Where can you buy a GPU (Nvidia Tesla, AMD Firestream)?

I want to make such a personal supercomputer, most likely will be doing it in the summer, has now decided to look for cars and light — a feeling that the GPU is almost sold out. More interesting AMD Firestream and on Ukrainian territory, but even in Russia something plohovato with it.

The problem is that queries such as "AMD Firestream buy", etc. derive most of the news. Someone from habrovky do all purchases of these devices and where can I find them?

p.s. I know you can also use "conventional" professional cards, but to drive fans without the need for graphical output as you do not want, and in the server to stick with no hands.
October 8th 19 at 03:00
6 answers
October 8th 19 at 03:02
And AMD 7XXX will not work?
Read the description, it's really interesting and also cheaper than what I'm looking for. The question is how these cards are designed for constant load. If they can work in hard mode a month or two or three? - Max_Baumbach16 commented on October 8th 19 at 03:05
You considered Bitcoin? :-) Actually these cards are the same as all the rest, prolonged stress to withstand should they focus on vychisleniya (OpenCL, DirectCompute, etc...). You read the specifications, you will be clearer. ;- ) they are the strongest on the market... - caterina.Emard61 commented on October 8th 19 at 03:08
well I just calculate and need, well, a month or two or three is a conditional I wrote. I would like to put in the server, the server in the data center for collocation and forget about what they are. And that thought constantly. :) Not to bticoinов, they are no longer counted, but I think a lot :) I Just watched the original NVIDIA Tesla, they are in relation to their so to say domestic counterparts are more advanced — reduced power consumption and at the same time not understated performance.

That's why I doubt it. - Max_Baumbach16 commented on October 8th 19 at 03:11
And correctly doubt the effectiveness of the playing cards in these calculations will be much less, and they burn for several months under constant load. - Jacinto_Oberbrunn commented on October 8th 19 at 03:14
About can burn the truth, but about efficiency — no. It all depends on the task. NVIDIA trims the speed doubles 4 times on the game card, but not everyone needs it. I like the GTX580 exactly 1.5 times faster than the Quadro 6000. And the last time commercials in 7-8 more. But I don't need doubles. - Cale.Lubowi commented on October 8th 19 at 03:17
October 8th 19 at 03:04
So with Amazon to order quite impossible? Then I remember recently HP servers for this case proud of the just, and you can have them immediately. Seems like it is the integrators in Ukraine who work with them.

It is possible a small question, what is considered a plan? Soft what your card supports and what works faster? And this choice can greatly depend.
If you do not reveal much of the card :), it is close to pattern recognition — you can play this way.

Software will be written with your based on the platform SDK and CUDA there may be OpenCL, so what else is there. Although now looked ru.wikipedia.org/wiki/OpenCV — at NVIDIA there is a contribution.

About HP servers now looked quite interesting. Though, probably, if separately collected, then will be cheap. :)

Another would be something adeseko find on the web. :) - Max_Baumbach16 commented on October 8th 19 at 03:07
Amazon is also an option — but I want a smaller one with a custom Tinker. - caterina.Emard61 commented on October 8th 19 at 03:10
With customs all the same will probably have to Tinker or you or the supplier. - Max_Baumbach16 commented on October 8th 19 at 03:13
If of software yet, then do not worry too much about the server, buy a couple of gaming vidyahu for the development of yeah write until soft and they testite inherently and optimize. Moreover, the next generation of NVIDIA is not far off. If you are lucky, you are all well raspravlyaetsya, and will fall on the iron, and is working, but if not lucky, the purchase of servers will be less headache:)
Before NVIDIA and ATI much different architecture used, now it seems ATI is moving in the right direction and their architecture is similar to NVIDIA's, but I have not played with them yet to be honest. Therefore, the choice of iron depends on how your algorithms it will earn. The difference can be more than one order for different cards... - Jacinto_Oberbrunn commented on October 8th 19 at 03:16
October 8th 19 at 03:06
For example Tesla can be bought for around 75 kilorubley we have, in St. Petersburg, a chain of stores known enough, the link will not give, for advertizing.
October 8th 19 at 03:08
Order or system integrators (they will take care of customs and everything, but they overpay 20-30%) or contact the local office of NVIDIA (like AMD — I do not know) and agree on delivery. Their phone in Moscow +7 495 981 03 00 work they are not very regularly, Tesla put for a long time (demand in Moscow they are almost no reason), maybe up to a month to go.
Thank you and all those who unsubscribed above! Tsacas as I see it there are 4 options: 1) it is the local office of the manufacturer (in Ukraine this is more difficult), 2) ready servers of the same type HP (as I saw it already more real, but it is not quite clear is there is already the card or separate purchase necessary), 3) via system integrators to order 4) order from Amazon or something like that.

Now I think, until summer there is still time.

In any case, if there are any thoughts from anyone else I will be glad to hear. - Max_Baumbach16 commented on October 8th 19 at 03:11
October 8th 19 at 03:10
Are you sure that you need a Tesla or something like that, you want the protein structure of the house calculate?))) Not the usual top some easier to take, if money is not critical, Yes, memory is smaller, but otherwise not worse.
>you want the protein structure of the house calculate
Something. :)

At the top vedaham I'm not sure that they are in the data center will not be covered through 3-4-5 months of constant work, yet Tesla and others more designed for this purpose. - Max_Baumbach16 commented on October 8th 19 at 03:13
Yuzaem three machines each with two 6870 as a poultice instead of the normal rent render farm. A couple of fucking days to write a normal translator to OpenCL not only for the viewport to drive, and normally all be considered in parallel with the CPU and everything is fine. The last four weeks working non-stop in normal household conditions, even without land — all right.

I think it's not critical. Tesla and Co. "more designed" just because they have support when used in similar projects, the prospects for the development of software and interfaces. - caterina.Emard61 commented on October 8th 19 at 03:16
October 8th 19 at 03:12
Have you considered the option of using ready-made server? For example:
Cluster Compute Instance on Amazon Web Services
( 2xNVIDIA Tesla M2050 "Fermi" GPUs., 2x quad-core Intel "Nehalem" X5570, 22 GB of RAM )
Hourly payment: 2.10$ per hour.
Advantages — you can start to use right now, scales as needed.
Disadvantages — price ( compared to a private server )

aws.amazon.com/hpc-applications/
The price really bites :) - Max_Baumbach16 commented on October 8th 19 at 03:15
How much will it cost your own server with similar features? Here Meijin.ru see supercomputers with the "locomotive" price. Heavy Utilization Reserved Instances — Quadruple Extra Large for 3 years will cost $10490 for 3 years and $0.494 per hour. - caterina.Emard61 commented on October 8th 19 at 03:18

Find more questions by tags E-commerceNVIDIAAMDProcessors