Can you fold/boinc/mine with a Xeon Phi?

[FruitBasketSilex]

Title, yo.

[Guest SkyFall]

As far as I know, none of these applications support xeon phi's.

[FruitBasketSilex]

Looks like we've already got contradicting answers. Yes, and no.

[FruitBasketSilex]

I would consider buying a two or three hundred dollar one to fold with, and just to have, if it would work.

[Guest SkyFall]

Title, yo.

 

I think that GPUs would be more cost- and power-efficient anyways.

[ShadowCaptain]

I would consider buying a two or three hundred dollar one to fold with, and just to have, if it would work.

 

Thats a lot of money just for folding, you would be better off donating the money for the processors and the energy used to fold on them to charity every month

 

AFAIK you cannot fold on them, but not 100%

[FruitBasketSilex]

I think that GPUs would be more cost- and power-efficient anyways.

60 cores, 1.2 jiggahertz 300w tdp 200$-300$

Vs.

R9 390x, unknown amount of cores, 1.1 jiggahertz 275w tdp 400$

Not even gonna take the time to try to look up Xeon Phi power consumption.

[Guest SkyFall]

60 cores, 1.2 jiggahertz 300w tdp 200$-300$

Vs.

R9 390x, unknown amount of cores, 1.1 jiggahertz 275w tdp 400$

Not even gonna take the time to try to look up Xeon Phi power consumption.

 

But the GPU can be used for others things, not only for calculating in a scientific scale.

Anyways, there's no real use for them if you're not an university which actually has software to run on these things.

 

 

Anyways, we are getting off topic...

[Madone6]

Title, yo.

nVidia Teslas and Xeon Phis don't work

[FruitBasketSilex]

nVidia Teslas and Xeon Phis don't work

Damnit.

EDIT: all to hell

[Tinkerfan]

Phi cards would maybe be amazing folders, owing to the co-processor architecture being able to potentially support crazy muli-threading. however, they aren't directly supported on either side, so you probably at the least wouldn't be able to take full advantage of their power. THEREFORE, because Intel only claims a peak processing power on the highest end cards of 1.5 ish teraflops, compred to the GTX980's 5 ish teraflops. So I'd say that while you can probably fudge things and get it to work, it is definitely not an optimal situation.

[Imakuni]

Damnit.

EDIT: all to hell

Wrong.

 

As far as the Boinc Platform goes, Xeon Phi and Teslas ARE supported. In fact, any single piece of hardware that can be recognized by your computer can work with boinc, by specifying it's features via app_info. If you were to get your hands on a secret Gtx Pascal, for example, you could very well make a file saying "Boinc, I have this co-processor (pascal) in my PC, it can do this this and that, and there's this application I'd like you to try and use it with". Assuming you configured it right, congratulations, you just added Boinc support for the GPU while you wait the official release...........

 

.....HOWEVER, that's not to say that the projects actually do. For example, Intel iGPUs: taking Primegrid's OpenCL codes (which, btw, is the same for Nvidia and AMD GPUs) actually runs on HD Graphics..... yet the server will never send tasks to the iGPU. Why? Because it was found that it was too slow, and just using the CPU core without the iGPU was faster. So even though the support is there, you won't get tasks for it. Alternatively, Boinc has a bug with newer r9 390x / Fury cards, saying they aren't capable of crunching; BUT, you can manually force client to run the app, and it'll work just fine, even though the support "isn't there".

 

 

"Allright, enought blabling: what about Teslas / Xeon Phi?"

 

Well, Teslas are easy. Once again, using Primegrid's software as an example, if your Tesla has CUDA / OpenCL of the appropriate version, it can get tasks; if I recall correctly, you don't even need to do that weirdness of adding manual support, it works right out of the box. You need to install drivers directly from nvidia (as Windows' drivers lacks a few needed features) and their performance isn't very good.... but at least they are fully supported. I don't know if other projects have support for it, but I know at least this one does.

 

Xeon Phis, on the other hand, are more tricky. For starters, you'd NEED to add manual support on Boinc's side (unlike the Teslas). Even then, I don't think there's a project with actual support for the thing, as very few developers / crunchers even have access to such cards. Once again, you could try a "compile the app and add support yourself" approach, but as far as out of the box support goes, the answer is "no".

[FruitBasketSilex]

Wrong.

As far as the Boinc Platform goes, Xeon Phi and Teslas ARE supported. In fact, any single piece of hardware that can be recognized by your computer can work with boinc, by specifying it's features via app_info. If you were to get your hands on a secret Gtx Pascal, for example, you could very well make a file saying "Boinc, I have this co-processor (pascal) in my PC, it can do this this and that, and there's this application I'd like you to try and use it with". Assuming you configured it right, congratulations, you just added Boinc support for the GPU while you wait the official release...........

.....HOWEVER, that's not to say that the projects actually do. For example, Intel iGPUs: taking Primegrid's OpenCL codes (which, btw, is the same for Nvidia and AMD GPUs) actually runs on HD Graphics..... yet the server will never send tasks to the iGPU. Why? Because it was found that it was too slow, and just using the CPU core without the iGPU was faster. So even though the support is there, you won't get tasks for it. Alternatively, Boinc has a bug with newer r9 390x / Fury cards, saying they aren't capable of crunching; BUT, you can manually force client to run the app, and it'll work just fine, even though the support "isn't there".

"Allright, enought blabling: what about Teslas / Xeon Phi?"

Well, Teslas are easy. Once again, using Primegrid's software as an example, if your Tesla has CUDA / OpenCL of the appropriate version, it can get tasks; if I recall correctly, you don't even need to do that weirdness of adding manual support, it works right out of the box. You need to install drivers directly from nvidia (as Windows' drivers lacks a few needed features) and their performance isn't very good.... but at least they are fully supported. I don't know if other projects have support for it, but I know at least this one does.

Xeon Phis, on the other hand, are more tricky. For starters, you'd NEED to add manual support on Boinc's side (unlike the Teslas). Even then, I don't think there's a project with actual support for the thing, as very few developers / crunchers even have access to such cards. Once again, you could try a "compile the app and add support yourself" approach, but as far as out of the box support goes, the answer is "no".

[Madone6]

Wrong.

 

As far as the Boinc Platform goes, Xeon Phi and Teslas ARE supported. In fact, any single piece of hardware that can be recognized by your computer can work with boinc, by specifying it's features via app_info. If you were to get your hands on a secret Gtx Pascal, for example, you could very well make a file saying "Boinc, I have this co-processor (pascal) in my PC, it can do this this and that, and there's this application I'd like you to try and use it with". Assuming you configured it right, congratulations, you just added Boinc support for the GPU while you wait the official release...........

 

.....HOWEVER, that's not to say that the projects actually do. For example, Intel iGPUs: taking Primegrid's OpenCL codes (which, btw, is the same for Nvidia and AMD GPUs) actually runs on HD Graphics..... yet the server will never send tasks to the iGPU. Why? Because it was found that it was too slow, and just using the CPU core without the iGPU was faster. So even though the support is there, you won't get tasks for it. Alternatively, Boinc has a bug with newer r9 390x / Fury cards, saying they aren't capable of crunching; BUT, you can manually force client to run the app, and it'll work just fine, even though the support "isn't there".

 

 

"Allright, enought blabling: what about Teslas / Xeon Phi?"

 

Well, Teslas are easy. Once again, using Primegrid's software as an example, if your Tesla has CUDA / OpenCL of the appropriate version, it can get tasks; if I recall correctly, you don't even need to do that weirdness of adding manual support, it works right out of the box. You need to install drivers directly from nvidia (as Windows' drivers lacks a few needed features) and their performance isn't very good.... but at least they are fully supported. I don't know if other projects have support for it, but I know at least this one does.

 

Xeon Phis, on the other hand, are more tricky. For starters, you'd NEED to add manual support on Boinc's side (unlike the Teslas). Even then, I don't think there's a project with actual support for the thing, as very few developers / crunchers even have access to such cards. Once again, you could try a "compile the app and add support yourself" approach, but as far as out of the box support goes, the answer is "no".

I'm sorry, I was talking about FAH not BOINC