CPU counts, FLOPs, and corporate partnerships

One question that has come up recently is that the count of the number of CPUs involved in FAH has gone down quite a bit in the last 3 to 6 months.  There are several factors going on here.

We frequently negotiate periods of time to run on the machines in corporations, as this can bring a lot of power to FAH.  In most cases, the donations we get from large corporations are for pre-set lengths of time. These companies have some campaign or some period of availability. We’re constantly negotiating with companies for such partnerships and so the fluctuations in the number of computers may be quite great, especially as the number of single-core clients changes rapidly.

Also, in recent partnerships,  GPUs and large-core CPUs were emphasized over lots of single core CPUs and so the count of the number of machines will likely go down quite a bit but the FLOPs will remain quite high (today we’re still about 15 Petaflops, which makes FAH a very, very powerful supercomputer, especially when you consider that other traditional supercomputers would underperform their reported FLOP count if they used FAH’s approach at calculating FLOPs).  In the end, a high FLOP count has a big impact on our work.

With that said, it is very important for me to emphasize that we value all contributions.  Indeed,  we have in the past (and will continue to show in the future) shown the power of how many computers, even single core PCs, can band together to do amazing things.  Towards those ends, we have in particular been working on a new server backend software which can more efficiently utilize the results from many single-core CPU machines.

As I see it, the future of FAH will have to be heterogeneous and may even in the future incorporate mobile devices.  In the coming weeks, we hope to roll out our plans there more publicly, with some new initiatives on the CPU side which I think could greatly increase the power of FAH.