AMD has been rather harshly criticized for its constant refusal to reveal any benchmark data, and instead post theoretical information about the expected performance of K10. Well the launch is coming up, September 10, and it seems the first real benchmarks has appeared. A user at coolaler.com have posted screenshots with SuperPi 1M, Cinebench, Fritz Chess Benchmark, wPrime and CPUmark. The system used is a dual-socket server with two quad-cores running at 2.0GHz. The least impressive of the benchmarks is the SuperPi 1M that results in measly 39.657 seconds. This is pretty much the exact same score a 2.0GHz K8 processor would get. On the other hand, Cinebench shows a very good multiprocessor scaling (7.01x), but is still slower than a similar Intel Core system.
wPrime, Fritz and CPUmark all tell the same story, but it seems that none of the benchmarks can be trusted. According to a post made by CPU-Z at XtremeSystems something is not entirely right with the system used for these benchmarks. The HT Link seems to be disabled and there is apparently a good chance parts of the cache are too.
Some sources say that the system used here is of an early stepping and that it does not reflect all of the changes made to Barcelona during the summer. However not all stepping B0 processors are bad, there are those who are free of bugs. All in all, this is probably not a true reflection of Barcelona, but may still be a hint. We will just have to wait and see.
