/2007, the statement would still be correct because there Intel has the high-end. Even if they manage to edge out the lower-clocked 1.86 Tigerton, it still doesn't look good for AMD.
      georgeou
      • Combine your facts and reevaluate

        "Even if they manage to edge out the lower-clocked 1.86 Tigerton, it still doesn't look good for AMD."

        You're becoming increasingly cautious about such statements in your blog post and still everybody seems to notice that general attitude. You spell it out in the talkback anyway: Your stories are supposed to show that Barcelona will be a disappointment for AMD.

        Now look at what you're actually saying: That "1.86 Tigerton" is the third best performing part in Intel's MP lineup in SPECint_rate2006. That means 4 of 6 SKUs are beaten by AMDs Barcelona over its price advantage and a lack of performance advantages on Intel's side. Even ruling out the SPECfp superiority of the Opteron, which exists however small the target market may be, my conclusion would be that it doesn't look good for Tigerton. How large do you think the market share is for the remaining 2 parts at their given price premium and performance advantage? In a black-and-white scenario those parts would be the only ones Intel would sell.
        CFKane
        • I said no such thing, you did.

          "Your stories are supposed to show that Barcelona will be a disappointment for AMD."

          I said no such thing. I said AMD will be competitive with their pricing and lack of FBDIMM on energy efficiency on the low-end with Barcelona 2.0. I've also said AMD can be more competitive in the mid-end if they can quickly get to 2.3+ GHz. The problem is that while AMD remains at 2 GHz, they will take a brutal beating on margins and AMD's pricing pretty much confirms this.

          It is a fact that people buy 4P servers not for value or energy efficiency. Energy efficiency and value on 4P servers is an oxymoron. The 4P market cares about performance first and foremost and AMD won?t be competitive on performance until they get to the high 2s on GHz. Furthermore, there is a halo effect of owning the leadership position. People seem to assume the entire platform is better if the vendor owns the leadership position.
          georgeou
          • Poor Intel

            "It is a fact that people buy 4P servers not for value or energy efficiency. Energy efficiency and value on 4P servers is an oxymoron."

            Too bad Intel will build up massive inventory of everything except the X7350 in their lineup and will suffer a multi-million dollar write-off when they replace the current offerings with their 45nm parts. How insane must they be to blow fuses and cripple the cache to end up with processors people don't buy? It's very fortunate for AMD that they do not have the capacity to produce nearly as many CPUs that they can't sell anyway...

            "People seem to assume the entire platform is better if the vendor owns the leadership position."

            That is for once right. Those people are fewer among deciders in companies which purchase large amounts of servers, though. It appears one finds them much more frequently in the blogsphere.
            CFKane
          • Energy Efficiency and Price per Performance of 4-way

            George,

            A good read of the preliminary swirl before formal processor launch. Just passing by, allow me to comment and correct some inaccurate assumptions on your part here, part of which I'd like to throw my own experience on, and part of which agree with some of your current published interpretations.

            I work in a governmental department and I'm not going further than that to state at what level, the specified area or where, but suffice to say, part 'n' parcel of my profession is scientific and high performance computing. All the way from high end workstations to 4-way to as many as possible with a given microprocessor, revamped by our own expertise most of the time, if need be. We are affiliated to many departments, research centers, and medical institutes, so my work is extended on per demand basis and is quite flexible since I have the required prerequisites and experience needed applicable to many of the tasks which may be on offer. Now, just before AMD Barcelona Opteron release I've been in contact with AMD and Intel in regards to their product offerings, both, since it's about the right time we usually purchase our updated computing stock and the board higher up from where I stand deem it appropriate to do so. I have the relevant feedback from both parties concerned.

            I'm the person here with formal authority for making an initial product evaluation submission based on performances and so forth.
            Let me make it unequivocally clear, when you purchase at this critical level with such [u]capacity[/u], performance value and financial input, you "[i]gnarly[/i]" well take into accord the best product delivered which meets many criterion's, inherently including and not limited to those of; total processor power consumption annually at throughout a range of given loads, total platform power consumption at a range of given loads, total power consumption at idle, overall power efficiency, performance scaling through processor additions, power usage and performance over a wide range of object and demand orientated applications and computing usage, processor scaling architecturally (yes, analysis of a given products architecture for variable uses and refinement projects), processor frequency scaling [etc), and first and foremost of what the board will question is the availability of, the availability of products for, the availability of expertise for such a platform/processor and the [b]aggregated annual and initial expense[/b] required to enable these centers to be up and running with the chosen platforms (including processors assessed on individual merits). Finance is hugely contended and fought for within the industry, so to flaunt and extravagantly, and quite frankly, recklessly waste it is by no way something that is remotely close to what we or others in similar positions I am aware of have, would or will do.

            Now, you can be rest assured, that a given processor doing well in one of our internal first segment critical analysis reports or performing successfully in one of our associated laboratories will have a knock-on-effect on the rest of our business interactions and affiliates, as whichever processor is analyzed by a authority higher up with more specific expertise and resources to do so reliably and thoroughly, it's quickly favored and adopted by many further below the similar market chains, causing a mini, product segment orientated monopoly in a certain sense, constrained within a technology timeline, if you may. A sense of pre-dominance of a certain brand at the time of release augments such success greatly, just as you mentioned..

            Say, over-simplistically and hyperbolically, you were allocated 50 million worth of currency funds in whichever developed occidental nation you choose, for example, to acquire, provide for, and sustain 5x 200 professional scientist + 250% that magnitude personnel data management, analysis, research and development centers, equipped with certain charters and to work inside specific territories of budget confinement... you have 5 years to work with such funds. Now, rule out 45 million straight away as proceeds confined to major premises, operational and project costs. Technological infrastructure gets 5 million of the working funds. Right.
            Now keep in mind you're to submit a full cost analysis and accounting for every base currency decimal possible every 2Q. Supercomputing at the in-house level usually seeds its success through architecturally strong, flexible, efficient, low TDP, relatively inexpensive and low clocked frequency processors.
            Tally onto this your need for additional technological devices for everyday standard work and function in the modern day, such as communication, transport and so forth over this forecoming period. Your budget is now down to ~1 million of this given currency for specific computing platform purchasing power. Moving on swiftly, you've now purchased the low-high end desktop and low-mid end workstation systems required under your management. You now have 500 thousand of the total initial allocated fund left. Can you really successively equip 1000 personnel with the required computing power and additional computing resources they will require within this given budget and yet, not really care about paying 1000 x100 over 2000 x 100 just for one processor + internal operating costs over such a period? Even 200 higher. Factor in 20% increased end working energy costs for any one processor overall, and the cheaper by virtue of both metrics will be a winner by a long margin as long as it's performance does not suffer a big hit compared to the more expensively priced offering. Provide 1x 500 high-end workstations on top of this, and now you see where budget management and compromises start to churn their way in very transparently at every angle, and your product cost-benefit analysis brief begins.

            Believe me, on paper the better, but in reality, the picture isn't a blooming pink rosy one when in such non-propensity positions, and price does in fact cost dearly, thus is a major, highly critical factor wherever it can creep up. ;)

            As an addenda, as technology enthusiast, you maybe aware that price is exactly what flopped the Opteron 8xxx series sales revenue throughout 2006, in comparison to the relatively very similar 2xxx series, though demand for this processor was high.

            Look forward to all and every sector and application benchmark across the various setups, configurations and platforms.
            qegs20012001@...
          • My point on 4P energy efficiency is that it is less significant

            My point on 4P energy efficiency is that it is less significant. When you're talking about buying a 4P server, you're someone who's willing to pay 3x the money for 1.7x the performance. You've easily spent an extra $20K on just 70% more performance. If you're looking at an extra 150W on power consumption and double it to account for extra cooling costs, we're looking at 8 cents per KWH, we're looking at $210/year or around $2000 for the lifespan of the product. You have factor that extra power in to the cost of ownership but it's relatively small compared to the premium you're paying on 4P servers.

            Come November when 2P Penryn systems on FSB1600 start threatening Barcelona 4P 2.0 GHz systems, things will get interesting again.
            georgeou
  • RE: Leaked - AMD Barcelona versus Intel Clovertown and Tigerton

    lol same ole same ole,where are the floating point numbers George with the 2 scores combind AMD wins again George,post all the scores not just the ones that favor Intel.I found this today,history of the AMD ,Intel debate,what the first poster says is true
    http://www.geek.com/forums/topic.php?id=33092&page
    This even explains Intel controlling the media

    Anyway since you havent posted all of the slides,Im gonna guess AMD beat Intel on every other test,I know they did on floating point I also no you think thats not important lol
    AMD OPTERON
    • AMD announced SPECfp already more than a month ago

      AMD announced SPECfp already more than a month ago and there's nothing new in the slides. I've also stated that AMD dominates SPECfp. AMD posted a bunch of other scientific HPC benchmarks but nothing on SAP, SPECweb, SPECjbb, TPC.
      georgeou
      • OEMs post most of those scores

        The key is what will AMD say about floating point and integer. OEMs submit those other scores, and we know all four big ones have lined up for Barcelona.
        Floating point has AMD taking Intel behind the woodshed.
        Anything less than dominance for Intel on integer on mainstream stuff and it has a real problem on its hands with all the huffing and puffing about dominance we've heard from Gelsigner, Otellini and Bryant. To hear them talk, the score is 72-0 Intel.

        And did you see Bryant say that Intel doesn't even know HOW to make a quad-core like AMD unless they are mature on 45nm? They so need the big cache to hang with AMD, they can't fit four cores on a die. Pathetic, really, for the biggest chipmaker.
        WarEagle
        • Putting in a 4-issue execution engine and a large cache is a design choice

          Putting in a 4-issue execution engine and a large cache is a design choice that vastly improves single threaded performance and the vast majority of desktop and server applications. AMD made a design choice to go with an integrated memory controller and Hyper-transport to improve memory latency and bandwidth which helps in HPC applications.

          What's pathetic is your insistence that once choice is better architecture than the other. Each architecture has pros and cons and Intel strategically used their transistor budget for better performance on the vast majority of applications. AMD's choice led to better HPC computing which is a minority of the market. I think it's very clear who made the right decision when you look at the financials.
          georgeou
          • ONLY BECAUSE INTEL CONTROLS YOU

            Its simple George Intel controls the media,people like you,you nit pick 1 or 2 percent some places and totally miss 20-30 persent advantages other places,on 30 OR MORE threads of data I dont care what it is,AMD will kill Intel and at present using less power on everything thats a fact.
            AMD OPTERON
          • If brains were eggs yours would be scrambled

            Get used to it. Right now Intel is better. Its not the end of the world. There is no need for you to make a complete fool of yourself clutching at falling straws to try and hold yourself up.

            Its likely at some point in the future AMD will make their gains again and Intel will be on the downside again. Take a pill and stop railing against reality. It makes you sound like an idiot.
            Cayble
          • you can read IC

            LMFAO CANT YOU READ
            AMD OPTERON
          • George, how do you spell Nehalem?

            B-A-R-C-E-L-O-N-A

            Let's see:
            - AMD takes x86 to 64-bits in 2003, Intel follows years later
            - AMD brings Direct Connect to market in 2003 and blows out x86 server bandwidth and scalability, Intel follows in 2008/2009
            - AMD brings performance-per-watt to the datacenter in 2003, Intel follows in 2006
            -AMD brings dual-core to x86 in the datacenter in 2005, Intel slaps two single-cores together the next year. Does a real dual-core two years later
            - AMD embeds memory controller in 2003 and kills the memory bottleneck, Intel follows in 2008/2009
            -AMD brings quad-core design to the x86 datacenter in 2007, Intel follows in 2008/2009.

            Yes George, by the time Barcelona was on AMD's roadmap, Intel had enough practice glueing processors together in a package with a front side bus to say "look we finished first". Lucky for Intel AMD is getting Barcelona out there when it is and not sooner.

            But the ultimate proof of this point is Intel's own roadmap. They utterly validiate every design choice AMD makes. It just takes them a long time.

            How many Intel engineers does it take to make a quad-core? Let's see, three of them. One to work the gluegun. One to photoshop the dual-core die shot into a quad-core die-shot (copy, paste, group the die shots side-by-side) and one to write the press release.

            And it is still pathetic that Intel doesn't know how to make a real quad-core with 65nm, and yet they say they are the gods of manufacturing. Please.
            WarEagle
          • You make AMD look bad. I know plenty of AMD folks that do not behave this

            "How many Intel engineers does it take to make a quad-core? Let's see, three of them. One to work the gluegun. One to photoshop the dual-core die shot into a quad-core die-shot (copy, paste, group the die shots side-by-side) and one to write the press release.

            And it is still pathetic that Intel doesn't know how to make a real quad-core with 65nm, and yet they say they are the gods of manufacturing. Please."

            You make AMD look bad. I know plenty of AMD folks that do not behave this way.
            georgeou
          • George, you said the *purpose of this blog* was to make AMD look bad?

            "A big part of the problem is that some in the press often gives AMD the baby glove treatment" - George Ou

            http://talkback.zdnet.com/5208-10533-0.html?forumID=1&threadID=35931&messageID=660342&start=0

            And remember, you've said when AMD has the better product, you'll let Intel have it.

            Oh and by the way, floating point isn't only for HPC. Ever here of a little thing called the PC gaming industry? 3D takes great FPU performance.

            And don't forget that Intel's best integer numbers are only on its HPC compiler. Take those scores down a big notch for everyday enterprise computing. Don't you think customers will see this on their real workloads?
            WarEagle
          • Intel dominates SPECfp on single and dual threads, which applies to gaming

            Gaming is predominantly single threaded and single socket. We're luck if a game can take advantage of two cores so it's extremely unlikely that any game will even come close to stressing the memory controller and FSB.

            Case in point since you like talking about SPECfp, Intel dominates SPECfp_2006 and SPECfp_rate2006. Intel has a 30.77% 21.22% clock-for-clock advantage at 3 GHz when you compare 65nm Core Micro-architecture to AMD K8 3.0 GHz single and dual thread on SPEC CPU 2006 FP.
            georgeou
          • And BTW, I never said "And remember, you've said when AMD has the better"

            "And remember, you've said when AMD has the better product, you'll let Intel have it."

            I said no such thing. The only thing I've ever said is that I'm loyal to my wallet. The product that meets my performance requirements and gives me the biggest bang/buck gets my business.
            georgeou
          • Gaming on a server chip?

            George,

            You're getting yourself mixed up now.

            We're talking servers here. Having better speed for Far Cry means jack in this market. (I'm willing to bet gaming on a 2U server box is an even smaller "niche" than you like to berate HPC as.)
            Robert Crocker