3 ms·

My understanding: The D-Wave machine isn't general quantum computing, but hardware supporting one specific algorithm: quantum annealing. Quantum annealing is si

My understanding: The D-Wave machine isn't general quantum computing, but hardware supporting one specific algorithm: quantum annealing. Quantum annealing is similar to Simulated annealing, and if you compare the D-Wave machine to simulated annealing implemented on normal, non specialized, hardware, the D-Wave is 100,000,000 times faster. In principal, even if the hardware only supports one algorithm, this could tell us something about whether quantum computers are really better than normal computers. Unfortunately the D-Wave machine/algorithm is still not much better than simply simulating a quantum machine in normal hardware (Quantum Monte Carlo) and worse than the conventional algorithm by Alex Selby. It's interesting, but it still doesn't shed much light on the important questions.

InvisibleCities 11y ago
One more thing that you didn't mention, which I think is important: if you run simulated annealing on specialized hardware and compare it to the performance of D-Wave's machine, there is virtually no speedup for D-Wave's machine.
nickpsecurity 11y ago
Sort. A counterpoint I read on this topic indicated it might be rigged a bit. The comparison was to an emulation of D-Wave's own annealing process on a single-CPU machine. That's... already an unfair representation of simulated annealing's performance on classical hardware. I'd compare the solution of a problem to optimized algorithms on NUMA machines, GPU clusters, and FPGA clusters that cost whatever D-Wave costs. Given that's all one knows about how it really works past their claims: input X dollars to get Y performance and Z energy-use. ;) Note: One could also compare to performance of single machine with CPU's, GPU's, FPGA's, and/or ASIC's to be more fair. Just needs to be optimal, classical implementation on HW designed for it vs optimal use of D-Wave.

InvisibleCities 11y ago

One more thing that you didn't mention, which I think is important: if you run simulated annealing on specialized hardware and compare it to the performance of D-Wave's machine, there is virtually no speedup for D-Wave's machine.

nickpsecurity 11y ago

Sort. A counterpoint I read on this topic indicated it might be rigged a bit. The comparison was to an emulation of D-Wave's own annealing process on a single-CPU machine. That's... already an unfair representation of simulated annealing's performance on classical hardware. I'd compare the solution of a problem to optimized algorithms on NUMA machines, GPU clusters, and FPGA clusters that cost whatever D-Wave costs. Given that's all one knows about how it really works past their claims: input X dollars to get Y performance and Z energy-use. ;) Note: One could also compare to performance of single machine with CPU's, GPU's, FPGA's, and/or ASIC's to be more fair. Just needs to be optimal, classical implementation on HW designed for it vs optimal use of D-Wave.