Will I be able to use my nVidia 8500gt to run GPU app?

and my nVidia 9600M GT ?

---

The problem is that only Geforce 2xx generation cards support the double precision that Milkyway@Home needs (all 'CUDA 1.3 capable' cards support it). I don't know if that can be worked around within the limits of CUDA, but it'd be a lot slower.

My 8500Gt runs "Cuda" already, so I guess thats a yes? Or do I need a "1.3 Cuda", if thats the case how do I find out if mine is 1.3 capable?

---

Here's a short list, as I guess there's no official one: http://setiweb.ssl.berkeley.edu/beta/forum_thread.php?id=1503&nowrap=true#36395

For GPU Grid the list of capable cards is different. The GPU Grid application takes a little more Oomph, and cards that can run SaH may not work well, or at all, on GPU Grid ...

Does the GPU Grid application support double precision? Because I know my card (9800GX2) can run it.

Does the GPU Grid application support double precision? Because I know my card (9800GX2) can run it.

The question is backwards... :)

The GPU Grid application does not need double precision. So, it can run on cards that only have single precision capabilities. The same is true with SaH ...

MW does much more double precision and so works best on hardware that does double precision fastest, which at the moment is ATI cards. Six months from now who knows ... though it is likely that it will stay this way for awhile ... Most applications don't do double precision and so will run fine on Nvidia cards assuming that the program can be vectorized ...

In fact, we just got a new student working with us, which her job being getting some GPU versions of milkyway@home available for you guys who really want to crunch some serious workunits :P

Travis, I'm very glad to hear that! I've determined that, once I get working again (currently unemployed), I should be able to afford something on the order of a 9600 GT. Maybe even two (my mobo is SLI). It would be really cool if she could make MW run on both cards in an SLI rig.

---

I have a 9600 Gt and it runs fine with the CUDA apps not that MW uses them but it runs nicely with GPU(PS3)grid
I just found that my laptop seems to be cuda able also not that i think it will be good for it :)
Boinc reports to see a Nvidia 8400M with 2 GFlops able to run 1.1 CUDA :)
I Guess that if it really would run its not gonna break any records :D

I have a 9600 Gt and it runs fine with the CUDA apps not that MW uses them but it runs nicely with GPU(PS3)grid
I just found that my laptop seems to be cuda able also not that i think it will be good for it :)
Boinc reports to see a Nvidia 8400M with 2 GFlops able to run 1.1 CUDA :)
I Guess that if it really would run its not gonna break any records :D

It would likely do better on SaH than GPU Grid ... :)

You mean the 9600 paul or the little 8400 ?

edit: omg lol a cuda SaH unit runs on the lappy :D

You mean the 9600 paul or the little 8400 ?

edit: omg lol a cuda SaH unit runs on the lappy :D

The 8400 ... I run a 9800GT and the 9600 (if I recall correctly) is not that far behind. YMMV ...

The only way to know for sure is to try I suppose. We don't have good speed measures of the GPUs and with such variety in clock rates, and the same models having different internals ... well ...

For my part, I want to replace the 9800GT and GTX280 with GTX295s or better ... That would get me almost 4 times the throughput of those two older cards.

Well to be honest i have got a 1 cuda unit on the 8400 but now it only gets the damn huge AP units which take too much time on that machine.

And i have no clue how to get Cuda units to load on it

Is MW not gonna make a Nvidia app as well ?

Well to be honest i have got a 1 cuda unit on the 8400 but now it only gets the damn huge AP units which take too much time on that machine.

And i have no clue how to get Cuda units to load on it

Is MW not gonna make a Nvidia app as well ?

In work, but, will only run on the very latest Nvidia GPUs because of the need for double precision ... and it will likely be slower because of the number of double precision units on the GPUs ... of course, with the 1060 Tesla that does not seem to be an issue.

hahah i dont know who is so rich he can buy these C1060 cards at $1,699.99
a piece just for running boinc, Btw i would not mind if someone donates me 3 of them ;)
if he/she does would have reach the max limit on computations every hour :D
i think untill boinc releases these limitations its not very usefull for others then scientists who can use it for complex models.
So for crunching its nice but for the points you should forget about them the max points is reached very fast with such power.
I also found machine downloads more other projects since MW seems to limit the workunits to 6 in the paste i got 48 units, but suddenly i only receive about 4 to 12 units.
Ofcourse the other projects thank MW for giving them more cpu time xD.

In work, but, will only run on the very latest Nvidia GPUs because of the need for double precision ... and it will likely be slower because of the number of double precision units on the GPUs ... of course, with the 1060 Tesla that does not seem to be an issue.

BUT will the Tesla be recognized by BOINC as a CUDA device with multiple GPUs, or as additional CPUs?

---

me@rescam.org

In work, but, will only run on the very latest Nvidia GPUs because of the need for double precision ... and it will likely be slower because of the number of double precision units on the GPUs ... of course, with the 1060 Tesla that does not seem to be an issue.

BUT will the Tesla be recognized by BOINC as a CUDA device with multiple GPUs, or as additional CPUs?

We don't seem to have good words to describe matters yet. As best as I can tell without buying one, the Tesla would be like a super GTX260/280 card. It would be seen as one GPU ... the issue is how fast would it improve matters. On SaH, with the limited tests I did, I saw no difference in speed between my 9800GT and my GTX280 card. Which raises the question in my mind if there is little to be gained by running more parallel units on their tasks (one of the main differences between the cards).

So, just having more stream units is of no avail if the task only needs 24... the rest are so much waste. Until someone does go out and get a tesla card we shall not know how much better it is than some other card. Again, for example, the GTX 280 is about 2-4 times faster on GPU Grid than the 9800GT (depending on the tasks and phases of the moon). Will the tesla be 4 times faster than the GT280?

At any rate, it would be seen as a GPU by BOINC, if it is seen at all ... :)

A good question for the Nvidia forums...

As to who might buy one? Well, I would ... Why not ... much cheaper than many upgrades I have done ...

I won't tell you how much my Mac Pro cost, only that my wife is happy when I only spend about 4K on a new machine ... she thinks that that is a bargain ...

I can easily see that a "good" configuration, were it workable, would be to have something like a GTX295, a pair of Teslas and a HD4890(2) in each system ...

But, to know what the best configuration would be we need some time to see what the various applications look like and how they perform on various configurations. When my SaH RAC dies and I get the urge I will be re-running the test to see the difference between the run speeds of the GTX280 and the 9800GT for some tasks ...

But, to know what the best configuration would be we need some time to see what the various applications look like and how they perform on various configurations. When my SaH RAC dies and I get the urge I will be re-running the test to see the difference between the run speeds of the GTX280 and the 9800GT for some tasks ...

If you want to see how a (EVGA) 9800GT runs just look at my tasks. My app_info will pass all WU versions to the GPU.

---

me@rescam.org

Well according to those who made that thing its 933 GFLOPS
My 9600 GT seems to give 38 GFLOPS but as paul stated its kinda difficult to give an impression what can be done with it

Since it has 4 GB ram at least but if only 1 application working on 1 unit would be done its gonna work at his max capacity

If anyone can make some kinda multi app which runs many units on the available processors depending on the memory then it starts to be more usefull

Nevertheless i expect it to be very fast in computations if those applications are memory demanding and/or use huge processing power.

Even better would be the combination of all cpu/gpu/tesla

Again the limits which are made by boinc and projects need to be stretched up.
Because i allready hear people complaining who are lucky owners of a 295 or more cards that these cards does not gain more credits because it bumps that limits all the time.

If this tesla cards give so much power more its obvious that no gain will be made then

Because i allready hear people complaining who are lucky owners of a 295 or more cards that these cards does not gain more credits because it bumps that limits all the time.

No limits that I have seen on GPU Grid. What limits are you talking about?