AMD Barcelona versus Intel Clovertown and Tigerton

[bluetooth]

IBM System x3455 delivers solid performance on SPEC CPU2006
x3455 posts competitive SPEC CPU2006 scores for a 2-socket quad-core system

September 10, 2007 ... The IBM® System x™ 3455 server excels at the compute-intensive applications used in high-performance clustering environments. The highly manageable, ultrathin 1U, rack-optimized platform is a 2-socket-, 8-core-capable server designed for high performance
at an affordable price. (1) The x3455 server features AMD Opteron™ quad-core processors and IBM’s Xcelerated Memory Technology™, which enables faster access to data, and helps provide outstanding performance
for large memory configurations with the ability to support 667MHz memory access speed of fully populated memory. The x3455 supports up to 48GB of PC2-5300 667MHz ECC DDR2 memory with IBM Chipkill technology for high performance and reliability. In addition, the AMD Opteron quad-core processors used by the x3455 incorporate AMD’s dual dynamic power management technology designed to increase performance per watt while reducing power consumption.

The x3455, using the quad-core AMD Opteron 2347 processor (1.9GHz, 512KB L2 cache per core—2 processors/8 cores/8 threads), demonstrated competitive performance for a 2-socket AMD system on the SPEC CPU2006 benchmark suite. These results were achieved using SUSE Linux® Enterprise Server 10 SP1.

The scores in the table are the first SPEC CPU2006 results published for this x3455 processor model.

SPEC CPU2006
Benchmark
x3455 – Quad-Core AMD Opteron Model 2347 Processor
(1.9GHz, 512KB L2 Cache per Core)
SPECint2006 11.3
SPECint_rate2006 83.2
SPECint_rate_base2006 72.8
SPECfp2006 11.2
SPECfp_rate2006 73.0
SPECfp_rate_base2006 68.5
Results are current as of September 10, 2007. The scores have been submitted to SPEC for review and will be posted on their Web site upon successful completion of the review. View all published results at www.spec.org.

(1) Planned availability for the x3455 model using the AMD Opteron Model 2347 processor
(1.9GHz, 512KB L2 cache per core) is November 16, 2007.

[bluetooth]

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