CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 745.

Epsylon3

legendary

Activity: 1484

Merit: 1082

ccminer/cpuminer developer

Quote from: sp_ on March 08, 2016, 01:33:33 PM

Quote from: s7icky on March 08, 2016, 12:53:34 PM

Quote from: sp_ on March 08, 2016, 10:10:56 AM

My fork was based on 1.7.3 wich have issues. I need to refork 1.7.4 and make a new build

but you sold it and knew it had issues? LOLOL money hungry asshole once again.

I have only modded the cuda code. My subscribers will get updates. Fixes where submitted in the stratum/c++ code afterwards(github)

there was not a bunch of commits since the 1.7.3 :p

CrazyRussian

newbie

Activity: 10

Merit: 0

Quote from: sp_ on March 10, 2016, 04:02:08 AM

Only 0.1 BTC donation for the windows subscription guys. Mine while it's hot.

Sent you donation 0.1BTC

TxId: 3a7c0032fab87d98d2e956422385339186d57bf7679455698d29c9e5ff2e37fb

fenomenhaa

sr. member

Activity: 248

Merit: 250

Quote from: sp_ on March 10, 2016, 03:29:03 AM

I have sendt decred sp-mod #2 to the donators. This is forked from 1.7.4 and contains some bugfixes and a new kernal. I get 10-15% higher rates on the pool with this version compared to release #1.

Happy mining

Recieved Thank you Sp

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Here are my testrigs on yiimp.ccminer.org with sp-mod #2 mining decred

Only 0.1 BTC donation for the windows subscription guys. Mine while it's hot.

Yiimp is currently paying 0.0090/GHASH

My profit is up from

0,18BTC / DAY (20GHASH)

to

0,225 / DAY (25GHASH)

chrysophylax

legendary

Activity: 2940

Merit: 1091

--- ChainWorks Industries ---

Quote from: sp_ on March 10, 2016, 03:29:03 AM

I have sendt decred sp-mod #2 to the donators. This is forked from 1.7.4 and contains some bugfixes and a new kernal. I get 10-15% higher rates on the pool with this version compared to release #1.

Happy mining

tanx sp ...

#crysx

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

I have sendt decred sp-mod #2 to the donators. This is forked from 1.7.4 and contains some bugfixes and a new kernal. I get 10-15% higher rates on the pool with this version compared to release #1.

Happy mining

chrysophylax

legendary

Activity: 2940

Merit: 1091

--- ChainWorks Industries ---

Quote from: bensam1231 on March 10, 2016, 02:03:32 AM

Quote from: chrysophylax on March 10, 2016, 01:27:07 AM

Quote from: sp_ on March 09, 2016, 01:52:46 AM

Quote

We have compiled ccMiner 1.7.4 tpruvot’s fork Windows binary in a version with the optimal settings for mining Vanillacoin (VNL) on more recent Nvidia-based GPUs as suggested by Alexis Provos and with his performance optimizations as implemented by tpruvot in the latest code. This Windows binary is a 32-bit one compiled with CUDa 6.5 and only with Compute 3.5 support as apparently this provides the optimal performance at least on Maxwell-based Nvidia GPUs (Faster than using Compute 5.2 on them). Below you can also see a comparison that we did with Nvidia GeForce GTX 970 from Gigabyte (WF3OC) and a reference design Gigabyte GTX 980 Ti to see the difference in performance with the regular version when using CUDA 7.5 and Compute 5.2 and with the optimized CUDA 6.5 and Compute 3.5 version available for download below. Do note that although this binary will work with other algorithms as well as the Blake 256 8-rounds used by VanillaCoin it may not provide the optimal performance for other algorithms as it is targeted to provide the best VNL mining hashrate.

This is interesting thanks for the tip. Never thought compute 3.5 could be bether. My kernal is tuned for compute 5.0 and cuda 7.5 32bit. I will be surpriced if it goes faster With compute 3.5, but Worth a try. But there is a difference between vanillacoin and decred...

(The article write 7% more hash with 12% more power. with compute 3.5)
(The decred sp-mod private #1 has 10-15% more hash with 0% more power)

sp ...

i had already mentioned this above requisite to you when testing ... did you not see what i wrote? ...

cuda 6.5 compute 3.5 - which is the optimum hashrate that particular miner gets under those compilation parameters ...

but most have an environment of cuda 7.5 and compile with compute 5.2 with your fork anyway - and my environment is the same ... so i cant use anything that does not have any real gain from cuda 7.5 compute 5.2 ...

this is why the version you will bring out - that DOES compile and increase hashrate in cuda 7.5 compute 5.2 - will be the one that is worth the effort to compile ...

im still interested in the optimizations done in that environment - NOT - in cuda 6.5 compute 3.5 ... no way am i reverting back to that ...

so if you can still bring out optimizations with the latest 'stuff' - ill still be interested ...

tanx ...

#crysx

So this is some ass backwards post to convince him to use some other version of compute that fits your needs? I mentioned it because cryptominingblog did testing on it. It's a special case. Nothing to do with you unless you mentioned this exact same thing somewhere else.

Seriously bro... Who cares what you're interested in and why should we?

do eveyone a favour bensam ( including your wife and kids ) and die in a corner somewhere ...

sp KNOWS what im talking about - its got nothing to do with YOU ...

'seriously bro' - im not interested in your musings nor the attacks that you continuously regurgitate on a consistent level ... damn! ...

you are so lucky you LIVE behind a profile - and not in my face right this minute - or you would have a piece of mind the australian way ... you are just bloody pathetic ...

there are things that get done outside of the public eye - and you of all people should know that ... always complaining and crying about private this and farm that ...

who the hell do you think 'one' of those people who were testing these optimizations were - LONG before they hit the critics? ...

twat! ...

#crysx

bensam1231

legendary

Activity: 1764

Merit: 1024

Quote from: chrysophylax on March 10, 2016, 01:27:07 AM

Quote from: sp_ on March 09, 2016, 01:52:46 AM

Quote

We have compiled ccMiner 1.7.4 tpruvot’s fork Windows binary in a version with the optimal settings for mining Vanillacoin (VNL) on more recent Nvidia-based GPUs as suggested by Alexis Provos and with his performance optimizations as implemented by tpruvot in the latest code. This Windows binary is a 32-bit one compiled with CUDa 6.5 and only with Compute 3.5 support as apparently this provides the optimal performance at least on Maxwell-based Nvidia GPUs (Faster than using Compute 5.2 on them). Below you can also see a comparison that we did with Nvidia GeForce GTX 970 from Gigabyte (WF3OC) and a reference design Gigabyte GTX 980 Ti to see the difference in performance with the regular version when using CUDA 7.5 and Compute 5.2 and with the optimized CUDA 6.5 and Compute 3.5 version available for download below. Do note that although this binary will work with other algorithms as well as the Blake 256 8-rounds used by VanillaCoin it may not provide the optimal performance for other algorithms as it is targeted to provide the best VNL mining hashrate.

This is interesting thanks for the tip. Never thought compute 3.5 could be bether. My kernal is tuned for compute 5.0 and cuda 7.5 32bit. I will be surpriced if it goes faster With compute 3.5, but Worth a try. But there is a difference between vanillacoin and decred...

(The article write 7% more hash with 12% more power. with compute 3.5)
(The decred sp-mod private #1 has 10-15% more hash with 0% more power)

sp ...

i had already mentioned this above requisite to you when testing ... did you not see what i wrote? ...

cuda 6.5 compute 3.5 - which is the optimum hashrate that particular miner gets under those compilation parameters ...

but most have an environment of cuda 7.5 and compile with compute 5.2 with your fork anyway - and my environment is the same ... so i cant use anything that does not have any real gain from cuda 7.5 compute 5.2 ...

this is why the version you will bring out - that DOES compile and increase hashrate in cuda 7.5 compute 5.2 - will be the one that is worth the effort to compile ...

im still interested in the optimizations done in that environment - NOT - in cuda 6.5 compute 3.5 ... no way am i reverting back to that ...

so if you can still bring out optimizations with the latest 'stuff' - ill still be interested ...

tanx ...

#crysx

So this is some ass backwards post to convince him to use some other version of compute that fits your needs? I mentioned it because cryptominingblog did testing on it. It's a special case. Nothing to do with you unless you mentioned this exact same thing somewhere else.

Seriously bro... Who cares what you're interested in and why should we?

chrysophylax

legendary

Activity: 2940

Merit: 1091

--- ChainWorks Industries ---

Quote from: sp_ on March 09, 2016, 01:52:46 AM

Quote

We have compiled ccMiner 1.7.4 tpruvot’s fork Windows binary in a version with the optimal settings for mining Vanillacoin (VNL) on more recent Nvidia-based GPUs as suggested by Alexis Provos and with his performance optimizations as implemented by tpruvot in the latest code. This Windows binary is a 32-bit one compiled with CUDa 6.5 and only with Compute 3.5 support as apparently this provides the optimal performance at least on Maxwell-based Nvidia GPUs (Faster than using Compute 5.2 on them). Below you can also see a comparison that we did with Nvidia GeForce GTX 970 from Gigabyte (WF3OC) and a reference design Gigabyte GTX 980 Ti to see the difference in performance with the regular version when using CUDA 7.5 and Compute 5.2 and with the optimized CUDA 6.5 and Compute 3.5 version available for download below. Do note that although this binary will work with other algorithms as well as the Blake 256 8-rounds used by VanillaCoin it may not provide the optimal performance for other algorithms as it is targeted to provide the best VNL mining hashrate.

This is interesting thanks for the tip. Never thought compute 3.5 could be bether. My kernal is tuned for compute 5.0 and cuda 7.5 32bit. I will be surpriced if it goes faster With compute 3.5, but Worth a try. But there is a difference between vanillacoin and decred...

(The article write 7% more hash with 12% more power. with compute 3.5)
(The decred sp-mod private #1 has 10-15% more hash with 0% more power)

sp ...

i had already mentioned this above requisite to you when testing ... did you not see what i wrote? ...

cuda 6.5 compute 3.5 - which is the optimum hashrate that particular miner gets under those compilation parameters ...

but most have an environment of cuda 7.5 and compile with compute 5.2 with your fork anyway - and my environment is the same ... so i cant use anything that does not have any real gain from cuda 7.5 compute 5.2 ...

this is why the version you will bring out - that DOES compile and increase hashrate in cuda 7.5 compute 5.2 - will be the one that is worth the effort to compile ...

im still interested in the optimizations done in that environment - NOT - in cuda 6.5 compute 3.5 ... no way am i reverting back to that ...

so if you can still bring out optimizations with the latest 'stuff' - ill still be interested ...

tanx ...

#crysx

bensam1231

legendary

Activity: 1764

Merit: 1024

So other people have about -25% speed consistently pool side compared to miners on Decred?

I understand there is variance, but this is consistently over days. I have about 98% accepted shares. Seems to be the same on both getwork and stratum.

Vaccomondus

full member

Activity: 224

Merit: 100

this does not support blake2s?

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote

We have compiled ccMiner 1.7.4 tpruvot’s fork Windows binary in a version with the optimal settings for mining Vanillacoin (VNL) on more recent Nvidia-based GPUs as suggested by Alexis Provos and with his performance optimizations as implemented by tpruvot in the latest code. This Windows binary is a 32-bit one compiled with CUDa 6.5 and only with Compute 3.5 support as apparently this provides the optimal performance at least on Maxwell-based Nvidia GPUs (Faster than using Compute 5.2 on them). Below you can also see a comparison that we did with Nvidia GeForce GTX 970 from Gigabyte (WF3OC) and a reference design Gigabyte GTX 980 Ti to see the difference in performance with the regular version when using CUDA 7.5 and Compute 5.2 and with the optimized CUDA 6.5 and Compute 3.5 version available for download below. Do note that although this binary will work with other algorithms as well as the Blake 256 8-rounds used by VanillaCoin it may not provide the optimal performance for other algorithms as it is targeted to provide the best VNL mining hashrate.

This is interesting thanks for the tip. Never thought compute 3.5 could be bether. My kernal is tuned for compute 5.0 and cuda 7.5 32bit. I will be surpriced if it goes faster With compute 3.5, but Worth a try. But there is a difference between vanillacoin and decred...

(The article write 7% more hash with 12% more power. with compute 3.5)
(The decred sp-mod private #1 has 10-15% more hash with 0% more power)

bensam1231

legendary

Activity: 1764

Merit: 1024

Quote from: sp_ on March 09, 2016, 12:25:01 AM

Quote from: bensam1231 on March 08, 2016, 08:37:58 PM

Yeah, similar issues with SP's private decred. I'm getting 100% accept with stratum. Getwork is still buggy, just like with Tpruvots.

Something I've been reading up, from the Suprnova thread on the Decred forums:

Quote from: ocminer

This is a confirmed bug in cgminer on windows. It affects all pools. This flags seem to help:

--no-submit-stale --scan-time 2 --queue 0

Unknown-work means you are sending work back to the pool that was never requested - thats why it gets rejected, on windows cgminer seems to wrong calculation of the hash sometimes which results in this behavior.

It will probably get fixed in one of the future releases, or simply try the flags above.

However, CCminer doesn't support any of the above options besides 'scan time', so you can't tweak the other two options.

Putting all of the issues with accepted shares aside. Does anyone have problems with the miner speed not matching pool speed? Tpruvot and SP's miner show 25% faster hashrate then the pool is showing, even with close to 100% accepted shares. The only place I've seen the pool speed show up correctly is yiimp. MNpool also has the same speed issue. This is averaged over about 24hours so it's not a outlier.

SP, not sure if you saw cryptominingblog today, but they were talking about improving performance by using compute 3.5, at least it works that way with Blake-14, have you tried it with Decred? I assume it may yield similar results as Decred also uses Blake.

http://cryptomining-blog.com/

My kernal is compiled with cuda 7.5. Cuda 6.5 is 10% slower

Compute 3.5 on 6.5. You can read the post on the link.

Quote

We have compiled ccMiner 1.7.4 tpruvot’s fork Windows binary in a version with the optimal settings for mining Vanillacoin (VNL) on more recent Nvidia-based GPUs as suggested by Alexis Provos and with his performance optimizations as implemented by tpruvot in the latest code. This Windows binary is a 32-bit one compiled with CUDa 6.5 and only with Compute 3.5 support as apparently this provides the optimal performance at least on Maxwell-based Nvidia GPUs (Faster than using Compute 5.2 on them). Below you can also see a comparison that we did with Nvidia GeForce GTX 970 from Gigabyte (WF3OC) and a reference design Gigabyte GTX 980 Ti to see the difference in performance with the regular version when using CUDA 7.5 and Compute 5.2 and with the optimized CUDA 6.5 and Compute 3.5 version available for download below. Do note that although this binary will work with other algorithms as well as the Blake 256 8-rounds used by VanillaCoin it may not provide the optimal performance for other algorithms as it is targeted to provide the best VNL mining hashrate.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: bensam1231 on March 08, 2016, 08:37:58 PM

Yeah, similar issues with SP's private decred. I'm getting 100% accept with stratum. Getwork is still buggy, just like with Tpruvots.

Something I've been reading up, from the Suprnova thread on the Decred forums:

Quote from: ocminer

This is a confirmed bug in cgminer on windows. It affects all pools. This flags seem to help:

--no-submit-stale --scan-time 2 --queue 0

Unknown-work means you are sending work back to the pool that was never requested - thats why it gets rejected, on windows cgminer seems to wrong calculation of the hash sometimes which results in this behavior.

It will probably get fixed in one of the future releases, or simply try the flags above.

However, CCminer doesn't support any of the above options besides 'scan time', so you can't tweak the other two options.

Putting all of the issues with accepted shares aside. Does anyone have problems with the miner speed not matching pool speed? Tpruvot and SP's miner show 25% faster hashrate then the pool is showing, even with close to 100% accepted shares. The only place I've seen the pool speed show up correctly is yiimp. MNpool also has the same speed issue. This is averaged over about 24hours so it's not a outlier.

SP, not sure if you saw cryptominingblog today, but they were talking about improving performance by using compute 3.5, at least it works that way with Blake-14, have you tried it with Decred? I assume it may yield similar results as Decred also uses Blake.

http://cryptomining-blog.com/

My kernal is compiled with cuda 7.5. Cuda 6.5 is 10% slower

bensam1231

legendary

Activity: 1764

Merit: 1024

Yeah, similar issues with SP's private decred. I'm getting 100% accept with stratum. Getwork is still buggy, just like with Tpruvots.

Something I've been reading up, from the Suprnova thread on the Decred forums:

Quote from: ocminer

This is a confirmed bug in cgminer on windows. It affects all pools. This flags seem to help:

--no-submit-stale --scan-time 2 --queue 0

Unknown-work means you are sending work back to the pool that was never requested - thats why it gets rejected, on windows cgminer seems to wrong calculation of the hash sometimes which results in this behavior.

It will probably get fixed in one of the future releases, or simply try the flags above.

However, CCminer doesn't support any of the above options besides 'scan time', so you can't tweak the other two options.

Putting all of the issues with accepted shares aside. Does anyone have problems with the miner speed not matching pool speed? Tpruvot and SP's miner show 25% faster hashrate then the pool is showing, even with close to 100% accepted shares. The only place I've seen the pool speed show up correctly is yiimp. MNpool also has the same speed issue. This is averaged over about 24hours so it's not a outlier.

SP, not sure if you saw cryptominingblog today, but they were talking about improving performance by using compute 3.5, at least it works that way with Blake-14, have you tried it with Decred? I assume it may yield similar results as Decred also uses Blake.

http://cryptomining-blog.com/

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: malekbaba on March 08, 2016, 03:04:18 PM

Quote from: sp_ on March 08, 2016, 01:33:33 PM

Quote from: s7icky on March 08, 2016, 12:53:34 PM

Quote from: sp_ on March 08, 2016, 10:10:56 AM

My fork was based on 1.7.3 wich have issues. I need to refork 1.7.4 and make a new build

but you sold it and knew it had issues? LOLOL money hungry asshole once again.

I have only modded the cuda code. My subscribers will get updates. Fixes where submitted in the stratum/c++ code afterwards(github)

which algo work best for 750ti cards? Can you please tell us the hashes a 750ti can make with your miner for vanilla and decred?

On a normalised hashrate basis lyra2v2 performs better than most other algos compared to other maxwell cards.
That doesn't mean it's more profitable than other algos, it's just more profitable than other cards when mining lyra2v2.
I haven't compared vanilla or decred.

malekbaba

legendary

Activity: 1526

Merit: 1026

Quote from: sp_ on March 08, 2016, 01:33:33 PM

Quote from: s7icky on March 08, 2016, 12:53:34 PM

Quote from: sp_ on March 08, 2016, 10:10:56 AM

My fork was based on 1.7.3 wich have issues. I need to refork 1.7.4 and make a new build

but you sold it and knew it had issues? LOLOL money hungry asshole once again.

I have only modded the cuda code. My subscribers will get updates. Fixes where submitted in the stratum/c++ code afterwards(github)

which algo work best for 750ti cards? Can you please tell us the hashes a 750ti can make with your miner for vanilla and decred?

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: s7icky on March 08, 2016, 12:53:34 PM

Quote from: sp_ on March 08, 2016, 10:10:56 AM

My fork was based on 1.7.3 wich have issues. I need to refork 1.7.4 and make a new build

but you sold it and knew it had issues? LOLOL money hungry asshole once again.

I have only modded the cuda code. My subscribers will get updates. Fixes where submitted in the stratum/c++ code afterwards(github)

s7icky

member

Activity: 106

Merit: 10

Quote from: sp_ on March 08, 2016, 10:10:56 AM

My fork was based on 1.7.3 wich have issues. I need to refork 1.7.4 and make a new build

but you sold it and knew it had issues? LOLOL money hungry asshole once again.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

My fork was based on 1.7.3 wich have issues. I need to refork 1.7.4 and make a new build

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 745. (Read 2347659 times)