Hi everyone, im new to this forum and i'd like to start off with a small discussion on distributed cluster GPU computing. Lets all imagine that Joe has 5 Microway Octoputer Systems with 8 AMD Firestream 9370 (2.65 TFLOP) Cards inside (With the dual 8 core cpus it puts out about 25ish Tflop/s). How can joe harness all this power into pure brute force password munching. What kind of cluster does he has to run in order to utilize his hardware to the max. I was also wondering if joe has another 5 Dell r620 systems with dual amd 9270 (about 1Tflop/ea) and dual Xeon E5-2960 (8 Cores/Ea), can he include those servers in his cluster? Presuming that each node has 4 1GbE links to the core switch. Can joe use different nodes for the same application? What server joe needs to manage the cluster? What kind of cluster does joe need? Will 4 GbE links suffice for the job? Thanks in advance...
You should write an application based on the SDK from BOINC cloud computing. You create a feeder and all the rest are computing nodes. Scallability to the maximum.