I'd love to know how to setup an FPGA with the raspberry pi so I can finally get a really cheap and low energy PC for getwork tasks while delegating the actual mining to the FPGA.I'd like a clear walkthrough with pics if someone has managed as well as which OS they used on the Raspberry Pi.
Topic: Raspberry Pi $25 PC - Could we run GPUs/FPGAs on this? (Read 14947 times)
I have experience running large amounts of FPGAs on what might be called an underpowered machine. I wouldn't plan on hosting more than five boards with this, but that is a very rough estimate.
If you don't want to wait to try it. I'm selling my freshly arrived pi here: https://bitmit.net/en/trade/i/2481-raspberry-pi-model-b/description
So if we are looking for trouble why not use a OpenWRT supporting router with USB port to run the FPGAs? haha...
Done
So if we are looking for trouble why not use a OpenWRT supporting router with USB port to run the FPGAs? haha...
Ok, ima get one of these, but i dont have a fgpa to hook it upto for testing, :/
That's OK, I do
By the time you get your Pi delivered you may as well (it'll be a little while yet).
Ok, ima get one of these, but i dont have a fgpa to hook it upto for testing, :/
Where do you guys get 6970 gets 683 GFLOPS.
6970 has 2.7 TFLOPS of floating point computational power and a 7970 is ~3.8 TFLOPS.
It's the double precision compute power6970 has 2.7 TFLOPS of floating point computational power and a 7970 is ~3.8 TFLOPS.
6970
2.7 TFLOPs Single Precision compute power
683 GFLOPs Double Precision compute power
Where do you guys get 6970 gets 683 GFLOPS.
6970 has 2.7 TFLOPS of floating point computational power and a 7970 is ~3.8 TFLOPS.
It's the double precision compute power6970 has 2.7 TFLOPS of floating point computational power and a 7970 is ~3.8 TFLOPS.
6970
2.7 TFLOPs Single Precision compute power
683 GFLOPs Double Precision compute power
Type B has 2 usb ports, type A only one.
The dubble usb port count is also worth something
The dubble usb port count is also worth something
Good point. I'm not saying it's impossible, but it's probably not worth the hassle to save $6.
I see you need the model B, $35 to get ethernet.
Not really. You can get a USB ethernet adapter for less than $4, or a USB wifi adapter for $8.
You got drivers to support that on the ARM platform?
I see you need the model B, $35 to get ethernet.
Not really. You can get a USB ethernet adapter for less than $4, or a USB wifi adapter for $8.
|
|
Dropping the extra $10 to get it built in might give you higher quality components though.
Where do you guys get 6970 gets 683 GFLOPS.
6970 has 2.7 TFLOPS of floating point computational power and a 7970 is ~3.8 TFLOPS.
6970 has 2.7 TFLOPS of floating point computational power and a 7970 is ~3.8 TFLOPS.
WOW. Did not realize the difference between a 5870 and 7970 was that big.
Now that you look at the TFLOPS and MHash/s it really does look like the 7970 is almost 1.7 times faster than a 5870 due to the much better overclocking ability of 28nm GPUs.
Yeah AMD integer performance took a nose dive with the new architecture. The 7970 simply puts up respectable numbers by brute force.
Where do you guys get 6970 gets 683 GFLOPS.
6970 has 2.7 TFLOPS of floating point computational power and a 7970 is ~3.8 TFLOPS.
6970 has 2.7 TFLOPS of floating point computational power and a 7970 is ~3.8 TFLOPS.
WOW. Did not realize the difference between a 5870 and 7970 was that big.
Now that you look at the TFLOPS and MHash/s it really does look like the 7970 is almost 1.7 times faster than a 5870 due to the much better overclocking ability of 28nm GPUs.
Where do you guys get 6970 gets 683 GFLOPS.
6970 has 2.7 TFLOPS of floating point computational power and a 7970 is ~3.8 TFLOPS.
6970 has 2.7 TFLOPS of floating point computational power and a 7970 is ~3.8 TFLOPS.
from a quick google it doesn't support opencl
interesting...
so assuming 14MH/s
to get anywhere near a 6970 in terms of raw MH/s power even at a conservative guestimate.... your probably looking at at least 25-50 of these units all hashing away at around 14MH/s to the same pool under a single account to get near say 600MH/s ( 42 rasberries )
Obviously really inefficient but would look really neat in a rack with loads of these things hooked up... sort of backwards retro
so assuming 14MH/s
to get anywhere near a 6970 in terms of raw MH/s power even at a conservative guestimate.... your probably looking at at least 25-50 of these units all hashing away at around 14MH/s to the same pool under a single account to get near say 600MH/s ( 42 rasberries )
Obviously really inefficient but would look really neat in a rack with loads of these things hooked up... sort of backwards retro
Where it says GPU is capable of 1Gpixel/s, 1.5Gtexel/s or 24GFLOPs with texture filtering and DMA infrastructure you'll have to excuse my ignorance here, FPGA stuff is like an alien world to me, but isn't 24GFLOPs a rather large number..... can anyone compare this to say a 6970? GFLOPs for GFLOPs
AMD's GFLOPs # for 6970 is 683, or >28x greater.So the 6970 gets 397 MH/s with 683 GFLOPS, or 1.72 GFLOPS / MH/s
So the raspberry pi GPU gets 24 / 1.72 = little under 14 MH/s
14 MH/s for $25 (A type is cheaper but with the same GPU) is $0.56 / MH/s
Conclusion, the normal GPUs for a normal pc are 3 to 6 times more efficient.
Another point, (most? all?) GPU miners are probably written with the opencl library, I don't know if there exists a version of it for the raspberry pi.
Where it says GPU is capable of 1Gpixel/s, 1.5Gtexel/s or 24GFLOPs with texture filtering and DMA infrastructure you'll have to excuse my ignorance here, FPGA stuff is like an alien world to me, but isn't 24GFLOPs a rather large number..... can anyone compare this to say a 6970? GFLOPs for GFLOPs
AMD's GFLOPs # for 6970 is 683, or >28x greater. Jump to: