CCminer(SP-MOD) Modded GPU kernels.

[serg_25]

SP_  (and all ccminer devs) you made really great job.  lyra2v2 hashing extremely fast on such small cards.
after couple hours of playing with configs  i could reach:

Code:

[2015-09-17 17:38:00] GPU #0: GeForce GTX 750 Ti, 5025 Temp= 50C Fan= 41%

crazy speed.

[sp_]

But Djm-34's private kernal is doing 6.2MHASH on the 750ti. That is why lyra2v2 is not profitable to mine...

[sp_]

I think problem start from commit "Removed some more instructions in Blake512, Bmw512,lyra2 and lyra2v2".

but it's not this commit. must be one of the other commits

found it now. Will submitt a fix.

[Cryptozillah]

I switched build from r61 to r67 on all my Nvidia gear and i got slightly slower speed per card reported by ccminer but noticed that my rejected hash at NiceHash went down a bit.




I still have problems with two rigs shutting themself down from time to time but i think that is due to faulty gpu:s involved in those rigs.

One day of mining donated and on its way:
https://blockchain.info/tx/a13c767b0d1940e769fa72927cbfccbc3787c040df1a02951fb20efc55200180

[djm34]

But Djm-34's private kernal is doing 6.2MHASH on the 750ti. That is why lyra2v2 is not profitable to mine...

lolwut !? yeah it's my fault... sure  

 please spare me this kind of bullshit...
(if it was an attempt at motivating me to release it... it pretty much failed...  )

[sp_]

But Djm-34's private kernal is doing 6.2MHASH on the 750ti. That is why lyra2v2 is not profitable to mine...

lolwut !? yeah it's my fault... sure  
 please spare me this kind of bullshit...
(if it was an attempt at motivating me to release it... it pretty much failed...  )

You have the fastest kernal in the world. You are the best optimizer in the lyra2v2 algo for the 750ti.
TSIV has already reached 5,7MHASH.

Without opensourcing, someone else will make a faster kernal. It's just a matter of time.

[sp_]

I have submitted fixes in the atomic conversions. Lyra2v2 should now be fine.

[sp_]

I switched build from r61 to r67 on all my Nvidia gear and i got slightly slower speed per card reported by ccminer but noticed that my rejected hash at NiceHash went down a bit.
One day of mining donated and on its way:
https://blockchain.info/tx/a13c767b0d1940e769fa72927cbfccbc3787c040df1a02951fb20efc55200180

Thanks alot:) Release 68 will be faster.

[sp_]

I still have problems with two rigs shutting themself down from time to time but i think that is due to faulty gpu:s involved in those rigs.

The motherboard could be out of power. remove a card and see if it helps over time.

[sp_]

SP_  (and all ccminer devs) you made really great job.  lyra2v2 hashing extremely fast on such small cards.
after couple hours of playing with configs  i could reach:

Code:

[2015-09-17 17:38:00] GPU #0: GeForce GTX 750 Ti, 5025 Temp= 50C Fan= 41%

crazy speed.

Do you mind to share your config?

[Slava_K]

I have submitted fixes in the atomic conversions. Lyra2v2 should now be fine.

Works! Hashrate +100 kH on 980, +60 kH on 960!

[sp_]

The lyra2v2 defaults are not optimal.

on 750ti the default intensity is -X 16, but -X 5 -X 10 and -X15 are bether. -X 5 use less power too(less memory)..

Will change it.

Also the tpb hardcoded in the code needs to be adjusted. 980ti/980/960 probobly need totally different values.

[Slava_K]

The lyra defaults are not optimal.

on 750ti the default intensity is -X 16, but -X 5 -X 10 and -X15 are bether. -X 5 use less power too(less memory)..

Will change it.

Also the tpb hardcoded in the code needs to be adjusted. 980ti/980/960 probobly need totally different values.

On 980 and 960 my best is 7. Plz test...

[sp_]

The lyra defaults are not optimal.
on 750ti the default intensity is -X 16, but -X 5 -X 10 and -X15 are bether. -X 5 use less power too(less memory)..
Will change it.
Also the tpb hardcoded in the code needs to be adjusted. 980ti/980/960 probobly need totally different values.

On 980 and 960 my best is 7. Plz test...

I don't have a 980 here. But you can also try to change the tbp:

Lyra2REv2.cu:

   else if (strstr(props.name, "980"))
   {
      intensity = 256 * 256 * 7;
      tpb = 7;
   }


if your values are higher than 9 (950,960,970,980,980ti you need to increase the limit as well:
cuda_lyra2v2.cu:

#define TPB52 9
#define TPB50 16

[sp_]

I submitted bether defaults now.

I get 5150KHASH on 1388Mhz core and 2700 MEM. (750ti 38W tdp)
the 970 is peaking at 9950KHASH

Corrected to around 5000 on a pool(benchmark is showing high numbers)

but high rejects on the pool (duplicate shares).

Will submit a fix.

fixed

building release 68

[bensam1231]

But Djm-34's private kernal is doing 6.2MHASH on the 750ti. That is why lyra2v2 is not profitable to mine...

He is selling it? Last I heard it wasn't for sale. Although that would probably have something to do with the decline of L2V2.

But Djm-34's private kernal is doing 6.2MHASH on the 750ti. That is why lyra2v2 is not profitable to mine...

lolwut !? yeah it's my fault... sure  
 please spare me this kind of bullshit...
(if it was an attempt at motivating me to release it... it pretty much failed...  )

You have the fastest kernal in the world. You are the best optimizer in the lyra2v2 algo for the 750ti.
TSIV has already reached 5,7MHASH.

Without opensourcing, someone else will make a faster kernal. It's just a matter of time.

Telling you guys, 2% miner fee and just release stuff like that... Then improve it a bit, release it again.

I use .67 A good result for the GTH 960.


Ps. I do not know how correctly sent to the creator of the beer, let it be income for the day. For _SP transaction ID
7d52b98dc6448bdfdcd64fafc3fffb20a2d641a8e867ac591878cdd4e2b298bc

Do you have a watt meter? How much power is it using?

[sp_]

But Djm-34's private kernal is doing 6.2MHASH on the 750ti. That is why lyra2v2 is not profitable to mine...

He is selling it? Last I heard it wasn't for sale. Although that would probably have something to do with the decline of L2V2.

No, he doesn't like money. He wants to be the best. 

[sp_]

But Djm-34's private kernal is doing 6.2MHASH on the 750ti. That is why lyra2v2 is not profitable to mine...

He is selling it? Last I heard it wasn't for sale. Although that would probably have something to do with the decline of L2V2.

But Djm-34's private kernal is doing 6.2MHASH on the 750ti. That is why lyra2v2 is not profitable to mine...

lolwut !? yeah it's my fault... sure  
 please spare me this kind of bullshit...
(if it was an attempt at motivating me to release it... it pretty much failed...  )

You have the fastest kernal in the world. You are the best optimizer in the lyra2v2 algo for the 750ti.
TSIV has already reached 5,7MHASH.
Without opensourcing, someone else will make a faster kernal. It's just a matter of time.

Telling you guys, 2% miner fee and just release stuff like that... Then improve it a bit, release it again.

Nah, If you have a 20% faster miner than everybody else, you build a BIG farm and ROI a 750ti-980ti card every day  
And then you wait for the competition to wake up...

I think it would be good money to make a etherum miner for the 750ti. Such a coin. Such a profit.  

[djm34]

The lyra defaults are not optimal.

on 750ti the default intensity is -X 16, but -X 5 -X 10 and -X15 are bether. -X 5 use less power too(less memory)..

Will change it.

Also the tpb hardcoded in the code needs to be adjusted. 980ti/980/960 probobly need totally different values.

On 980 and 960 my best is 7. Plz test...

what kind of hashrate are you getting with these numbers ?
Also bear in mind that cuda only uses these number to produce a number of threads for each kernels. So if there are large discrepencies between the various kernel. you might run the full chain of algo on a number which can be rather different from the intensity you gave... meaning the reported hashrate (intensity/time) can be wrong...  

[sp_]

-Faster lyra2v2
-Faster quark
-tiny speedups in all algos.
-Some bugfixes

1.5.68(sp-MOD) is available here: (17-sep-2015)

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

The sourcecode is available here:

https://github.com/sp-hash/ccminer

Post your stats here. Card name/gpu clock/memclock