I want to experiment with DL (specifically TensorFlow) to see if I can gain any traction on some computational biology problems. At first I'm hoping to use one of my current machines -- but they have PCI 2.0, and so I assume critically limited long term. No point in getting a card that will be limited by my PCI bottleneck, but I really have no idea how or with what card that will occur. If I see promise I would get a newer machine, but that would probably be 6 months or so from now.
My datasets, once prepped into features, aren't huge by today's standards -- eg 50K samples, maybe 1000 features. Might do useful experiments with smaller sets. I'm guessing the size of the datasets are the critical issue for slow PCI, but I don't know that.
Card could go in either of these machines, if compatible:
1) Supermicro X8DAH+ mobo, dual Xeon 5690, 192 GB RAM. My primary workstation that I do feature prep on, fair bit of parallel code, and currently run SVM machine learning.
2) Dell Precision T7500, dual Xeon 5650, 96 GB RAM. Mostly a database server, probably best here.
I've been looking at the GTX 1000 line (upper limit would be 1070 at this point) if PCI bus wasn't a deal breaker, but there may be something else I don't know about that would suggest an older card. At least a CUDA compatibility of 3.5 needed.
Thanks in advance for any recommendations or thoughts.
UPDATE: FWI, Currently Supermicro has a GeForce GT 220 and the Dell an AMD Bonaire XTX Radeon R7 260X or 360(?)
My datasets, once prepped into features, aren't huge by today's standards -- eg 50K samples, maybe 1000 features. Might do useful experiments with smaller sets. I'm guessing the size of the datasets are the critical issue for slow PCI, but I don't know that.
Card could go in either of these machines, if compatible:
1) Supermicro X8DAH+ mobo, dual Xeon 5690, 192 GB RAM. My primary workstation that I do feature prep on, fair bit of parallel code, and currently run SVM machine learning.
2) Dell Precision T7500, dual Xeon 5650, 96 GB RAM. Mostly a database server, probably best here.
I've been looking at the GTX 1000 line (upper limit would be 1070 at this point) if PCI bus wasn't a deal breaker, but there may be something else I don't know about that would suggest an older card. At least a CUDA compatibility of 3.5 needed.
Thanks in advance for any recommendations or thoughts.
UPDATE: FWI, Currently Supermicro has a GeForce GT 220 and the Dell an AMD Bonaire XTX Radeon R7 260X or 360(?)