These work units are from a different series. So the calculations on them are going to be possibly harder or easier (apparently harder from what all have said). These two batches were hand picked by the researchers to give them a good sample set to know if we are able to go live with the updated application version. Of course there is no time table on going to production, need to be thorough with our testing first.

Thanks again for all who have participated,
-Uplinger

9 WU all valid. Not many hours but more than I had before.

Returned 5 units yesterday all valid giving me another 7.02 hours

Got about a dozen WU's across the 8 PC's but lost 6 of them to server abort which a real annoyance.....

With the halving of HCC WU CPU times a while back, plus this beta test, it would seem that HCC scientists/programmers have been analysing & optimising their code for speed. 2 comments/questions:

1. Do any of the WCG sciences use the CPUs' vector instructions (MMX, AVX, ...) or do they perform matrix arithmetic etc using scalar instructions and software loops? I think I read here some time ago that they only used scalar instructions because vector instructions on different CPU models might give slightly different results, giving WU validation problems. OTOH, if use of GPUs for HCC is still being considered, the validation problem must be solvable, and I suggest that using the vector instructions might be a better alternative to GPGPU because this would be much easier to program plus most or all x86 CPUs have vector instructions but only a minority of computing devices are equipped with GPGPU-capable graphics cards.

2. 64-bit HCC? The increase in speed for the 64-bit version of C4CW seems disappointing. A 64-bit version of HCC might do better because HCC uses mostly integer arithmetic while most/all other WCG sciences I think use mostly floating point, and the 64-bit BOINC integer benchmark is about 20% faster than for 32-bit BOINC (Intel) while the fp benchmarks are almost identical.