Slashdot Mirror
Intel V8 Octa-Core System, Full Performance Tests
MojoKid writes "In the April time frame,
details of Intel's dual-socket 8-core system dubbed
'V8' became available but only preliminary performance numbers were shown.
The platform consists of quad-core Xeon processors in an Intel 5000X
chipset-based motherboard, along with FBDIMM (Fully Buffered DIMM) serial
memory. A follow-on article at HotHardWare goes into significantly more detail on the platform and showcases many more performance metrics on a Windows Vista 64-bit installation. The
POV-Ray and Cinebench 95 benchmark numbers alone are something to smile
about. 'Intel's V8 isn't about promoting a platform as much as it is a show of strength and a glimpse of things to come. What V8 and QuadFX show is that both Intel and AMD are on a path to offering true, enthusiast-class, dual-socket platforms. And that's a good thing. Perhaps AMD is a little further down the path thanks to a more tweaker-friendly motherboard in the QuadFX-compatible Asus L1N64-SLI WS, but until consumers have more motherboards to choose from and perhaps quad-core processors from AMD, we can't very well declare that the time for QuadFX has arrived. One motherboard does not a platform make.'"
Figuring out how to redesign a program to run in parallel is a terribly difficult thing to do, for the most part. There are sometimes linear algebra problems that appear in science applications that lend themselves to parallel coding, but those aren't things that most users are trying to implement. *They cannot give up on making sequential processing faster.* Making a platform as massively parallel as this is (for a personal computer) will never accomplish what improving other facets of the architecture like memory latency, cache size, and of course the chip frequency. So we have been using machines with four processors for 11 years, and for the most part only one processor gets utilized even after extensive efforts to make our applications run in parallel. The overhead for farming out work is worthwhile only when you have very large chunks of computing that doesn't have to be sequentially processed. I really see this multicore processing stuff as a bit of a cop out.