Something I want to start working with is massively parallel computation, and the new Mac Pro would be a good candidate for that. Meanwhile I was reading an article about a former teacher of mine Andrew Ng about a paper he wrote on how you can build Googles $1M artificial brain on the cheap (see link above).
What they did is outfit a bunch of workstations with Nvidia graphics for $20,000.
It's not clear if they are saying the GPU's are 1TFLOP or the workstation is 1TFLOP.
At any rate the new mac pro they showed is at 7 TFLOPS I believe, making it quite capable for deep learning algorithms. If it comes in at around $6 then it will be quite cheap too.
What they did is outfit a bunch of workstations with Nvidia graphics for $20,000.
Quote:
|
Our cluster is comprised of 16 servers, each with 2 quad-core processors. Each server contains 4 NVIDIA GTX680 GPUs and an FDR Infiniband adapter. The GPUs each have 4GB of memory and are capable of performing about 1 TFLOPS (single-precision) with well-optimized code. |
At any rate the new mac pro they showed is at 7 TFLOPS I believe, making it quite capable for deep learning algorithms. If it comes in at around $6 then it will be quite cheap too.