If you've ever wanted to play with a Xeon Phi card, here's your chance

[Radium_Angel]

https://www.ebay.com/itm/Intel-Xeon-Phi-5110P-60-Core-1-053GHz-Coprocessor-C1P87A-708360-001-225w-PCIe/123686943549?_trkparms=aid%3D111001%26algo%3DREC.SEED%26ao%3D1%26asc%3D20160908105057%26meid%3D45d3bbc2fb294985affc92f543d30610%26pid%3D100675%26rk%3D4%26rkt%3D15%26mehot%3Dpp%26sd%3D174438122583%26itm%3D123686943549%26pmt%3D1%26noa%3D1%26pg%3D2380057%26brand%3DIntel&_trksid=p2380057.c100675.m4236&_trkparms=pageci%3Adc10d67f-51ba-11eb-bb74-724c434b27c6|parentrq%3Ae252f5221760ac3cd79bb828ffed0ff8|iid%3A1

 

Less than 40$ US

Now, you can't just drop this into a Win10 system and suddenly see all the extra cores, but you can, sort of.

https://www.pugetsystems.com/labs/hpc/Intel-Xeon-Phi-with-Windows-510/

and the driver you'll need for Windows to "See" it

https://software.intel.com/content/www/us/en/develop/articles/intel-manycore-platform-software-stack-mpss.html?wapkw=mpss#wn38rel

 

I've been considering one, just to own a piece of history. I'd love to get a gaggle of programmers together to make some kind of "man in the middle" program/passthrough, to utilize all those cores for less esoteric use.

[WhitetailAni]

1 minute ago, Radium_Angel said:

https://www.ebay.com/itm/Intel-Xeon-Phi-5110P-60-Core-1-053GHz-Coprocessor-C1P87A-708360-001-225w-PCIe/123686943549?_trkparms=aid%3D111001%26algo%3DREC.SEED%26ao%3D1%26asc%3D20160908105057%26meid%3D45d3bbc2fb294985affc92f543d30610%26pid%3D100675%26rk%3D4%26rkt%3D15%26mehot%3Dpp%26sd%3D174438122583%26itm%3D123686943549%26pmt%3D1%26noa%3D1%26pg%3D2380057%26brand%3DIntel&_trksid=p2380057.c100675.m4236&_trkparms=pageci%3Adc10d67f-51ba-11eb-bb74-724c434b27c6|parentrq%3Ae252f5221760ac3cd79bb828ffed0ff8|iid%3A1

 

Less than 40$ US

Now, you can't just drop this into a Win10 system and suddenly see all the extra cores, but you can, sort of.

https://www.pugetsystems.com/labs/hpc/Intel-Xeon-Phi-with-Windows-510/

and the driver you'll need for Windows to "See" it

https://software.intel.com/content/www/us/en/develop/articles/intel-manycore-platform-software-stack-mpss.html?wapkw=mpss#wn38rel

 

I've been considering one, just to own a piece of history. I'd love to get a gaggle of programmers together to make some kind of "man in the middle" program/passthrough, to utilize all those cores for less esoteric use.

Larrabee II within my reach.

Except it isn't...
Oh well.
Think it will work on an AMD system?

[Radium_Angel]

Just now, ragnarok0273 said:

Larrabee II within my reach.

Except it isn't...
Oh well.
Think it will work on an AMD system?

Well, it's not as straightforward as that, it's a many core computing unit with its own built in uOS, which is why you can't just drop it and and get it up and running like a normal device.

But the deal was too good to pass up, so I just snagged one myself, even if I never use it properly, it's a piece of computing history.

[jaslion]

I remember seeing these pop up quickly in 2018 with someone trying to make a render farm with these (or a similar add in card) it worked but a ryzen system was comparable.

[WhitetailAni]

1 minute ago, Radium_Angel said:

Well, it's not as straightforward as that, it's a many core computing unit with its own built in uOS, which is why you can't just drop it and and get it up and running like a normal device.

But the deal was too good to pass up, so I just snagged one myself, even if I never use it properly, it's a piece of computing history.

If you're ever cleaning things out, I'm happy to take it off your hands

[Radium_Angel]

2 minutes ago, ragnarok0273 said:

If you're ever cleaning things out, I'm happy to take it off your hands

I tend to give away tons of hardware from time to time, I'll keep that in mind. I just gave away 2 laptops, 3 desktops and a pair of servers because I didn't need them taking up so much space.

4 minutes ago, jaslion said:

I remember seeing these pop up quickly in 2018 with someone trying to make a render farm with these (or a similar add in card) it worked but a ryzen system was comparable.

Oh no doubt, these are "merely" 1Ghz in core speed, so yeah, easily beaten, but there are 60 cores, which, if the right middleware ever came out, could really help do some heavy listing on an existing system, especially if the main system was older, like an Ivy-Bridge Xeon workstation.

[jaslion]

1 minute ago, Radium_Angel said:

I tend to give away tons of hardware from time to time, I'll keep that in mind. I just gave away 2 laptops, 3 desktops and a pair of servers because I didn't need them taking up so much space.

Oh no doubt, these are "merely" 1Ghz in core speed, so yeah, easily beaten, but there are 60 cores, which, if the right middleware ever came out, could really help do some heavy listing on an existing system, especially if the main system was older, like an Ivy-Bridge Xeon workstation.

The main problem is that they function with the MIMD rendering protocol instead of the SIMT one which every renderer uses because it expects to just run on a normal cpu it can directly interface with and not through a api layer. There was something they did in opencl that allowed them to feed every single core one part of the render and use a common knowledge pool which lead to a massive leap in performance but like a ryzen 1600 matched it.

[Tristerin]

F

 

I just spent money yesterday on some hardware.  I dont want to explain this purchase to the wife lol.  I still need a new drive ordered and will have to explain that (need moar nvme)

 

Tempting, its only $40.

[Radium_Angel]

Just now, Tristerin said:

its only $40.

That's just lunch money, take her out to a nice place for dinner (or order in, depending on COVID where you are) and you'll be fine.

1 minute ago, jaslion said:

The main problem is that they function with the MIMD rendering protocol instead of the SIMT one which every renderer uses because it expects to just run on a normal cpu it can directly interface with and not through a api layer.

Yeah, these are not general purpose cards, they never were, but they're cheap enough (now) to maybe bang on and do something with...maybe.

[Tristerin]

Im going to call my brother rq (hes on vacation in CO so may not answer) and ask him about these for mining.  60 slow cores on low level coins may yield results worth keeping.

 

@Radium_Angel

[Tristerin]

@Radium_Angel nope thats a hard pass on mining, but I may still get one

[Radium_Angel]

13 minutes ago, Tristerin said:

@Radium_Angel nope thats a hard pass on mining, but I may still get one

Yeah they won't mine.

[8tg]

IMO if you want to spend 40$ on some dumb hardware, buy a fermi Tesla and game on a Tesla.

 

[Radium_Angel]

6 minutes ago, 8tg said:

IMO if you want to spend 40$ on some dumb hardware, buy a fermi Tesla and game on a Tesla.

 

Assuming you game, that is...

[8tg]

5 minutes ago, Radium_Angel said:

Assuming you game, that is...

I could argue that in almost any regard, a cheap housefire Tesla with a dvi port is more useful than a Xeon phi for anyone.

They’re cool but the list of things they can do without going full diy is like 3 lines long.

[Radium_Angel]

2 minutes ago, 8tg said:

I could argue that in almost any regard, a cheap housefire Tesla with a dvi port is more useful than a Xeon phi for anyone.

They’re cool but the list of things they can do without going full diy is like 3 lines long.

So, on a K40 (assuming that is the cheap Fermi card you are talking about) how much of a PITA is it to make it functional as a "gaming" card, and what kind of performance could be expected.

Failing that, what would one do with a K40 otherwise?

[8tg]

21 minutes ago, Radium_Angel said:

So, on a K40 (assuming that is the cheap Fermi card you are talking about) how much of a PITA is it to make it functional as a "gaming" card, and what kind of performance could be expected.

Failing that, what would one do with a K40 otherwise?

C2050/2070/2090

they have a dvi port and in their nvidia driver settings is a toggle for compute or general use

then it’s a gtx 470/480

[Radium_Angel]

5 minutes ago, 8tg said:

then it’s a gtx 470/480

Yikes, thanks, I'll stick the Phi

[8tg]

Just now, Radium_Angel said:

Yikes, thanks, I'll stick the Phi

I too like the compute performance of an 8600GT with little to no off the shelf software to utilize it 

[Radium_Angel]

1 minute ago, 8tg said:

I too like the compute performance of an 8600GT with little to no off the shelf software to utilize it 

That's half the fun!

[8tg]

Just now, Radium_Angel said:

That's half the fun!

I get that, pardon my snide comments. If that’s what you want to do with the hardware, then yeah it’s probably pretty neat to play with a phi.

But for a more generalized audience who needs some kind of gpu compute boost, I would suggest a cheap fermi Tesla. Since they are basically just consumer cards with (relatively) large quantities of ECC memory and decent compute performance, but they’re also capable of regular desktop or workstation use.

Its like if you took a Quadro type of card and gave it some Tesla features, that’s what they are.

And that I would argue is more useful for a hobbyist on a budget, or someone who just wants to fiddle with a cheap gpu.

 

My C2050 is good for that, helping with autodesk airflow simulations substantially over an equivalent Quadro or gtx 470. But then this same card also isn’t going to struggle rendering a scene, or displaying said scene smoothly. Very multipurpose.

[whm1974]

Didn't this Tech from Intel went Belly Up fairly soon after it was released? As I recall, very few Users could even get even benefit from using this Card at all. Didn't his Require Special Applications to make use of? 

[Radium_Angel]

1 minute ago, 8tg said:

I get that, pardon my snide comments. If that’s what you want to do with the hardware, then yeah it’s probably pretty neat to play with a phi.

But for a more generalized audience who needs some kind of gpu compute boost, I would suggest a cheap fermi Tesla. Since they are basically just consumer cards with (relatively) large quantities of ECC memory and decent compute performance, but they’re also capable of regular desktop or workstation use.

Its like if you took a Quadro type of card and gave it some Tesla features, that’s what they are.

And that I would argue is more useful for a hobbyist on a budget, or someone who just wants to fiddle with a cheap gpu.

 

My C2050 is good for that, helping with autodesk airflow simulations substantially over an equivalent Quadro or gtx 470. But then this same card also isn’t going to struggle rendering a scene, or displaying said scene smoothly. Very multipurpose.

Apples to oranges, the Fermi cards are GPU rendering, the Phi is....not.

I totally get your point, which is why the original topic was "IF" not, "YOU HAVE TO NOW!" as I'm not certain what I am going to do with this, except own a part of computer history.

[Radium_Angel]

12 minutes ago, whm1974 said:

Didn't this Tech from Intel went Belly Up fairly soon after it was released? As I recall, very few Users could even get even benefit from using this Card at all. Didn't his Require Special Applications to make use of? 

Correct, has a shelf life of about 13 years (2020 was the kill date) and yes, it's for very specialized purposes, for a very narrow field of science.

But they were (IIRC) 6k$ when new, so here's a chance to own some serious (if highly specialized) firepower for pennies

[whm1974]

6 minutes ago, Radium_Angel said:

Correct, has a shelf life of about 13 years (202 was the kill date) and yes, it's for very specialized purposes, for a very narrow field of science.

But they were (IIRC) 6k$ when new, so here's a chance to own some serious (if highly specialized) firepower for pennies

Reading Wikipedia, it seems you could you use for General Purpose Applications if you wanted to. However since the Core were P5 Pentium based with modern Instructions added. And at higher clockspeeds then way back then.

 

But it wouldn't be worth the trouble, due to Reqular Software can't make use of that many cores to begin with. And the Expense of ~$6,000...