Can we beat Kasparov with a single virtual machine?
In June 1997, Deep Blue was the 259th most powerful supercomputer achieving 11.38 GFLOPS running 30 parallel RS/6000 SP Thin P2SC-based system nodes, with each node containing a 120 MHz P2SC microprocessor enhanced with 480 special purpose VLSI chess chips.
But how does that relate to computing power in 2011 and especially to virtual computing power running VMware vSphere 4.1?
An Intel Xeon X5680 is a popular processor nowadays, running 3,3 GHz on six processor cores. This Intel Xeon X5680 can achieve 80 GFLOPS (Floating Point Operations per Second) running all of its six cores. Taking into account a virtualization overhead of 7% (true, this is rather large) leaves 75 GFLOPS (93% of 80 GFlops). Which is 12.5 GFLOPS per processor core.
Deep Blue II was the 259th most powerful supercomputer in 1997 achieving 11.38 GFLOPS.
This means that we can use a single VMware virtual machine running a single vCPU to achieve the same amount of processing power. :-)
But let’s take this a little further.
When using the same Intel Xeon X5680 mentioned above with a four socket mainboard, we could achieve 300 GFLOPS per ESX host (4 x 75 GFLOPS). A VMware vSphere 4.1 cluster can contain a maximum of 32 hosts per cluster. When we multiply this we get 9.6 TFLOPS.
When we take this even further.
The fastest supercomputer in 2010 is the Tianhe-IA which is located in the National Supercomputing Center in Tianjin, China, which can achieve a stunning 2.566 PFLOPS.
This, in VMware vSphere 4.1, equals 268 clusters for which we need 30 data centers with 9 linked vCenter servers!
But no worries, when history repeats itself, we will achieve this computing power on a single vCPU virtual machine in 2024. :P
Just wait and see…..