Generating Bitcoins with your video card (OpenCL/CUDA)

[Ground Loop]

Runs fine on the i5 Macbook Pro.  Generates a lot of heat, but it's more stable than the speedy Windows x64 version by far.

[1714020506]

1714020506

[1714020506]

1714020506

[omegadraconis]

I finally got around to testing this on my hackintosh pc and it seems to be running just fine, although it is definitely not ready for the masses. The speed keep jumping around (although this maybe related to my hackintosh nvidia driver), The debug log shows it is coming in @
**Perf - thread 1 : 1636k iter/sec
**Perf - thread 4 : 1616k iter/sec
**Perf - thread 5 : 1295k iter/sec
**Perf - total : 4547k iter/sec (3 threads)

**Perf - thread 1 : 3363k iter/sec
**Perf - thread 3 : 2109k iter/sec
**Perf - total : 5472k iter/sec (2 threads)

**Perf - thread 1 : 2630k iter/sec
**Perf - thread 3 : 2069k iter/sec
**Perf - total : 4699k iter/sec (2 threads)

Seems like the speed is dropping over time after running for a short bit of time, anyone else having this issue?

At it's peak, this is about 2200khash better then in windows with the x64bit optimized version. The normal mac client is only coming in at 858khash/s for some reason with two cores. I think is is real promise with using opencl to add khashs.

My hardware is an intel Dual-core celeron @3.8ghz, My GPU is an Geforce 8800GTX, 2GB ram, and I am running OSX 10.6.3.

[nimnul]

While this is definitely really cool, it won't benefit anyone if you release it. After two weeks, the difficulty system will adjust and we'll be back to generating bitcoins at the same rate but using much more computer power. Although I suppose this will benefit those with fast video cards instead of just fast CPUs.

On the contrary, it will benefit everyone. Remember, bitcoin is not about generating free bitcoins - it's about creating a market. And more complexity means more stable market as it becomes harder to rent a farm of servers and put generated coins to market to destabilize it.

[kwukduck]

Yes, it brings more power to the desktop user instead of the server people that are just farming coins in big numbers, servers don't have strong GPU's so it just pulls the desktop machines up to level.

Win 7 32bit + Ubuntu 10.04LTS
nVidia 8800GT

[LZ]

Fedora 13, Mac OS 10.5, Windows 7
nVidia GeForce GTS 250

[puddinpop]

I've created a Windows 32 bit CUDA enabled client.  Consider it a beta right now.  You'll need the Visual Studio 2005 runtime to run it.  I spent a considerable amount of effort to create this, so I'm asking that if you use it, you send 5 me BTC for every block you generate while using it.  I suggest you try it out first with 2 clients connected to each other and an empty block database, so you can see that it works correctly for you relatively quickly.

See my signature for the latest download.

[BitLex]

.. so I'm asking that if you use it, you send 5 me BTC for every block you generate while using it. ..

your askin'?
would be cool, but seems like you hardcoded your 10%-share to auto-send you coins,
as much as i'd like to reward your efforts, i really don't like that.

besides that and the fact, that generating totally slows down my system (not even surfing is fun),
it seems to work, it shows (!) me  ~33.000khs on a gtx260, not bad (if it's true) compared to the regular ~3.800khs the triple-core-cpu gets.
can't do anything else besides running it though, so i shut it down again after a short test.

[Gespenster]

Not sure if I trust this, might be a scam to rip off your coins or something.

[puddinpop]

Yes, the transaction is automatic if you have an available balance.  It is detailed in the readme.txt file.  Please don't use it if you feel uncomfortable with it.  I just want to give a better chance of generating blocks to those with CUDA enabled hardware.  There is already 1 person with 20% of the hashing capability, so I thought this would even the field better.

besides that and the fact, that generating totally slows down my system (not even surfing is fun),

While the kernel is running, the GPU is 100% dedicated to running the kernel only, not updating the screen.  This is why you will see sluggishness.  Actually, you could squeeze out some more performance at the cost of more sluggishness.  This is perfectly normal.  If you have 2 graphics cards, you could use the slow one for your display and the fast one just for CUDA.

Reflector says it is not a .NET assembly.

That is quite correct.  Bitcoin is a c++ application, so obviously there won't be any .NET code.

how do we know you won't take more or do other things? Your intentions seem to be muddy.

Please test it in a sandbox environment if you are uneasy.  The only thing it does differently form the standard client is generate using CUDA, and send 5 BTC for every block you generate while using the CUDA miner.  Everything else is standard.

[Gespenster]

There is already 1 person with 20% of the hashing capability, so I thought this would even the field better.

Could you elaborate on that?

[puddinpop]

There is already 1 person with 20% of the hashing capability, so I thought this would even the field better.

Could you elaborate on that?

ArtForz, on IRC, has like 12 or so 5770s running his own OpenCL client.  The entire network is doing something around 5-6Ghash/s, and he alone has stated that he has over 1Ghash/s

[MoonShadow]

While the kernel is running, the GPU is 100% dedicated to running the kernel only, not updating the screen.  This is why you will see sluggishness.  Actually, you could squeeze out some more performance at the cost of more sluggishness.  This is perfectly normal.  If you have 2 graphics cards, you could use the slow one for your display and the fast one just for CUDA.

Is there some way to tie the GPU portion of generation to a screensaver-like start/stop?

[puddinpop]

Is there some way to tie the GPU portion of generation to a screensaver-like start/stop?

Sure, if someone wanted to code a screensaver they could do that.

[Ground Loop]

Everything above is true here too.

My GT240 is pumping 16,339 khash/sec.  Yow!
And yes, the machine is damn near unuseable.

By my math, I'm generating coins at 3x the previous rate, so if there's a 10% author tax, I can live with that.

[omegadraconis]

My 8800GTX is currently crunching at a blistering 25000-24000 khashs with one cpu core pegged @ 50%. The hardware is the exact same as above where I posted my results from the mac OSX Cuda Client but, running windows 7 ultimate 64bit. My system is still usable (browsed the web with no noticeable difference) but, windows menus and other aero enabled windows are laggy. Not trying to be "that guy" but, I am a little un-trusting of a closed-source client that sends bitcoins to a random address without knowing what else is under the hood. Then again if it nets me coins and only cost 5 thats not bad as long as that is all it does. Anyone using it generate coins yet? Also how many coins would you want to opensource the client?

[EDIT]
Wanted to add a note here that I have been running the client for about 8 hrs now and it is stable no crashes. Seems that puddinpop did a good job with the code. I think I can live with the 10% tax for now once I see the coins start rolling in.

[mizerydearia]

Although I don't use windows, I'll donate some bitcoins to puddinpop if they release the client source code as open source.  Otherwise I'll donate some bitcoins to the first person to release a hacked/modified version of the client that disables the 10% autopayment on block generation, whichever comes first.

puddinpop would ultimately establish far more profits if the client remains closed, at least until using the client is no longer beneficial as the difficulty will have increased far too much for it to be as beneficial as it was compared to noncuda clients.  However, reverting back to a noncuda client would be worse than continuing to use the cuda client and therefore unless an alternative is established, puddinpop will continue to receive plenty of profits from usage of the client.  While it will be nice to donate a nice amount to puddinpop for their efforts, I personally feel 10% for all block generations from everyone is too much, and I would like to encourage open source or otherwise rewarding those that defeat the proprietary profiting addition.

Keep in mind while khash/sec is increased providing higher chance to generate a block, more electricity is used to do so.  I don't believe puddinpop is willing to pay 10% of electricity bill or cost of extra electricities used from usage of this software.

Also, Satoshi and sirius-m Have put forth a lot of hard work into the open source Bitcoin client and do not force donations.  I would like to donate to Satoshi for his efforts. ^_^

[omegadraconis]

Although I don't use windows, I'll donate some bitcoins to puddinpop if they release the client source code as open source.  Otherwise I'll donate some bitcoins to the first person to release a hacked/modified version of the client that disables the 10% autopayment on block generation, whichever comes first.

I will donate for an opensource client.

[TTBit]

I'm not a programmer or expert user, so go easy on me please.

I have an old athlon machine that can generate ~1850 khash/s. I have a geforce 8500 gt on that machine, and used this to up production to ~2150 khash/s (says GPU 2150, so I know I'm using the -gpu). How can I get these 20,000 khash/sec that everyone else is? Is my athlon the bottleneck here, or can I just buy a nicer video card?

[omegadraconis]

I'm not a programmer or expert user, so go easy on me please.

I have an old athlon machine that can generate ~1850 khash/s. I have a geforce 8500 gt on that machine, and used this to up production to ~2150 khash/s (says GPU 2150, so I know I'm using the -gpu). How can I get these 20,000 khash/sec that everyone else is? Is my athlon the bottleneck here, or can I just buy a nicer video card?

For Cuda I think (emphasis on think, not know) the best thing to look for would be the number of Stream Processors, which is like the number of cores. If you are using an old Athlon and your stuck with agp I think this will slow you down abit vs a pci-e bus but, I really don't know. Also you would want to make sure you cpu can keep up with pumping data to your gpu. Like I said in my post one of my cpu's cores is pegged @ 100% usage while the gpu crunches away.

[puddinpop]

If I were to receive a sizable donation, I might be willing to open source the code.  However, any talk about hacking or cracking the current client makes me less likely to do this, and I will not accept donations from anyone who has already advocated for such measures.

Here's a little story to think about.

There is an apple tree growing in some public place where everyone is free to take apples whenever they wish.  Everyone is happy and content eating the free apples directly from the tree.  Now some start to think about all the possibilities that the apples can provide.  You can make apple juice, apple cider, apple pie, etc.  Most only think about the possibilities and don't take action.  Either because they can't cook, or they don't have the time or resources.  Now someone comes along and decides to make apple juice.  He sets up a stand and sells the juice to the people.  Some people are fine with this, and happily pay.  They understand the value added to the apple by processing it into juice.  It took time, effort, and resources to do so.  Others are upset that the apples they had been getting for free have been processed and that they now have to pay for the refined product.  They would rather take the apple juice by force rather than paying for it or making their own.

I understand that some of you are uncomfortable with the client.  If that is the case the obvious solution is not to use the client.