CCminer(SP-MOD) Modded GPU kernels.

[bensam1231]

There is no optimization to do on pascal yet, its almost the same hardware, reduced in size

Maybe the next cuda versions will show some hidden functions, but im not confident about that

Maybe 'optimize' wasn't the right word... How about bug fixing? Skein only scales +50% on 1070 compared to 970, when it should be closer to +100%.

so in the last day :

zpool some segfaults issues
decred team ask me to check gominer
mrM4D ask me quark changes

and im working on neoscrypt... not enough ?

i already checked this SIA several times... but i forgot again why i stopped...

I don't know, who pays more and what's needed more?


Still need a dual miner for Eth + other algo. Apparently Claymore now has a Eth + Sia miner. We need something similiar for Nvidia. If you're wondering where the donations are, it's where the money is.

I just wanted to bring it to your attention. You can implement if it makes sense at your leisure. I thought you might stand to gain some revenue from a bunch of people jumping on the SIA bandwagon with the dual miner.

Yup... Earning me more money makes me inclined to donate (I assume it works this way for other miners as well). If there was a miner fee, there would be a direct correlation between the two. Dual miner has been talked about here for close to seven months, especially after Claymore released his. Still no one biting at it... Eth is going to be PoS before one shows up.

There is also Cryptonote floating around. The 'elusive' Cryptonote no one has really touched in years that has craptons of volume floating around multiple coins.

[scryptr]

SP_ AND EPSYLON3--

SiaCoin uses the Blake2b  algorithm.  Would either one of you, or both, be able to incorporate the algo into CCminer?  Several othe Blake algos are currently supported, and they have been given CUDA enhanced optimization.  My 980ti hashes at about 1.3GH/s, better than the reported 1.0GH/s for a 290X.  Gominer, reported on by CryptoMining-Blog today, uses only Open-cl  coding, and CUDA enhanced code should really improve the performance.  Gominer runs smoothly, no problems, and is open source. 

Blake2b would be a good addition to the CCminer list of supported algos. SiaCoin is a coin with more value than just another clone coin.  Like ETH, it offers a practical side, the use of the blockchain for data storage in this case.

What do you say?       --scryptr

[Amph]

the problem with sia is that it has no pool, so unless you have 10 rig, it's pointless to mine, diff is too high at the moment, for a little farm

[antantti]

the problem with sia is that it has no pool, so unless you have 10 rig, it's pointless to mine, diff is too high at the moment, for a little farm

http://sia.nanopool.org/

[Amph]

the problem with sia is that it has no pool, so unless you have 10 rig, it's pointless to mine, diff is too high at the moment, for a little farm

http://sia.nanopool.org/

oh tnx didn't know that there was one, but the diff is even higher now, almost on par with other top altcoin, in terms of profit

it does not work with the cuda version

[scryptr]

the problem with sia is that it has no pool, so unless you have 10 rig, it's pointless to mine, diff is too high at the moment, for a little farm

http://sia.nanopool.org/

oh tnx didn't know that there was one, but the diff is even higher now, almost on par with other top altcoin, in terms of profit

it does not work with the cuda version

ODDS ARE BETTER THAN WINNING THE LOTTERY--

I solo-mined SiaCoin overnight with Gominer  (open-cl).  When I woke up, my 980ti was mining at 1.4GH/s.  It should hit a block within 2-4 weeks, and a block is worth about 0.25BTC.  I am going back to Lyra2v2 after I type this, though.  I also need to check my $2 lottery ticket to see if I am a new millionaire, the drawing was last night.  No SiaCoin in the wallet this morning...

--scryptr

[Epsylon3]

Look like the 1070 mem perf was fixed in driver 368.69, i now get 950 kH/s on windows (+30%)

but the 1080 seems worst sigh

GPU #0: Gigabyte GTX 1080, 391.72 kH/s
accepted: 15/15 (diff 0.003), 1324.49 kH/s yes!
GPU #1: GeForce GTX 1070, 949.84 kH/s
accepted: 16/16 (diff 0.004), 1325.44 kH/s yes!

[Amph]

Look like the 1070 mem perf was fixed in driver 368.69, i now get 950 kH/s on windows (+30%)

but the 1080 seems worst sigh

GPU #0: Gigabyte GTX 1080, 391.72 kH/s
accepted: 15/15 (diff 0.003), 1324.49 kH/s yes!
GPU #1: GeForce GTX 1070, 949.84 kH/s
accepted: 16/16 (diff 0.004), 1325.44 kH/s yes!

oh nice, neoscrypt right? and windows 10 or 7? someone can confirm that the same perf are there for win 7?

also can you try to oc +200 first and then +400 on the core? it should be able to hit almost 1300

[Epsylon3]

its the FE and 30°c today but yep i guess 1MH is reachable with the "normal" P0 mem clock

http://ccminer.org/preview/ccminer.exe

[Amph]

its the FE and 30°c today but yep i guess 1MH is reachable with the "normal" P0 mem clock

http://ccminer.org/preview/ccminer.exe

yeah it's 30 here too, but just a test to see how much the oc will perform, you can try for 5 min, i doubt you will screw the fe edition

[Epsylon3]

its not the problem, the problem is the perf are reduced with the thermal limits

nvm, as you can see the problem was not in ccminer code

[bensam1231]

Look like the 1070 mem perf was fixed in driver 368.69, i now get 950 kH/s on windows (+30%)

but the 1080 seems worst sigh

GPU #0: Gigabyte GTX 1080, 391.72 kH/s
accepted: 15/15 (diff 0.003), 1324.49 kH/s yes!
GPU #1: GeForce GTX 1070, 949.84 kH/s
accepted: 16/16 (diff 0.004), 1325.44 kH/s yes!

Already installed the driver, haven't checked Ethereum and Neo hash as other 'users' reported it didn't fix it. Shame on me for listening to them I guess.

NeoS gets around 1000~ Mh/s on W8.1 x64. Using DJMs last build.

Definitely still room for tuning. Should be closer to 1200. Watching the GPUs, none of them are 'maxed' out and sit at about 85% utilization and MCU is also at about 80% utilization. This may be a intensity issue, but I kind of doubt it considering what Pallas was saying. Intensity seems about ballpark with a little bit of testing as well.

[Amph]

its not the problem, the problem is the perf are reduced with the thermal limits

nvm, as you can see the problem was not in ccminer code

mmh ok, btw are you on win 10 with those perf?

[giagge]

Look like the 1070 mem perf was fixed in driver 368.69, i now get 950 kH/s on windows (+30%)

but the 1080 seems worst sigh

GPU #0: Gigabyte GTX 1080, 391.72 kH/s
accepted: 15/15 (diff 0.003), 1324.49 kH/s yes!
GPU #1: GeForce GTX 1070, 949.84 kH/s
accepted: 16/16 (diff 0.004), 1325.44 kH/s yes!

Wow , with my nvidia gtx 980 OC 1500gpu i have now 780 kh/s on windows 10 x64 on neoscrypt with FTC .

[Vegetablepin]

Look like the 1070 mem perf was fixed in driver 368.69, i now get 950 kH/s on windows (+30%)

but the 1080 seems worst sigh

GPU #0: Gigabyte GTX 1080, 391.72 kH/s
accepted: 15/15 (diff 0.003), 1324.49 kH/s yes!
GPU #1: GeForce GTX 1070, 949.84 kH/s
accepted: 16/16 (diff 0.004), 1325.44 kH/s yes!

Already installed the driver, haven't checked Ethereum and Neo hash as other 'users' reported it didn't fix it. Shame on me for listening to them I guess.

NeoS gets around 1000~ Mh/s on W8.1 x64. Using DJMs last build.

Definitely still room for tuning. Should be closer to 1200. Watching the GPUs, none of them are 'maxed' out and sit at about 85% utilization and MCU is also at about 80% utilization. This may be a intensity issue, but I kind of doubt it considering what Pallas was saying. Intensity seems about ballpark with a little bit of testing as well.

Well I installed new driver, then at some point it said I was on .38 driver again after a few restarts. Reinstalled .69 and 1070 is hashing at 600~(w10), I guess custom clean install is needed instead of express.

[bensam1231]

Just tried Ethereum, it's still broken in OCL and in Cuda mode. I get a queuebuffer error in OCL and in Cuda I get a 'unspecified launch failure' after it finishes building the dag.

Perhaps a chunk style loading might be needed (even if devices have enough memory) on W7/8. Since in W10, the newer WDDM version increases the chunk size.

[Epsylon3]

I dont have win10... for me its an unstable and unfinished os... i stay on 7 for a few years

[Amph]

Look like the 1070 mem perf was fixed in driver 368.69, i now get 950 kH/s on windows (+30%)

but the 1080 seems worst sigh

GPU #0: Gigabyte GTX 1080, 391.72 kH/s
accepted: 15/15 (diff 0.003), 1324.49 kH/s yes!
GPU #1: GeForce GTX 1070, 949.84 kH/s
accepted: 16/16 (diff 0.004), 1325.44 kH/s yes!

Already installed the driver, haven't checked Ethereum and Neo hash as other 'users' reported it didn't fix it. Shame on me for listening to them I guess.

NeoS gets around 1000~ Mh/s on W8.1 x64. Using DJMs last build.

Definitely still room for tuning. Should be closer to 1200. Watching the GPUs, none of them are 'maxed' out and sit at about 85% utilization and MCU is also at about 80% utilization. This may be a intensity issue, but I kind of doubt it considering what Pallas was saying. Intensity seems about ballpark with a little bit of testing as well.

Well I installed new driver, then at some point it said I was on .38 driver again after a few restarts. Reinstalled .69 and 1070 is hashing at 600~(w10), I guess custom clean install is needed instead of express.

so you fixed it now? they should do the same on win 10 like they do on win7/8

also on mcu utilization being only 80%, fb usage which is the sam eis at 80% on maxwell too, so that can not be improved

[Vegetablepin]

Look like the 1070 mem perf was fixed in driver 368.69, i now get 950 kH/s on windows (+30%)

but the 1080 seems worst sigh

GPU #0: Gigabyte GTX 1080, 391.72 kH/s
accepted: 15/15 (diff 0.003), 1324.49 kH/s yes!
GPU #1: GeForce GTX 1070, 949.84 kH/s
accepted: 16/16 (diff 0.004), 1325.44 kH/s yes!

Already installed the driver, haven't checked Ethereum and Neo hash as other 'users' reported it didn't fix it. Shame on me for listening to them I guess.

NeoS gets around 1000~ Mh/s on W8.1 x64. Using DJMs last build.

Definitely still room for tuning. Should be closer to 1200. Watching the GPUs, none of them are 'maxed' out and sit at about 85% utilization and MCU is also at about 80% utilization. This may be a intensity issue, but I kind of doubt it considering what Pallas was saying. Intensity seems about ballpark with a little bit of testing as well.

Well I installed new driver, then at some point it said I was on .38 driver again after a few restarts. Reinstalled .69 and 1070 is hashing at 600~(w10), I guess custom clean install is needed instead of express.

so you fixed it now? they should do the same on win 10 like they do on win7/8

also on mcu utilization being only 80%, fb usage which is the sam eis at 80% on maxwell too, so that can not be improved

No I did a clean install and its still hashing 600~, not using insider though if that matters for neoscrypt?
Other algos are hashing fine though besides eth

[Amph]

Look like the 1070 mem perf was fixed in driver 368.69, i now get 950 kH/s on windows (+30%)

but the 1080 seems worst sigh

GPU #0: Gigabyte GTX 1080, 391.72 kH/s
accepted: 15/15 (diff 0.003), 1324.49 kH/s yes!
GPU #1: GeForce GTX 1070, 949.84 kH/s
accepted: 16/16 (diff 0.004), 1325.44 kH/s yes!

Already installed the driver, haven't checked Ethereum and Neo hash as other 'users' reported it didn't fix it. Shame on me for listening to them I guess.

NeoS gets around 1000~ Mh/s on W8.1 x64. Using DJMs last build.

Definitely still room for tuning. Should be closer to 1200. Watching the GPUs, none of them are 'maxed' out and sit at about 85% utilization and MCU is also at about 80% utilization. This may be a intensity issue, but I kind of doubt it considering what Pallas was saying. Intensity seems about ballpark with a little bit of testing as well.

Well I installed new driver, then at some point it said I was on .38 driver again after a few restarts. Reinstalled .69 and 1070 is hashing at 600~(w10), I guess custom clean install is needed instead of express.

so you fixed it now? they should do the same on win 10 like they do on win7/8

also on mcu utilization being only 80%, fb usage which is the sam eis at 80% on maxwell too, so that can not be improved

No I did a clean install and its still hashing 600~, not using insider though if that matters for neoscrypt?
Other algos are hashing fine though besides eth

i think insider build does not matter anymore, they reverted the wddm to 2.0(check your dxdiag now to see what you have)