RaspberryPi litecoin CPU and GPU miner bounty

[matthewh3]

Is anyone interested in chipping into a bounty for someone to develop an optimised LTC miner for the RaspberryPi that mines on its ARM1176JZF-S CPU and its Broadcom VideoCore IV GPU ?


LTC donation address - LKQ9ZKKBNHYcPjSkSK2GnkbYb7PeFgnSj7

BTC donation address - 1JooT9sKj3YG2k8i6MrH5DTo72g8rF7gKb

NMC donation address - NCvUBqH8bPGwybfWHWycF8dYiLYb8gg3VN


On the case the project not getting developed all donations will go to projects on http://litebit.co/

[1714172321]

1714172321

[1714172321]

1714172321

[grimd34th]

ill chip in 200 ltc if you give me an addr

[tatsuchan]

How about we chip in on getting more companies involved in making these mini power efficient computers?  What do we have now, a 2 year wait list?  Good luck buying in bulk ANYTIME soon.  I'm all for development though, if someone wants to set up a pool, I'll chip in more as it gains relevance. 

[matthewh3]

ill chip in 200 ltc if you give me an addr

Before I can take donations we need more interest of a bounty of at least £100 and potential developers who can program its GPU (not easy I think and a £100 bounty may seem pity-full?).  The LTC CPU miner can probably be recompiled (we just need it optimising for the ARM instruction set) and maybe the "Reaper" program can be adjusted for its GPU and DSP?  I think all this will be eventually done anyway as the RaspberryPi is a development board I was just thinking a nice bounty might speed up the development.  As the RaspberryPi only cost £30 and runs on less than 4W so a dedicated optimised easy to use dedicated mining software for the board may see a lot of interest from hobby miners wanting to get a board or two. 

[pooler]

The Raspberry Pi has pretty much the same CPU as my iPhone 3G, only running at a higher clock speed.
On that CPU, the current version of cpuminer can do about 100 hashes/second at 412 MHz, so on the 700-MHz processor of the Pi (assuming everything else constant, which is almost certainly not the case) it should do less than 200 h/s. Of course this is using a C implementation of scrypt, but even if someone wrote optimized code for the ARM I doubt it would be possible to reach 1 kh/s.

As regards the GPU, it doesn't support OpenCL. It does support OpenGL ES, but that's not a viable option for GPGPU computing. (source)

[matthewh3]

The Raspberry Pi has pretty much the same CPU as my iPhone 3G, only running at a higher clock speed.
On that CPU, the current version of cpuminer can do about 100 hashes/second at 412 MHz, so on the 700-MHz processor of the Pi (assuming everything else constant, which is almost certainly not the case) it should do less than 200 h/s. Of course this is using a C implementation of scrypt, but even if someone wrote optimized code for the ARM I doubt it would be possible to reach 1 kh/s.

As regards the GPU, it doesn't support OpenCL. It does support OpenGL ES, but that's not a viable option for GPGPU computing. (source)

Thanks I've also realised the Broadcom GPU doesn't have a "Cuda" or a "Catalyst" so would need Broadcoms help.  Anyway I've posted the idea in the RaspberryPi forum to see what developers interested in the board think.

The only positive thing I read from that link was -

"its a complete implementation of OGL ES2.0, so does support shaders. What you do with them is up to you!"

[matthewh3]

LTC donation address - LKQ9ZKKBNHYcPjSkSK2GnkbYb7PeFgnSj7

BTC donation address - 1JooT9sKj3YG2k8i6MrH5DTo72g8rF7gKb

NMC donation address - NCvUBqH8bPGwybfWHWycF8dYiLYb8gg3VN

On the case the project not getting developed all donations will go to projects on http://litebit.co/

[matthewh3]

LTC donation address - LKQ9ZKKBNHYcPjSkSK2GnkbYb7PeFgnSj7

BTC donation address - 1JooT9sKj3YG2k8i6MrH5DTo72g8rF7gKb

NMC donation address - NCvUBqH8bPGwybfWHWycF8dYiLYb8gg3VN

On the case the project not getting developed all donations will go to projects on http://litebit.co/

I have received some interest on the RaspberryPi forums - http://www.raspberrypi.org/forum/projects-and-collaboration-general/raspberrypi-litecoin-cpu-and-gpu-miner-bounty#p62032



"I'll put it another way:
The ONLY people that can write code that runs on the GPU is the manufacturer (Broadcom).
The manufacturer provides users info so we can make OpenGL API calls. We, the users, don't get to write any code that runs on the GPU. All we can do is send requests to draw stuff. We can try and fake some elements of parallel processing but there are inherent limitations that don't make it worthwhile.
 
I think your colleagues assement on your linked thread of Bitcointalk of this thread:
http://www.raspberrypi.org/for…..n-the-r-pi
pretty much sums up the situation – when you re-read it, bear in mind that JamesH is a Broadcom employee and therefore really knows his stuff about such things.
 
However, if there is going to be some money offered for anything, I'll start pledging for porting the X display driver to OpenGL ES for the RaspberryPi."



So get the donations coming  

[pooler]

I've just finished writing some optimized assembly code for the ARM11. Here are the results for scrypt on my iPhone-3G, running at 412 MHz:

C implementation* (cpuminer 2.2):  103 h/s
New hand-optimized assembly:       160 h/s

Of course a 55% speedup is not bad at all, but still if we assume that hash rate scales linearly with CPU clock speed, we can only expect the Raspberry Pi to do about 270 h/s.

I still have to do some testing to make sure that the code behaves well on more recent ARM processors like the Cortex-A8, but I'm pretty sure this new code will get included in the next release of cpuminer, which will probably be out in a week or two.

* The C implementation was compiled with gcc 4.2.1; I know that's not exactly recent, but it's all I can get on this device. It is possible that more recent versions of gcc produce faster code.

[matthewh3]

I've just finished writing some optimized assembly code for the ARM11. Here are the results for scrypt on my iPhone-3G, running at 412 MHz:

C implementation* (cpuminer 2.2):	103 h/s
New hand-optimized assembly:	160 h/s

Of course a 55% speedup is not bad at all, but still if we assume that hash rate scales linearly with CPU clock speed, we can only expect the Raspberry Pi to do about 270 h/s.

I still have to do some testing to make sure that the code behaves well on more recent ARM processors like the Cortex-A8, but I'm pretty sure this new code will get included in the next release of cpuminer, which will probably be out in a week or two.

* The C implementation was compiled with gcc 4.2.1; I know that's not exactly recent, but it's all I can get on this device. It is possible that more recent versions of gcc produce faster code.

Cool but the bounty is still very small maybe we could apply for DevCoin funding to keep it a work in progress?

[ShadowAlexey]

I've just finished writing some optimized assembly code for the ARM11. Here are the results for scrypt on my iPhone-3G, running at 412 MHz:

C implementation* (cpuminer 2.2):  103 h/s
New hand-optimized assembly:       160 h/s

Of course a 55% speedup is not bad at all, but still if we assume that hash rate scales linearly with CPU clock speed, we can only expect the Raspberry Pi to do about 270 h/s.

I still have to do some testing to make sure that the code behaves well on more recent ARM processors like the Cortex-A8, but I'm pretty sure this new code will get included in the next release of cpuminer, which will probably be out in a week or two.

* The C implementation was compiled with gcc 4.2.1; I know that's not exactly recent, but it's all I can get on this device. It is possible that more recent versions of gcc produce faster code.

This might not be the case for rbp, but are you planing to use neon instruction set?

[pooler]

This might not be the case for rbp, but are you planing to use neon instruction set?

Yes, I plan to start working on a NEON-based implementation later on this week.

[daybyter]

Hi!

Why do you want to mine on such a machine? Better use it to run a tradebot 24/7...

[Remember remember the 5th of November]

Hi!

Why do you want to mine on such a machine? Better use it to run a tradebot 24/7...

I agree, it's already so fragile, you will only break it. This is why uninformed people shouldn't request things. Sorry if I sound harsh, but it's true.

[pooler]

Hi!

Why do you want to mine on such a machine? Better use it to run a tradebot 24/7...

I agree, it's already so fragile, you will only break it. This is why uninformed people shouldn't request things. Sorry if I sound harsh, but it's true.

Nothing makes me think the Pi is fragile. Its developers say it is perfectly safe to have it running 24/7 under heavy loads.
Also consider that ARM processors are way more power efficient than Intel CPUs, so heat dissipation is much less of a problem.

As for the why: mining is not all about the profit. It is both interesting and useful to assess the capabilities of a given hardware platform.
And it's fun.

[matthewh3]

Hi!

Why do you want to mine on such a machine? Better use it to run a tradebot 24/7...

I agree, it's already so fragile, you will only break it. This is why uninformed people shouldn't request things. Sorry if I sound harsh, but it's true.

Nothing makes me think the Pi is fragile. Its developers say it is perfectly safe to have it running 24/7 under heavy loads.
Also consider that ARM processors are way more power efficient than Intel CPUs, so heat dissipation is much less of a problem.

As for the why: mining is not all about the profit. It is both interesting and useful to assess the capabilities of a given hardware platform.
And it's fun.

Also why am I uniformed when LTC mining has already been ported to iPhone and Android.  Arm CPU's are way more efficient per watt than x86 there is even ARM based super-computers and servers being built.  So it makes sense to mine LTC which is supposed to be CPU biased on the most power efficient CPU family.  Sorry to sound harsh but maybe uninformed people shouldn't claim to be a font of knowledge on the subject.

[daybyter]

That was before the reaper LTC GPU miner was released and the LTC rate dropped to nil...

[GTRsdk]

That was before the reaper LTC GPU miner was released and the LTC rate dropped to nil...

I don't think the GPU miner had much of an effect (to be honest).

It isn't nearly as optimized as the bitcoin ones.

[pooler]

After some more optimization, I think I have finally broken the 300 hashes/sec barrier for the ARM11 at 700 MHz.
I say "I think" because I have only tested the code on my iPhone 3G at 412 MHz, which can now do 181 hashes/sec.

I don't know if anybody is actually interested in ARM11 mining, but this sure is challenging!

P.S. Sorry but NEON optimizations will take longer than expected, as my Cortex-based device is currently away for repair.

[matthewh3]

Maybe this with a A10 single-core 1.5GHz ARM CPU and a Mali 400 GPU would hash better plus it's only $74 again probably using only a few watts. http://arstechnica.com/gadgets/2012/05/new-74-android-mini-computer-is-slightly-larger-than-a-thumb-drive/