Holy cow…I am definitely going to get one as soon as I get some money together for one. I was getting ready to buy a GTX260 or GTX280 to do some double precision work…but at 50% off I don’t think I can pass up a C1060!
to me it’s not whether it’s “50% off” but whether it’s a good deal. This completely depends on if you need the 4gb of memory. Otherwise, it looks like it’s still around $500-600, and two gtx 285 cards would probably be better… if you have that sort of money. I’m happy with the gtx 260… I need to do a lot more optimization before I can justify a new card.
That’s very true. It’s really a toss-up between the GTX295 and the C1060, since with the discount they are close to the same price…but I’ve had some GPU programs in mind that will be very memory-intensive, so that’s where the Tesla will come in handy. If anything, I may even get a GTX260 as well, so I can test out some multi-GPU apps as well.
There’s only a few things stopping me from getting one right now, the most important of which is that my current dev. machine only has one PCIe 1.0 x16 slot in it, so there’s no room for the Tesla. That, and I’m holding out a bit to see what news I hear about PCI Express 3.0 (which is supposedly due towards the end of the year).
My original 8800 GTX development machine (built 2 years ago) has been turned into a 24/7 job running machine. I’ve got ~4 months of contiguous run time on it now without a single issue. Although, I did have some random crashes in the very beginning that turned out to be caused by overheating. A small case modification to get more airflow fixed that.
The biggest issues with a cluster of GTX 295s are 1) cooling, 2) power supply and 3) mechanics (fitting them in the case). If you can solve all three (which it sounds like you have) you shouldn’t have any other trouble.
Just curious, but why wansn’t S1070 an option (besides cost)? It puts 2 GPUs on each PCIe connection so you can build an 8 GPU box if you want to (somebody correct me if this is not a supported configuration). And it solves issues 1,2, and 3 above while keeping it all packed in a very small rackmount case, making it trivial to deploy dozens of them.
The GTX295 has two GPU’s per board, while the C1060 only has one. However, the Tesla has 4GB of on-board RAM while the GTX295 only has 896MB per GPU (on the currently available boards). The extra RAM is the difference…for some applications, allowing the developers to keep data in RAM rather than transferring it back and forth is a huge deal. You’ll have to decide whether it is worth the cost (and half the GPU’s per box) for your applications.