Really?!
I was expecting a small percentile improvement from a little tweak here and there. An actual improvement that large is very significant. They did mention that the current code was derived from the GPGPU work, and I guess there's a bit of room to refine it to more SPE orientated. Well, then there ya go. I left last night, had a power outage came back today, and was set to download at the same time that my prior work unit had not yet uploaded... I had limited information with which to make a deduction!
I don't think my work unit started over, but I didn't look that closely. I mainly looked at the remaining time, before and after, and laughed at the Bullet-Time protein sequence. And its too late for me to tell for certain now.
So twice as fast?!
I don't think it's that strange. Perhaps they've added new functionality to support new projects.
Then is it possible they're sending out new projects that rely on lower resolution data (and thus faster processing)?
The only meaningful benchmark is points per day averaged over a reasonable timespan. The points are calibrated by Stanford, and since everyone's PS3 is "identical" in computing power, you can be sure that the points will be an accurate reflection of improved code.
Well IMO it could do with a swish broadcast of important and worthy info. It's free advertising for Sony!
That's good news. That means every PS3 working has the update, which means FP performance improvements will be reflected absolutely in the scores at Stanford when the average includesonly updated work. Any ideas when this will occur? Active PS3's return units in 5 days, so by then, all active PS3's will be updated. That should mean by this time next week, we get an average 1.1's performance. At the moment, from the current snapshot, PS3 is at 13 GFlops a unit (380 TFs / 29k PS3s). We should also see that average creeping up noticeably if performance has received a large boost.
