CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 960.

chrysophylax

legendary

Activity: 2940

Merit: 1091

--- ChainWorks Industries ---

Quote from: bathrobehero on September 05, 2015, 06:06:44 PM

Quote from: scryptr on September 05, 2015, 05:51:39 PM

GIMP ON YIIMP--

For the last couple hours I mined MUE coin (Monetary Unit) on YiiMP (yiimp.ccminer.org). MUE is a quark algorithm coin, and according to CoinWarz, currently capable of producing ~200-400 dollars a day in mining revenue. There must be a decimal place or two missing in the calculation!

I mined over 1000 MUE in the space of time that I spent on YiiMP, but they are currenly worth about 15 satoshis each. I will be earning about .00015 BTC on the exchange when my MUE coins mature, I think. Not so sure!

Every block that I hit, I got all the coins save the pool share. Somehow, CoinWarz led me to believe that I'd be earning 100X more, at least. I solved ~40 blocks at 40 MUE coins per block. Because YiiMP is more or less a hobby/experimental pool, it was as if I was solomining. There was no other miner to share the blocks with me, just the 1% pool share to deduct.

If I could only find the right coin...

--scryptr

MonetaryUnit (MUE)
Exchange volume (24h): 0.02 BTC.

No offense, but I still chuckle everytime when I see a multipool picking up coins with 0 volume.

If I ever launch a coin it will have coin maturity of about a week and it will change the algo every few weeks just to throw multipools off.

id be in on that Wink

...

#crysx

bathrobehero

legendary

Activity: 2002

Merit: 1051

ICO? Not even once.

Quote from: scryptr on September 05, 2015, 05:51:39 PM

GIMP ON YIIMP--

For the last couple hours I mined MUE coin (Monetary Unit) on YiiMP (yiimp.ccminer.org). MUE is a quark algorithm coin, and according to CoinWarz, currently capable of producing ~200-400 dollars a day in mining revenue. There must be a decimal place or two missing in the calculation!

I mined over 1000 MUE in the space of time that I spent on YiiMP, but they are currenly worth about 15 satoshis each. I will be earning about .00015 BTC on the exchange when my MUE coins mature, I think. Not so sure!

Every block that I hit, I got all the coins save the pool share. Somehow, CoinWarz led me to believe that I'd be earning 100X more, at least. I solved ~40 blocks at 40 MUE coins per block. Because YiiMP is more or less a hobby/experimental pool, it was as if I was solomining. There was no other miner to share the blocks with me, just the 1% pool share to deduct.

If I could only find the right coin...

--scryptr

MonetaryUnit (MUE)
Exchange volume (24h): 0.02 BTC.

No offense, but I still chuckle everytime when I see a multipool picking up coins with 0 volume.

If I ever launch a coin it will have coin maturity of about a week and it will change the algo every few weeks just to throw multipools off.

scryptr

legendary

Activity: 1797

Merit: 1028

GIMP ON YIIMP--

For the last couple hours I mined MUE coin (Monetary Unit) on YiiMP (yiimp.ccminer.org). MUE is a quark algorithm coin, and according to CoinWarz, currently capable of producing ~200-400 dollars a day in mining revenue. There must be a decimal place or two missing in the calculation!

I mined over 1000 MUE in the space of time that I spent on YiiMP, but they are currenly worth about 15 satoshis each. I will be earning about .00015 BTC on the exchange when my MUE coins mature, I think. Not so sure!

Every block that I hit, I got all the coins save the pool share. Somehow, CoinWarz led me to believe that I'd be earning 100X more, at least. I solved ~40 blocks at 40 MUE coins per block. Because YiiMP is more or less a hobby/experimental pool, it was as if I was solomining. There was no other miner to share the blocks with me, just the 1% pool share to deduct.

If I could only find the right coin...

--scryptr

t-nelson

member

Activity: 70

Merit: 10

Quote from: sp_ on September 05, 2015, 07:51:30 AM

Submitted a speedup 50KHASH in x11 on the 980ti. With code from the klaus_t fork. and constant rotates.

Seems on par with 65 on 750ti and 960. At least no regression noticeable.

t-nelson

member

Activity: 70

Merit: 10

Quote from: pallas on September 05, 2015, 05:21:06 AM

Quote from: djm34 on September 05, 2015, 05:08:38 AM

Quote from: Slava_K on September 05, 2015, 04:40:21 AM

-- SNIP --

needs to include cuda_vector.h instead of cuda_helper.h

I didn't modify the includes.
Furthermore, it compiles fine on Linux...
I know the build files on windows have been enhanced recently, maybe it's related to that.
I can't help because I never built on win.

The header deps are a mess. Odds are it was pulled in by some other header on Linux and you got (un?)lucky.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Submitted a speedup 50KHASH in x11 on the 980ti. With code from the klaus_t fork. and constant rotates.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: pallas on September 05, 2015, 07:42:25 AM

Quote from: sp_ on September 05, 2015, 07:21:21 AM

Quote from: pallas on September 05, 2015, 07:16:12 AM

with the current commit (reverting quark_jh and quark_skein but not groestl) quark is a bit faster... that's puzzling :-)
"in the lyra2v2 algo I now get +50KHASH on the gtx 970"
but lyra2v2 is not using quark/, Algo256/ or groestl... still puzzled ;-)

yes, but you also changed the rotl32 in the blake256 And it seems to be a little bether there. So I kept it.

I see. But that means that x11 (and others based on the same file) should also be faster :-)

x11 uses blake 512.

the 256 algos are neoscrypt,lyra, and lyra2

GingerAle

legendary

Activity: 1260

Merit: 1008

For those that follow this thread in the hopes that some improvements will happen to Monero (i.e. cryptonight) nvidia mining, the Monero Forum Funding system for djm34's proposed work is up and running:

https://bitcointalksearch.org/topic/m.12334860

well, that takes you to a link to the link, here's a direct:

https://forum.getmonero.org/8/funding-required/2355/improved-nvidia-mining-software

Sorry for the off topic. This thread has become the hotspot for this stuff though.

pallas

legendary

Activity: 2716

Merit: 1094

Black Belt Developer

Quote from: sp_ on September 05, 2015, 07:21:21 AM

Quote from: pallas on September 05, 2015, 07:16:12 AM

with the current commit (reverting quark_jh and quark_skein but not groestl) quark is a bit faster... that's puzzling :-)
"in the lyra2v2 algo I now get +50KHASH on the gtx 970"
but lyra2v2 is not using quark/, Algo256/ or groestl... still puzzled ;-)

yes, but you also changed the rotl32 in the blake256 And it seems to be a little bether there. So I kept it.

I see. But that means that x11 (and others based on the same file) should also be faster :-)

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: pallas on September 05, 2015, 07:16:12 AM

with the current commit (reverting quark_jh and quark_skein but not groestl) quark is a bit faster... that's puzzling :-)
"in the lyra2v2 algo I now get +50KHASH on the gtx 970"
but lyra2v2 is not using quark/, Algo256/ or groestl... still puzzled ;-)

yes, but you also changed the rotl32 in the blake256 And it seems to be a little bether there. So I kept it.

pallas

legendary

Activity: 2716

Merit: 1094

Black Belt Developer

with the current commit (reverting quark_jh and quark_skein but not groestl) quark is a bit faster... that's puzzling :-)

"in the lyra2v2 algo I now get +50KHASH on the gtx 970"
but lyra2v2 is not using quark/, Algo256/ or groestl... still puzzled ;-)

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

I think it's because the compiler is unable to precalculate and remove instructions that operate on contant data when we use inline asm. rotl(const,x)

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: pallas on September 05, 2015, 06:52:32 AM

Quote from: sp_ on September 05, 2015, 06:45:12 AM

I reverted the jh512 change and the rotr change.

That is suspect: quark jh and x11 bmw were redefining rotr32 with the original sph code, while all the rest use the cuda_helper version which has specific cuda code.
If the cuda specific code is slower, than we should replace the rotr32 definition in cuda_helper because all the other files will benefit from it.
But I highly doubt that it's the case: cuda_helper has been worked on for months and it can't be slower than sph code, unless the compiler is so smart that it can detect it's a rotate and replace it with even more optimized code.

In brief, I wouldn't reput the double definitions (it's a warning...) but adjust the cuda_helper version, so we have the best rotr32 on all algos.

The compiler is smart. it converts the

ROTL32c(x, n) ((x) << (n)) | ((x) >> (32 - (n)))

to one assembly instrucion. the funnelshift. Take a look at the generated .ptx code in the release folder.

But in the lyra2v2 algo I now get +50KHASH on the gtx 970, so sometimes it's faster to use the method in cuda helper.

pallas

legendary

Activity: 2716

Merit: 1094

Black Belt Developer

Quote from: sp_ on September 05, 2015, 06:45:12 AM

I reverted the jh512 change and the rotr change.

That is suspect: quark jh and x11 bmw were redefining rotr32 with the original sph code, while all the rest use the cuda_helper version which has specific cuda code.
If the cuda specific code is slower, than we should replace the rotr32 definition in cuda_helper because all the other files will benefit from it.
But I highly doubt that it's the case: cuda_helper has been worked on for months and it can't be slower than sph code, unless the compiler is so smart that it can detect it's a rotate and replace it with even more optimized code.

In brief, I wouldn't reput the double definitions (it's a warning...) but adjust the cuda_helper version, so we have the best rotr32 on all algos.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

I reverted the jh512 change and the rotr32 change. The compiler is already using the funnelshift to do the rotl32. And when you use the "fast" method. the exefile get 3kb bigger.

I think it's because the compiler is unable to precalculate and remove instructions that operate on contant data when we use inline asm. rotl(const,x)

the jh512 change is bad on the 750ti loose 100khash

pallas

legendary

Activity: 2716

Merit: 1094

Black Belt Developer

Quote from: sp_ on September 05, 2015, 05:59:13 AM

Quark is a bit slower with the last commits.

old hasrate:
55091

last hashrate:
54550

I don't think it can be because of the vector operations.
So it must be the rotr32 of cuda_helper, but that has been used on most of the other files for months.
As a test, could you replace rotr32 definition with "rotate" and see if it's faster?

ldp5500

full member

Activity: 173

Merit: 100

Someone with GTX750 mining it? Quark and X11 speed? Which version of the software used? THX!

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quark is a bit slower with the last commits.

old hasrate:
55091

last hashrate:
54550

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

I submitted a fix for the build

pallas

legendary

Activity: 2716

Merit: 1094

Black Belt Developer

Quote from: djm34 on September 05, 2015, 05:08:38 AM

Quote from: Slava_K on September 05, 2015, 04:40:21 AM

Error   18   error : identifier "uint8" is undefined   c:\ccminer-windows\groestl_functions_quad.cu   489   1   ccminer
Error   19   error : expected an expression   c:\ccminer-windows\groestl_functions_quad.cu   489   1   ccminerError   18   error : identifier "uint8" is undefined   c:\ccminer-windows\groestl_functions_quad.cu   489   1   ccminer
Error   19   error : expected an expression   c:\ccminer-windows\groestl_functions_quad.cu   489   1   ccminer
Error   20   error : expected a ")"   c:\ccminer-windows\groestl_functions_quad.cu   489   1   ccminer
Error   21   error : expected an expression   c:\ccminer-windows\groestl_functions_quad.cu   489   1   ccminer
Error   22   error : expected a ")"   c:\ccminer-windows\groestl_functions_quad.cu   489   1   ccminer
Error   23   error : expected an expression   c:\ccminer-windows\groestl_functions_quad.cu   496   1   ccminer
Error   24   error : expected a ")"   c:\ccminer-windows\groestl_functions_quad.cu   496   1   ccminer
Error   25   error : expected an expression   c:\ccminer-windows\groestl_functions_quad.cu   496   1   ccminer
Error   26   error : expected a ")"   c:\ccminer-windows\groestl_functions_quad.cu   496   1   ccminer
Error   27   error : expected an expression   c:\ccminer-windows\groestl_functions_quad.cu   497   1   ccminer
Error   28   error : expected a ")"   c:\ccminer-windows\groestl_functions_quad.cu   497   1   ccminer
Error   29   error : expected an expression   c:\ccminer-windows\groestl_functions_quad.cu   497   1   ccminer
Error   30   error : expected a ")"   c:\ccminer-windows\groestl_functions_quad.cu   497   1   ccminer
Error   31   error : expected an expression   c:\ccminer-windows\groestl_functions_quad.cu   501   1   ccminer
Error   32   error : expected a ")"   c:\ccminer-windows\groestl_functions_quad.cu   501   1   ccminer
Error   33   error : expected an expression   c:\ccminer-windows\groestl_functions_quad.cu   501   1   ccminer
Error   34   error : expected a ")"   c:\ccminer-windows\groestl_functions_quad.cu   501   1   ccminer
Error   35   error MSB3721: The command ""C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v6.5\bin\nvcc.exe" -gencode=arch=compute_50,code=\"sm_50,compute_50\" -gencode=arch=compute_52,code=\"sm_52,compute_52\" --use-local-env --cl-version 2013 -ccbin "C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\bin" -I. -Icompat -I"compat\curl-for-windows\curl\include" -Icompat\jansson -Icompat\getopt -Icompat\pthreads -I"compat\curl-for-windows\openssl\openssl\include" -I"compat\curl-for-windows\zlib" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v6.5\include" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v6.5\include" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v6.5\include" --keep --keep-dir Release -maxrregcount=80 --ptxas-options=-v --machine 32 --compile -cudart static --ptxas-options="-O3" -DWIN32 -DNDEBUG -D_CONSOLE -D_CRT_SECURE_NO_WARNINGS -DCURL_STATICLIB -DUSE_WRAPNVML -DSCRYPT_KECCAK512 -DSCRYPT_CHACHA -DSCRYPT_CHOOSE_COMPILETIME -D_MBCS -Xcompiler "/EHsc /W3 /nologo /Ox /Zi /MT " -o Release\cuda_groestlcoin.cu.obj "C:\ccminer-windows\cuda_groestlcoin.cu"" exited with code 2.   C:\Program Files (x86)\MSBuild\Microsoft.Cpp\v4.0\V120\BuildCustomizations\CUDA 6.5.targets   593   9   ccminer

needs to include cuda_vector.h instead of cuda_helper.h

I didn't modify the includes.
Furthermore, it compiles fine on Linux...
I know the build files on windows have been enhanced recently, maybe it's related to that.
I can't help because I never built on win.

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 960. (Read 2347664 times)