Shame, Petz. So much FUD, and no attempt at equal comparison.
Allow me to do the work for you. I'm choosing base because there are not enough peak scores to do a thorough comparison.
In SPECint_rate_base:
IBM System x 3655 (AMD Opteron 2212, 2GHz) - 37.4
IBM System x 3455 (AMD Opteron 2222SE, 3GHz) - 50.2 (1.34x)
IBM System x 3550 (Intel Xeon 5130, 2GHz) - 45.4
IBM System x 3550 (Intel Xeon 5160, 3GHz) - 60.8 (1.34x)
(IBM does not have the best comparison, but here's the new Intel submission as well....)
IBM BladeCenter HS21 XM (Intel Xeon E5335, 2GHz) - 77.4
IBM System x 3650 (Intel Xeon X5365, 3GHz) - 96.4 (1.25x)
Supermicro X7DB8+ (Intel Xeon processor E5335, 2.00 GHz) - 78.5
Supermicro X7DB8+ (Intel Xeon processor X5365, 3.00 GHz) - 98.4 (1.25x)
Your conclusions might be:
- Intel scales equally well to AMD, dual core vs dual core
- Intel scales worse with quad core vs Intel dual core
Your conclusions cannot be:
- Intel quad core will scale poorly in frequency vs Barcelona, since we don't know how poorly Barcelona will scale
In SPECfp_rate_base:
IBM System x 3655 (AMD Opteron 2212, 2GHz) - 37.8
IBM System x 3455 (AMD Opteron 2222SE, 3GHz) - 48.6 (1.29x)
IBM BladeCenter HS21 XM (Intel Xeon 5130, 2GHz) - 33.8
IBM BladeCenter HS21 XM (Intel Xeon 5160, 3GHz) - 41.4 (1.22x)
(No Clovertown results from IBM, but here are 2 from other vendors)
Supermicro X7DB8+ (Intel Xeon processor E5335, 2.00 GHz) - 53.4
Supermicro X7DB8+ (Intel Xeon processor X5365, 3.00 GHz) - 63.1 (1.18x)
PowerEdge 2950 (Intel Xeon processor E5335, 2.00 GHz) - 52.0
Dell Precision 690 (Intel Xeon X5365, 3.00 GHz) - 61.3 (1.18x)
Your conclusions might be:
- Intel scales slightly worse than AMD, dual core vs dual core
- Intel scales slightly worse in quad core vs Intel dual core
Your conclusions cannot be:
- Intel quad core frequency scaling will suck vs Barcelona, because we don't know Barcelona scaling, yet.
In both cases, the data does not show any substantial AMD advantage, except that they have slightly better frequency scaling in SPECfp_rate (big surprise). It's not clear whether Barcelona will scale well, or even as well as their dual core implementation, and no: Native Quad Core doesn't automatically guarantee better scalability.
In other words, Petz, your post was complete FUD. Intel's dark little secret is that neither they nor AMD can scale in SPEC_rate perfectly well with frequency. But of course, most people expected this anyway. |
