Apple has issued SPEC benchmarks for its new 2GHz Dual CPU workstation, pitting it against a two-way Dell Xeon and a 3Ghz Pentium 4 machine.
The results are in line with Intel's historically strong performance in integer performance and PowerPC's equally strong floating point results. But according to Apple the gcc compiler used did not support the AltiVec floating point instructions, and yet the G5 machine managed to trump both Dells on trial, scoring 840 to the Dell Pentium's 693 and the dual Xeon's 646 in the SPECfp_base2000 benchmark.
The G5 lagged behind both Dells in integer performance, notching up 800 to the Pentium's 836 and the Xeon's 839.
However with the parallel "rate" benchmarks, which tax both of the CPUs in the test machines, the G5 edged out the Xeon 17.2 to 16.7 in the integer score and 15.7 to 11.1 in the floating point tests, suggesting Apple makes far better use of its two CPUs than the Xeon machine. It isn't clear if the Dell machine was optimized for SMT, or Hyper Threading, but the results augur well for Apple G5 performance in technical and scientific computing environments and for playing games.
Apple engaged VeriTest to perform the benchmarks. The most recent submissions to SPEC for integer and floating point benchmarks show higher results than the figures disclosed by Apple. The SPEC submissions typically use an Intel Fortran compiler.
A clutch of other benchmarks have been released by Apple, including a Quake 3 benchmark at 1024x768 giving the G5 a score of 337 frames per second compared to the Pentium's 275, and impressive Photoshop and Cubase plug-in scores.
Update: New document, old SPEC scores: the VeriTest benchmarks are the same tests that were debunked here, and defended here - in great detail. Curiously, Apple's SPEC scores are lower than figures for early G5 silicon supplied to The Register last October. ®