A slightly more open approach to bitcoin on the GPU

[Syke]

nelisky, I'll provide a card. Check your PMs.

[1713542964]

1713542964

[1713542964]

1713542964

[mizerydearia]

Everyone developing this GPU thing seems to have an agenda.

Maybe noagendamarket should do it! ^_^

[puddinpop]

I just tried your latest patch.  I noticed a few things:

• Your post, and the patch name, makes it seem like the patch is against revision 157, which does not yet exist.  Upon looking at the content of the patch, it is obvious the patch is against revision 155.
• You have an extraneous curly brace on line ~3077 when FOURWAYSSE2 is not defined
• I get about 6200 khash/s with your patch using the GPU only (limit set to 1 CPU).  However without the CPU limit, and using 2 CPUs, I only get 6500 khash/s according to the counter.  That can't be right.

[nelisky]

I just tried your latest patch.  I noticed a few things:

• Your post, and the patch name, makes it seem like the patch is against revision 157, which does not yet exist.  Upon looking at the content of the patch, it is obvious the patch is against revision 155.
• You have an extraneous curly brace on line ~3077 when FOURWAYSSE2 is not defined
• I get about 6200 khash/s with your patch using the GPU only (limit set to 1 CPU).  However without the CPU limit, and using 2 CPUs, I only get 6500 khash/s according to the counter.  That can't be right.

Thanks for trying it out, finally some feedback

You are correct about the revision number being wrong. It was actually against r154. The fact I did a hacky job at including the miner means that problems do exist. I find that using all the cores + the GPU actually slows everything, and that reducing the number of mining threads on the fly doesn't do the right thing. One has to, as best as I can understand, start with 1 and up to 2 or more, but reducing threads leaves the system in unexpected states.

As for using a limit of -1 slowing things down, it makes some sense, as the CPU still is at 100% when using the GPU, doing memcpy's and finding nonces that are potential winners.

I'm attaching a new version that unrolls pretty much everything and nets me an extra 1MH/s, but this is finely tuned for my own system. This comes from a slightly more patched code base, so I hope I didn't leave any extra stuff.

[nelisky]

I just tried your latest patch.  I noticed a few things:

• Your post, and the patch name, makes it seem like the patch is against revision 157, which does not yet exist.  Upon looking at the content of the patch, it is obvious the patch is against revision 155.
• You have an extraneous curly brace on line ~3077 when FOURWAYSSE2 is not defined
• I get about 6200 khash/s with your patch using the GPU only (limit set to 1 CPU).  However without the CPU limit, and using 2 CPUs, I only get 6500 khash/s according to the counter.  That can't be right.

Hey puddinpop, I'm curious. On what OS are you compiling this? If it's not OSX, care to share the makefile? Also, what is you graphics card and how does my patch compare to your version on it? I got about 20% increase on mine compared to yours, but then again some things like the threads per block or the total number of blocks have a huge impact on the overal performance, and I didn't try to optimize your version.

[nelisky]

I just tried your latest patch.  I noticed a few things:

• Your post, and the patch name, makes it seem like the patch is against revision 157, which does not yet exist.  Upon looking at the content of the patch, it is obvious the patch is against revision 155.
• You have an extraneous curly brace on line ~3077 when FOURWAYSSE2 is not defined
• I get about 6200 khash/s with your patch using the GPU only (limit set to 1 CPU).  However without the CPU limit, and using 2 CPUs, I only get 6500 khash/s according to the counter.  That can't be right.

Hey puddinpop, I'm curious. On what OS are you compiling this? If it's not OSX, care to share the makefile? Also, what is you graphics card and how does my patch compare to your version on it? I got about 20% increase on mine compared to yours, but then again some things like the threads per block or the total number of blocks have a huge impact on the overal performance, and I didn't try to optimize your version.

I'm sure puddinpop has good reasons to not reply to my questions, but it feels weird. It's almost like he has something against me. Do you, puddinpop? Can't we just be friends?

Anyhow, no joke, I have generated my first real block using the CUDA miner! Woohoo, 50 coins accounted for, *now* it was all worth it, the sleepless nights, the complaining family due to my lack of attention, everything!

If I get another block I may start developing an opencl version, while I'm in profit! That is, if puddinpop didn't make one first and sold it for 20k coins

[GeorgeH]

Has anyone built this for windows?

[em3rgentOrdr]

I just tried your latest patch.  I noticed a few things:

• Your post, and the patch name, makes it seem like the patch is against revision 157, which does not yet exist.  Upon looking at the content of the patch, it is obvious the patch is against revision 155.
• You have an extraneous curly brace on line ~3077 when FOURWAYSSE2 is not defined
• I get about 6200 khash/s with your patch using the GPU only (limit set to 1 CPU).  However without the CPU limit, and using 2 CPUs, I only get 6500 khash/s according to the counter.  That can't be right.

Hey puddinpop, I'm curious. On what OS are you compiling this? If it's not OSX, care to share the makefile? Also, what is you graphics card and how does my patch compare to your version on it? I got about 20% increase on mine compared to yours, but then again some things like the threads per block or the total number of blocks have a huge impact on the overal performance, and I didn't try to optimize your version.

I'm sure puddinpop has good reasons to not reply to my questions, but it feels weird. It's almost like he has something against me. Do you, puddinpop? Can't we just be friends?

Anyhow, no joke, I have generated my first real block using the CUDA miner! Woohoo, 50 coins accounted for, *now* it was all worth it, the sleepless nights, the complaining family due to my lack of attention, everything!

If I get another block I may start developing an opencl version, while I'm in profit! That is, if puddinpop didn't make one first and sold it for 20k coins

Nelisky (or anyone else), could you provide the makefile and what system you compiled this on.  I'm really having trouble compiling this.  I have Ubuntu 10.04 with a GFX 9800 Gx2 with the latest Nvidia CUDA drivers installed (and I verified that my CUDA test programs do work). 

If no one wants to provide the makefile, could someone explains how I would go about modifying the existing unix makefile to compile the cuda version?

Also, what specific SVN reversion are we patching against?

[nelisky]

Nelisky (or anyone else), could you provide the makefile and what system you compiled this on.  I'm really having trouble compiling this.  I have Ubuntu 10.04 with a GFX 9800 Gx2 with the latest Nvidia CUDA drivers installed (and I verified that my CUDA test programs do work). 

If no one wants to provide the makefile, could someone explains how I would go about modifying the existing unix makefile to compile the cuda version?

Also, what specific SVN reversion are we patching against?

It's been a while since I last worked on this, but I have compiled on OSX, against revision 156 and the osx makefile changes are on the patch a few messages above. You will probably need to change the paths to the libs, but should be fairly simple to do so.

[LZ]

nelisky, can you make r205 patch?

[nelisky]

nelisky, can you make r205 patch?

Not right now, no, I lack the time. But I'm curious, why not use one of the much more optimized by now OpenCL implementations?

[LZ]

Our strength is in the diversity of solutions. Sometimes your miner is the optimal solution.