I've been skimming through Anandtechs CPU benchmarks and Intel really just BLOWS away AMD.
Intel has low-end dual core processors that can outperform AMD's high end quad (and even hex) cores in a variety of tests; why is this?
Very intersting thread. Many answers, many simplifications or simple speculations.
However I can give you a deeper answer so it is not an easy answer:
1. The AMD design K7-K10.5 is a design for high parallel throughput with average latency.
2. Anandtechs benchmark set is not capable of utilizing parallel throughput.
3. Intel's design has bad(P4)/average(Core) throughput but excellent latency (Core). Intel P/PII/PIII/Core is a design for minimum latency and since Core with reasonable throughput. (PIV hat bad throughput and only average latency).
4. Anandtech benchmark set is extremly sensitive to latency.
5. Those effects are amplified by most (all?) Anandtech benchmarks compiled with Intel compilers which not only emphasize latency but in addition create assembly code to hurt AMD performance. That is why if you look at the SPEC benchmark set the picture is somehwat different where no Intel compilers are taken for benchmarking AMD.
Therefore this heavy throughput focus of AMD gives and gave it bad results in benchmark sets used by Anandtech. Or other way, Intel always had an advantage in latency (with the exception of PIV).
It has as well to do with R&D expenses as providing performance gains by high throughput can be achieved with lower R&D expenses, but providing performance gains by low latency require higher R&D expenses.
It has little to do with manufacturing as Intel and AMD are both way ahead in manufacturing. But you could say, that the high expenses of AMD in maufacturing allows less money to be used for R&D. Intel pays a lot of extra money to be ~6+ month ahead of AMD in using a smaller process technology on the other hand AMD has the superior process technology with SOI. I would not say that there is any distinct impact on performance by this.
So far regarding your question why this is like you describe.
AMD Bulldozer will likly completly reverse this picture as they nearly half the latency and double the core count which would speed up AMD CPUs by almost factor 3! However the Anandtech benchmarking software set will show a little lower performance improvement because of scaling issues.
As AMD had generally no problem being behind Intel regarding top performance in Anandtech benchmark set as they just sell sell their chips at competitive performance/price points, a reversed situation could be really completly new for Intel. This will not endanger Intel as a company but it could have heavy impact on their profits.
I assume that Intel will react with providing more cores and pushing out higher speed bins to reduce the gap. This year could get really interesting.
Or and this comes back to your initial question, AMD invested a lot of time and resources to make their new Bulldozer design and that reduced resources to narrow the gap in Anandtech Benchmark set compared to Core CPUs with improvements of their current designs.
Facebook
Twitter