RPCMiner 2013 (CUDA only) [Win32 only] (+ 5-7% eff.)

[CoalChamber]

I did some modifications to original puddinpop's Bitcoin RPCMiner, and now it works slightly better!
Of course, this is true only for its CUDA version, since I have a nVidia card, and all modifications I did only for these GPUs.

Unfortunately, all of changes are incompatible with CPU & OpenCL versions: you can compile CPU (Crypto++ or 4-way) or OpenCL version, but it just will not work. But this is not a problem: Crypto++ & 4-way algorithms are not so effective, as CPU algorithm in UfaSoft's miner, so if you want to do a CPU mining, you can use that miner from UfaSoft.
Also, there is more effective OpenCL miner, such as CGMiner, if you use ATI card for mining...

So I decided to concentrate on the work on the CUDA miner...  


Kernel code was significantly changed, but it looks like there are some other ways of improvement. Generally speaking, I want to do GREAT research on SHA-256 internals, but it will take much time, and time is money, as you know...  

So, if you want to help me with that research, and, of course, also make a contribution to the development of this project (if you find it useful), you can just donate me some BTCBTCBTCs for it.  Say, 12-15 BTC is enough.Offer above is still valid, but you can spend your BTCs more effectively



UPDATE 18.11.2013
New alpha version is released! Check download links below!


Where to download

RPCMiner 2013 page at SourceForge: https://sourceforge.net/projects/rpcminer2013/ [Source files are available in the "Code" section]

You can download executable and only required CUBIN file separately, or download all-in-one SFX RAR package (it contains .exe and all CUBINs for all 8 possible compute capabilities).

Download separate files (beta) here
-OR-
Download all-in-one package (beta) here
-OR-
Download all-in-one package (latest alpha) here

Note:
Only Windows 32-bit version [Win32] is supported. There are no benefits to compile it to x64 - because SHA256 deals with 32-bit data chunks, 64-bit registers (if available for GPU) are useless.

BTW, I test RPCMiner2013-cuda.exe on Win7x64  and it works with the same hashrate, as on Win7x86 and WinXP!



Features (in comparison with original RPCMiner):
1. Kernel was completely revised, and it works now up to 5-7% effective.
2. Added some support for monitoring nVidia GPU Temperature & Fan Speed (may not be available with your card).
3. Best grid blocks/threads auto-detection now works more accurate and effectively.
4. Winning nonce is now displayed.
5. (Alpha only) "Nonce-bouncing" feature is now implemented.

Bug fixes:
 Fixed serious bug with incorrect handling of any target that isn't a kind of 0x00000000ffffff....ffffff.

Known problems:
Your PC can stop responding randomly,  if you try to close application with "Close" button, or by pressing Ctrl-C/-Break combination. Usually it happens, when you use too high values of "-aggression" parameter.
You can avoid this problem, when using RPCMiner2013  with mining-proxy software.  Just close mining-proxy.exe, wait a minute until you see "No blocks are being hashing right now", and after that you can safely close application.

Known limitations:
Only GetWork protocol is supported. If you want Stratum, you must run mining-proxy software, available from your favorite pool.



How to use (examples):

1. With Deepbit pool:

RPCMiner2013-cuda.exe -url=http://pit.deepbit.net:8332 -user=<YourMailbox_at_SomeWhere.ext>_worker -password=<Password> -workrefreshms=25000 -gpu -aggression=8 -gpugrid=16 -gputhreads=192

2. With Slush's pool:
Slush's pool doesn't support Getwork, so you must run mining-proxy.exe at first (without any parameters). Your command line will looks like following:

RPCMiner2013-cuda.exe -url=http://localhost:8332 -user=<UserName>.worker1 -password=<Password> -workrefreshms=25000 -gpu -aggression=7 -gpugrid=64 -gputhreads=64

3. With GUIMiner:
Just rename RPCMiner2013-cuda.exe to rpcminer-cuda.exe and after that you can easily use this app with GUIMiner!

Notes:
1. Default value for "-workrefreshms" is 4000 (4s). But it's not enough, if you have a slow card! So, if you raise it to 15000-20000 or even to 50000, it will increase chances to find nonce.
2. CUDA grid launch parameters that shown are best suited for my old-but-gold GTS 250. Of course, if you have the latest card, you can use larger values. Or you can omit "-gpugrid" and/or "-gputhreads" to run auto-detection of these parameters.

Some performance numbers (for Gygabyte's "green" GTS 250):
Input parameters: aggression = 8, gpugrid = 64, gputhreads = 64
Perf for default settings (Core = 675 MHz, Mem = 1000 MHz, Shdr = 1620 MHz):  30373 khashes/s
Perf for OC'ed settings  (Core = 762 MHz, Mem = 1225 MHz, Shdr = 1958 MHz):  36111 khashes/s

Planned features (will be available in next alpha release):
1. Non-zero starting nonce. (not applicable)

[1715583398]

1715583398

[1715583398]

1715583398

[1715583398]

1715583398

[1715583398]

1715583398

[Eli0t]

interesting, might be useful for ZET mining

[-ck]

Planned features (will be available in next alpha release):
These two are for real bitcoin mining cheaters:  
1. Non-zero starting nonce.
Makes possible to set starting nonce, that is non-zero.
For example, let we noticed that nonce value is very seldom less than 0x01000000. So why don't we start with the search of nonces ​from this starting point?  
2. "Nonce-bouncing".
Sets new starting nonce value, which depends from the previously found nonce(s). It also can change during the hashing; increase, then decrease, then increase again ... - i.e., "bouncing", or fluctuating around some median value.

Logic fail. These will have no effect.

[ssateneth]

Anyone have performance numbers?

[HellDiverUK]

So, what you're saying is "Send me 15BTC to work on some abandoned program with no point because no-one uses GPUs any more, and even when they did they didn't use nVidia ones".

Or did I miss something? 

[Notanon]

Anyone have performance numbers?

43,733 kilohashes/sec on a Gigabyte Geforce GTX 660 OC

[patoberli]

I get around 5150 khash/s with my Nvidia Quadro FX580. This is around 400 more than what I get with GUIMiners default opencl miner.

[CoalChamber]

New alpha version is already available on SourceForge! Check the download links in the first message!

P.S.
BTW, working now on Primecoin CPU & GPU miners. So you can download binaries at https://sourceforge.net/projects/primecoincpugpuminers/ 

[paulus51]

2. With Slush's pool:
Slush's pool doesn't support Getwork, so you must run mining-proxy.exe at first (without any parameters). Your command line will looks like following:

RPCMiner2013-cuda.exe -url=http://localhost:8332 -user=<UserName>.worker1 -password=<Password> -workrefreshms=25000 -gpu -aggression=7 -gpugrid=64 -gputhreads=64

3. With GUIMiner:
Just rename RPCMiner2013-cuda.exe to rpcminer-cuda.exe and after that you can easily use this app with GUIMiner!

cant get it to work ??

in guiminer ext.path : D:\bitcoins\guiminer\miners\puddinpop\rpcminer-cuda.exe  [ renamed as you told ]

no extra flags ect ect
what is wrong here ?

[CoalChamber]

cant get it to work ??

in guiminer ext.path : D:\bitcoins\guiminer\miners\puddinpop\rpcminer-cuda.exe  [ renamed as you told ]

no extra flags ect ect
what is wrong here ?

You must fill all the fields like that (for deepbit pool, as an example):



This version of GUIminer is old, but it works

[souljahinc]

based on various experimentations i've discovered the way of using the rpcminer-cuda effectively.

GPU : GT 540M (1GB dedicated memory)
CUDA Core / Shaders : 96

my rpcminer commandline will be as per followings :-
RPCMiner2013-cuda.exe -url=http://hostname:port -user=username.workerxx -password=passwd -workrefreshms=18000 -gpu=0 -aggression=7 -gpugrid=96 -gputhreads=384

when i firstly used the proggy i've let it run on auto mode detection trying to see what the best value for its gpugrid and gputhreads. after countless of experimentations, i've concluded that gpugrid might be referering to its cuda-cores and gputhreads will be best and less overheat issue if its x4 the value of its cuda-cores. therefore, i've set my -gpugrid=96 and -gputhreads=384 with interval of 18000ms (18 seconds). i hope that the information provided might be some good use to other cuda gpu miners. cheers and happy mining.

p.s. i've got a stable 9.6MH/s-11.6MH/s (excluding the 1st 30-120 secs of burst speed at 16.xMH/s++) flowing rather than relying on the auto detection mode which is always giving random figures and my mining speed stale at 7.7MH/s-8.2MH/s.

[djm34]

on what are you running ?

[Godpikkeno]

i'm using gtx 970, and when i started guiminer, it appear unable to load cuda module.
I'm using last nvidia driver.

[djm34]

i'm using gtx 970, and when i started guiminer, it appear unable to load cuda module.
I'm using last nvidia driver.

that's the kind of things which happen when you resurrect old thread 

[Bombadil]

i'm using gtx 970, and when i started guiminer, it appear unable to load cuda module.
I'm using last nvidia driver.

that's the kind of things which happen when you resurrect old thread  

It gets even worse the older the thread is... and this one is verrrry old
I didn't know there were other cudaminers before the real cudaminer (and now ccminer) ^^ Thanks for reviving this one