nemonemo

19 ms·

151.
by nemonemo 12y ago
This article contains good tips for building a GPU cluster with RDMA. One thing I would like to add is that there are two types of GPUDirect depending on CUDA versions. Previous CUDA supported GPUDirect through CPU memory, and now CUDA supp
152.
by nemonemo 13y ago
I think the fundamental reason of not being finisher may come from several reasons: lack of detailed planning (oh, there's an unanticipated obstacle), lack of energy to finish things, and little barrier of attention shift. I would say
153.
by nemonemo 13y ago
This is a very bold statement. IO-bound means that the bottleneck is at IO. GPU has little benefit in IO. Also, mapreduce is not embarrassingly parallel. Out of mapreduce, map is the only embarrassingly parallel phase. It has input parsing,
154.
by nemonemo 13y ago
GPU acceleration sounds all great.. until anyone actually generalizes the framework, implements applications, and measures even the most trivial workload. There should be some niche applications you may get 2-10x benefit, but 1000x is eithe
155.
by nemonemo 14y ago
Your comment conveys another implication that UN is supposed to be impartial to issues between member countries. That cannot explain veto power of five permanent members of UN security council, which no other country has. On consistency: I
156.
by nemonemo 14y ago
NK is aiming the weapon at US and its allies, unlike India or Pakistan who aim at each other. B.R. Myer's talk explains the motive behind NK's nuclear program pretty well. http://www.booktv.org/Watch/11315/The+Cleanest+Race+How+Nort...