Printable View
Yeah, SuperMicro, Tyan, and ASRock are three of the few companies I know of that sell a lot of the same stuff retail as they do to big customers.
Haha, we'll probably be getting rid of some in a couple years, actually...but only if you want to buy 10-20 racks at a time... :whistler:Quote:
Originally Posted by Konrad
In the meantime, you can always rent one for $0.65/hour/GPU. ;)
Damn, $5700 per year for each GPU?
Aw, no thanx lol. Even a pair of $1300 Titan X cards is already pretty excessive for my needs, in price and in performance. I know people who spend less buying cars then I did on computers, but I still can't resist great old GPUs selling for ten bucks each, lol.
I am one of those people who spent more on my computer than my transportation (motorcycles). Kind of puts a crimp in the computer budget to be honest.
I have a couple of Quadro FX 4800's lying around now....
How often are you actually using those GPUs' power though? Let's say for the sake of argument that the Titan X offers the same performance as one of those instances (talking GPGPU or other stuff that can be done remotely; obviously gaming/etc is not yet a feasible application for these). I don't remember what model card they use, but it's likely to be in the same rough ballpark. If you use the card less than ~23% of the year (that's total time, not just time you're awake), then it would be cheaper to use the EC2 instance.Quote:
Originally Posted by Konrad
Think of it less as "this would cost me $[X] per year with it running the whole time" and more as "what would it cost if I only paid for this when I actually used it".
lol, I lust after a Titan X pair, but I have to make do with "only" a pair of 980 Classifieds. About $900 each (with backplates), but I think I'm getting great value out of them, and I expect to continue getting value out of them for at least a couple more years.
I do CAD crunch/render/sim stuff enough to want the best I can justify buying, lol. Hate those frustrating little pauses (seconds, minutes, even hours) every single time I execute even the most trivial little schema or syntax changes. I game a bit, not all that much, and fps isn't really my thing unless I can find games which have particularly engaging single-player stories/environments or unless I can find some not-immature multi-player opponents with roughly the same "skill" level I possess. For me, the most appealing feature of the Titan X is the 12GB onboard VRAM, mainly because that seems to be where my GPGPU bottlenecks occur the most. I'm an oddball since I game a little, I compute a lot, and I don't really need the DPFP or godlike support/warranty which define value for workstation GPUs. It's handy to "have it all" in one machine, but I'm just as content to work (or game) one one while another crunches away at some other work task.