CCminer(SP-MOD) Modded GPU kernels.

[Namsbreh]

I am working on an update on the spreadcoin private miner #10. It's been 1.5 years since the last update. testing with cuda 8.0 now.

Great news, thank you.

The first tests shows that when my spreadcoin #9 is compiled with cuda 7.5 and cuda 8.0 I loose 4.5-7.25% (tested on maxwell 960M)

cuda 7.5: -7,25%
cuda 8.0: -4,47%

Do you have any Pascal cards to test with?

[bensam1231]

SP you should consider selling your DCR miner to Claymore. Dual mining effectively killed off exclusively mining DCR.

Not sure how much he's would be willing to give you, but depending on how much better it is compared to his current miner it could be lucrative.

If Epsylon is still improving SIA he should consider selling his improvements to Claymore instead of just releasing them as well. He essentially makes money off of how fast his miner is, so it's something he might be interested in (which is why a miner fee is great).

[sp_]

Do you have any Pascal cards to test with?

I have the 1070.

[sp_]

If anybody is interested i have buildt a version of Pallas m7 miner for windows. (https://github.com/pallas1/ccminer-m7-branch/)


https://github.com/sp-hash/ccminer/releases/tag/1.2

Compiled with cuda 6.5. Unmodified sourcecode from Pallas github exept lower the intensity to 256*256*32 to fit 2GB cards. (-X 32)

It does around 5MHASH on the 750ti

[Amph]

If anybody is interested i have buildt a version of Pallas m7 miner for windows. (https://github.com/pallas1/ccminer-m7-branch/)


https://github.com/sp-hash/ccminer/releases/tag/1.2

Compiled with cuda 6.5. Unmodified sourcecode from Pallas github exept lower the intensity to 256*256*32 to fit 2GB cards. (-X 32)

It does around 5MHASH on the 750ti

5MH on the 750ti and 15MH only on a 1070 does not sound right

[pallas]

If anybody is interested i have buildt a version of Pallas m7 miner for windows. (https://github.com/pallas1/ccminer-m7-branch/)


https://github.com/sp-hash/ccminer/releases/tag/1.2

Compiled with cuda 6.5. Unmodified sourcecode from Pallas github exept lower the intensity to 256*256*32 to fit 2GB cards. (-X 32)

It does around 5MHASH on the 750ti

5MH on the 750ti and 15MH only on a 1070 does not sound right

15.5 @125W on 1070
17 @180W on 980Ti

[Amph]

If anybody is interested i have buildt a version of Pallas m7 miner for windows. (https://github.com/pallas1/ccminer-m7-branch/)


https://github.com/sp-hash/ccminer/releases/tag/1.2

Compiled with cuda 6.5. Unmodified sourcecode from Pallas github exept lower the intensity to 256*256*32 to fit 2GB cards. (-X 32)

It does around 5MHASH on the 750ti

5MH on the 750ti and 15MH only on a 1070 does not sound right

15.5 @125W on 1070
17 @180W on 980Ti

what is the usage on a 1070?

[pallas]

If anybody is interested i have buildt a version of Pallas m7 miner for windows. (https://github.com/pallas1/ccminer-m7-branch/)


https://github.com/sp-hash/ccminer/releases/tag/1.2

Compiled with cuda 6.5. Unmodified sourcecode from Pallas github exept lower the intensity to 256*256*32 to fit 2GB cards. (-X 32)

It does around 5MHASH on the 750ti

5MH on the 750ti and 15MH only on a 1070 does not sound right

15.5 @125W on 1070
17 @180W on 980Ti

what is the usage on a 1070?

99-100
But I think the power usage fluctuates too much, probably the power used is less than 125w.

[Namsbreh]

Do you have any Pascal cards to test with?

I have the 1070.

Me too!

[Amph]

If anybody is interested i have buildt a version of Pallas m7 miner for windows. (https://github.com/pallas1/ccminer-m7-branch/)


https://github.com/sp-hash/ccminer/releases/tag/1.2

Compiled with cuda 6.5. Unmodified sourcecode from Pallas github exept lower the intensity to 256*256*32 to fit 2GB cards. (-X 32)

It does around 5MHASH on the 750ti

5MH on the 750ti and 15MH only on a 1070 does not sound right

15.5 @125W on 1070
17 @180W on 980Ti

what is the usage on a 1070?

99-100
But I think the power usage fluctuates too much, probably the power used is less than 125w.

based on this and the hash of the 980ti, it should do nearly 20MH or around that, but i would not be surprised if it can do more, like in lbry where the 980ti is a joke in comparison

[antantti]

based on this and the hash of the 980ti, it should do nearly 20MH or around that, but i would not be surprised if it can do more, like in lbry where the 980ti is a joke in comparison

How much is 980ti doing on lbry?

[mirny]

based on this and the hash of the 980ti, it should do nearly 20MH or around that, but i would not be surprised if it can do more, like in lbry where the 980ti is a joke in comparison

How much is 980ti doing on lbry?

for about 200/220Mhs, depends on your settings

[bensam1231]

based on this and the hash of the 980ti, it should do nearly 20MH or around that, but i would not be surprised if it can do more, like in lbry where the 980ti is a joke in comparison

How much is 980ti doing on lbry?

for about 200/220Mhs, depends on your settings

980ti should always be doing about 10% less then a 1070. If it isn't, that means something isn't working properly. 1070 will usually be double of a 970, otherwise, once again, something isn't working properly.

Lyra for instance, a 1070 doubles a 970s performance (give or take).

[Nikolaj]

In my experience a 1070 isn't always 2x, better 1.85 if you compare it to a 970.

It depends a lot by the algos. I've got a very complex chart and these are the ratios.. (personal use, I can't share it right now)

[sp_]

I see that you have used ccminer 1.8 pruvot to do your comparison. My "All nicehash algos optimized" have a 62% faster quark kernal included.


Sp-Mod #5 ( All nicehash algos optimized) default intensities hashrates in MHASH/s. (980ti 1290 coreclock/1650 memclock) faster than ccminer 1.7.5 cuda 7.5 32 build

nist5: 43.80 (+17%)
quark: 27.85 (+61,91%)
lyra2v2 37.10
x11:   12.85 (+18%)
x13:   10.27 (+19.97)
x15:    8.83 (+18.52 )
neoscrypt: 0.99 (+73.68%)

[Nikolaj]

I can insert your data in my db if you wish, have you got some numbers for a 1070 / 1080 too ? 

[bensam1231]

In my experience a 1070 isn't always 2x, better 1.85 if you compare it to a 970.

It depends a lot by the algos. I've got a very complex chart and these are the ratios.. (personal use, I can't share it right now)

It's not about what is already written, rather new optimizations that are made for Pascal specifically. The 1070 has been out for a little over a month. Those algos weren't written with Pascal in mind. Lyra with the improvements written by Nanashi were made specifically for the 1070 and it shows.

If it can be double for Lyra, it can be double for other algos... save Dagger as that's a very special case scenario. There are very few algos that peg the memory controller.

[sp_]

In my experience a 1070 isn't always 2x, better 1.85 if you compare it to a 970.
It depends a lot by the algos. I've got a very complex chart and these are the ratios.. (personal use, I can't share it right now)

It's not about what is already written, rather new optimizations that are made for Pascal specifically. The 1070 has been out for a little over a month. Those algos weren't written with Pascal in mind. Lyra with the improvements written by Nanashi were made specifically for the 1070 and it shows.
If it can be double for Lyra, it can be double for other algos... save Dagger as that's a very special case scenario. There are very few algos that peg the memory controller.

I have recieved my first 1070 rig now. I think I will do the pascal mod as I did with the Maxwell. But with a 2% fee. Alot of work is waiting..

[Amph]

In my experience a 1070 isn't always 2x, better 1.85 if you compare it to a 970.
It depends a lot by the algos. I've got a very complex chart and these are the ratios.. (personal use, I can't share it right now)

It's not about what is already written, rather new optimizations that are made for Pascal specifically. The 1070 has been out for a little over a month. Those algos weren't written with Pascal in mind. Lyra with the improvements written by Nanashi were made specifically for the 1070 and it shows.
If it can be double for Lyra, it can be double for other algos... save Dagger as that's a very special case scenario. There are very few algos that peg the memory controller.

I have recieved my first 1070 rig now. I think I will do the pascal mod as I did with the Maxwell. But with a 2% fee. Alot of work is waiting..

received? you mean built? 2% like claymore can be okish, if you can boost all the algo hashrate for pascal

[sp_]

received? you mean built? 2% like claymore can be okish, if you can boost all the algo hashrate for pascal

Yes built.

All algos optimized, your 2% will finance my hard work. Everybody will win.