Replies to post #11887 on Intel Corp (INTC)

[wbmw]

Duke, Re: Fujitsu has been making Spark processors for a while now. I am not sure what is new in this announcement.

Fujitsu's SPARC processor is more competitive than Sun's, so it is an improvement. OTOH, neither SPARC variety can hold a candle to IPF. Here are some comparisons.

SPECint:
Fujitsu Limited PRIMEPOWER650 (SPARC64 V 1350MHz) 776 905
Hewlett-Packard Company HP Integrity Server rx4640 (1500 MHz, Itanium 2) 1380 1404
SGI SGI Altix 350 (1400MHz/3MB, Itanium 2) 1078 --
Sun Microsystems Sun Blade 2000 (USIII Cu 1.2GHz) 642 722

The fastest Fujitsu submission is 21%/25% ahead of the fastest Sun submission in base/peak score, but still 28% behind the low cost, DP optimized Itanium 2 SGI platform in base, and 44%/36% behind the fastest HP submission in base/peak.

SPECfp:
Fujitsu Limited PRIMEPOWER650 (1350MHz) 1096 1340
Hewlett-Packard Company HP Integrity Server rx4640 (1500 MHz, Itanium 2) 2161 2161
SGI SGI Altix 350 (1400MHz/3MB, Itanium 2) 1931 1942
Sun Microsystems Sun Blade 2000 (1.2GHz) 1074 1344

The fastest Fujitsu submission is about on par with Sun's fastest, though it's likely that Sun's compiler optimizations would help the Fujitsu system to score better. The scores are still well behind IPF (43%/31% behind the low cost, DP optimized SGI submission and 49%/38% behind the fastest HP submission).

SPEC OMPL2001:
Fujitsu Limited PRIMEPOWER HPC2500 (SPARC64 V 1.3GHz) (124 thread) 262140 --
SGI SGI Altix 3000 (1500MHz, Itanium 2) (128 thread) 536169 --
Sun Microsystems Sun Fire E25K (72 processor, 143 thread) 240622 316182

A difficult comparison, given the different number of cores and processors, but the single core Fujitsu SPARC64 V processor performs well against the dual core UltraSparc IV processor. OTOH, with a similar number of processors, it is still behind the SGI Itanium 2 submission by 51%.

SPEC JBB2000:
Fujitsu Limited Fujitsu Limited PRIMEPOWER1500 HotSpot Server VM on Solaris/SPARC,version1.4.1_01 (1.35GHz SPARC64 V, 32 processors) 492683
Hewlett-Packard Company Hewlett-Packard Company HP Integrity Superdome server (32 processors, Itanium2 6M) HP Hotspot 1.4.2.00 (32-bit) on HP-UX 11i v2 for Itanium2 574912
Sun Microsystems, Inc. Sun Microsystems, Inc. Sun Fire E6900 HotSpot Server VM on Solaris/SPARC, version 1.4.2 (48 cores, 24 chips, 2 cores/chip) 421773

Another hard comparison, given that the only Sun submission has 1.5x as many cores as the best SPARC64 V submission, but if you do the math, it's easy to see who comes out on top. Fujitsu still scores 14% below the similarly configured HP Itanium 2 submission.

There's not many other benchmarks that have all three architectures side by side, but I think Sun's logic is that Fujitsu has a far better single core processor, and the dual core processor due in early 2006 will be better than anything Sun could have done on their own. Of course, the Sun announcement does have implications on the schedule of the supposed "Rock" processor. Originally, this was due out in 2006, but an agreement with Fujitsu might mean that it was delayed until 2007 or 2008. If that's the case, then Sun does not have a competitive roadmap going forward, and SPARC's days are numbered.