Author

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 579. (Read 2347641 times)

legendary
Activity: 1512
Merit: 1000
quarkchain.io
Playing a bit with the powerLimit on a 1070s SJS rig , giving me between 305-308MH...
For one card I was able to get 312mhs@154watt stable.

Yeah , I saw , I was talking about 6cards rig.
legendary
Activity: 1510
Merit: 1003
Playing a bit with the powerLimit on a 1070s SJS rig , giving me between 305-308MH...
For one card I was able to get 312mhs@154watt stable.
legendary
Activity: 1512
Merit: 1000
quarkchain.io
Playing a bit with the powerLimit on a 1070s SJS rig , giving me between 305-308MH...
legendary
Activity: 1510
Merit: 1003
You didn't get it from me. So I guess someone else have sold it to you cheaper.
Sorry, but for 1 card owner 0.1btc is nonsence, so I did.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
You all remember - I said many bad words about _sp and his greediness. I can repeat this. But after trying his mod I must say - it is good. He eliminated the mem usage, reduced power consumption and his code is more stable on high clocks (or medium clocks with very low voltage). This can be used for maximum hashrate or for maximum power saving with decent speed. High price + hi % devmining is bad. But I got this with great discount so I will use it.

You didn't get it from me. So I guess someone else have sold it to you cheaper.
legendary
Activity: 1510
Merit: 1003
You all remember - I said many bad words about _sp and his greediness. I can repeat this. But after trying his mod I must say - it is good. He eliminated the mem usage, reduced power consumption and his code is more stable on high clocks (or medium clocks with very low voltage). This can be used for maximum hashrate or for maximum power saving with decent speed. High price + hi % devmining is bad. But I got this with great discount so I will use it.

BTW. My gtx750 1gb running slower with this sp_private3 comparing to Epsylon opensource. 42mhs against 48. Maybe it is bad tuned for little maxwell.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
Not after the penta miner.
Never from me again.

The penta miner worked as advertised with the correct boost in the hashrate. Even today the pentablake sp-mod private is +25% faster than the opensource on the gtx 1070. Are you blaming me for the collapse of Joincoin? Penta has been more profitable to mine than LBR for a long time if you only put a couple of cards to mine it.
legendary
Activity: 1400
Merit: 1000
At exact same setting my version is around

5-7% faster on the gtx 1070 and 1080.( +10% with oc and tuning).
maxwell compute 5.2 l cards are 10-25% faster
compute 5.0  750ti cards are 5-7% faster

OK then send me your kernel and I will down clock your kernel to the open source to compare the speed. That way it is speed for speed.

I can send you a copy if you donate 0.1BTC

Not after the penta miner.

Never from me again.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
Take a look again. With the rig running 2 GPU's.(the 2nd picture)  Gpu #1 is hashing @ 300mhash with a coreclock of 2012mhz.
With 6 cards the result is reduced a bit, but this has nothing to do with the kernel.
Perhaps the PSU is not powerful enough to provide the extra power needed.
Perhaps the rig is without powered risers.

legendary
Activity: 3248
Merit: 1070
i have only 1x8 pin card, so this can't be good for me, since i'm already at 276MH or 283, need to check it again, again 290, not so much of an improvement

You should reach 300+ on all your cards with the correct setup. Please read Bensams post.
Rednow  has  (as usual) the overclock record with 322MHASH with the sp-mod #3
He managed to run the opensource @ 290MHASH on his (crazy clocks).

+11%

bensam didn't say anything worth reading, all i know is that your screen show 290 for 2025 core and 93% tdp

if i put those exact setting with the open source i get 276, against your 290, there is no 300 here

and frankly i don't want to go above 80% tdp as the gpu tend to consume a lot, and i have only 1000w feeding six gpu

so if you cna do a test at 80% tdp and showing the core i can know at 100% if this si worth it
legendary
Activity: 1512
Merit: 1000
quarkchain.io
i have only 1x8 pin card, so this can't be good for me, since i'm already at 276MH or 283, need to check it again, again 290, not so much of an improvement

You should reach 300+ on all your cards with the correct setup. Please read Bensams post.
Rednow  has  (as usual) the overclock record with 322MHASH with the sp-mod #3
He managed to run the opensource @ 290MHASH on his (crazy clocks).

+11%

..Yes indeed , if you have single card on the rig. Things are different if you fill the PCI-E line with 6...
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
i have only 1x8 pin card, so this can't be good for me, since i'm already at 276MH or 283, need to check it again, again 290, not so much of an improvement

You should reach 300+ on all your cards with the correct setup. Please read Bensams post.
Rednow  has  (as usual) the overclock record with 322MHASH with the sp-mod #3
He managed to run the opensource @ 290MHASH on his (crazy clocks).

+11%
legendary
Activity: 3248
Merit: 1070
2 cards 300 mhash
6 cards 294mhash

But this can be solved. Red antantti last posts. The 1x 8pin cable is not able to deliver enough power, and with 6 cards oc drawing power from the mainboard you need to lower the tdp of all the cards.





i have only 1x8 pin card, so this can't be good for me, since i'm already at 276MH or 283, need to check it again, against 290, not so much of an improvement
legendary
Activity: 1764
Merit: 1024
You guys that are complaining about the 70% TDP are being dumb. A couple pages ago it was asked to compare it at lower TDP so we could get a comparison for same wattage. Those two cards are operating at the same TDP. That TDP is dynamic, meaning the amount of clockspeed you'll get on one card is higher then the other because SP's MINER STRESSES CARDS LESS!

That means it's more efficient. Which is why he's saying you can OC the stuffing out of the cards. MINERS DO NOT ALL OPERATE AT THE SAME WATTAGE! Even for the same algo, Epsylon can code one, Wolf can code one, and SP can code one, and you'll get different watt readings on all of those... IT'S ENTIRELY POSSIBLE to have a more efficient miner, meaning you'll get a higher hashrate for LESS watts. Since Nvidia uses dynamic clock speeds, when you limit by same wattage, which Anttanti did, you end up at different clock speeds and hashrates, because SP can PUT OUT MORE HASH at the SAME wattage.

If you notice in the second test, ANT limited instead by CLOCK SPEED and as such, you receive MORE HASH at LESS WATTAGE. Look at the TDP. The clocks are the same, the wattage however is not (TDP).

Jesus. There are no schenanagins happening here. What SP is saying is what you're getting, it's been confirmed multiple times by multiple posts.

You don't need to see results on a six card rig. That's dumb and stupid logic. Nothing will change. Matter of a fact you might even end up with even better hashrate depending on how well you have your power delivery setup (more cards using less watts means better power delivery).


In other news Lbry is poping again. It's still more profitable to mine ETH+SIA then Lbry on Epsylons, Lbry is more profitable using SPs miner. Taking into account fluctuations in the market, it's probably still a good idea to stay on ETH+SIA, unless you think it'll continue to rise.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
ROUND 2

1.8.1 vs sp#3 70% 120% tdp, strix 970 vs strix 970, intensity 25, windows 7x64 368.81

Cards are sitting next to each other in that very same rig hashing together. Left with opensource right with sp #3. Same card, same memory, same bios, same asic quality, same default voltage. 1x8-pin power connector.

Intensity is 25 because that is max what opensource can do, core is 1486 because that is max what opensource can keep steady. For this test tdp is set to 120% which means 196 watts which is max what these cards can do without bios mod.

Let's get ready to rumble!



-edit- well it looks like left wasn't steady at 1486 but you get the idea.

Thanks antantti. With -i 27 and higher core clock you should reach 190+. Don't know about the strix, but the gigabyte G1 gaming cards can go all the way up to 1650mhz with this kernel. (but they come with 4pin + 3pin power) and 250W tdp

legendary
Activity: 1764
Merit: 1024

Hey look you regurgitated the part about intensity I posted a couple days ago... Good job bro and you have no idea what intensity actually does. Higher intensity does not mean you'll have better hashrate. Higher intensity can actually produce worse hashrate and use more power.

It's not a magic value too, where you increase intesnity by 5 and it's magically that much better. Your lack of understanding of... well... anything is astounding and you try to use it as 'proof' something is working wrong. Reminds me of the guy complaining that Wolf0 broke his rigs because he purchased a miner that utilized his GPUs better and produced more heat while drawing more wattage... then pretended like his cards would always use the same wattage.

If you don't know what you're talking about, don't act like you do.


You can't do math. What Yiimp lists as actual is what it paid for actual MH/BTC in the last day.

this guy is funny, keep posting retarded sentence, like regurgitating, no what i mean about intensity is that he changed how intensity work, by assigning it other value, but as usual you didn't understood shit, and i was hoping for a code rewrite instead....

and yes you can do the math about any coin, only because you don't know how to do it, it does not mean that it can not be done, please stay away with your ignorance is infecting me

Assigned it a different value like 29 instead of 25? As I mentioned before, that number often times has little association with 'improvements'. Just because intensity is higher doesn't mean it's a better miner or vice versa.

He DID rewrite code, you don't magically get better performance by increasing intensity. Try it with Epsylons miner.

Of course I can do math, as can other people... You apparently can't and are saying yiimps actual pool estimate are wrong. We've already discusses Suprnova and Coinmine. Coinmine has been improved, but it's often times still wrong.

lol, my ignorance is infecting you? Don't be a dumbass that tries to pretend he knows what he's talking about when you have absolutely no idea and we wont have these conversations.
legendary
Activity: 1176
Merit: 1015
ROUND 2

1.8.1 vs sp#3 70% 120% tdp, strix 970 vs strix 970, intensity 25, windows 7x64 368.81

Cards are sitting next to each other in that very same rig hashing together. Left with opensource right with sp #3. Same card, same memory, same bios, same asic quality, same default voltage. 1x8-pin power connector.

Intensity is 25 because that is max what opensource can do, core is 1486 because that is max what opensource can keep steady. For this test tdp is set to 120% which means 196 watts which is max what these cards can do without bios mod.

Let's get ready to rumble!



-edit- well it looks like left wasn't steady at 1486 but you get the idea.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
LBRY credits +20% today on a 281.58 BTC Volume.

http://coinmarketcap.com/currencies/library-credit/

Go get those coins!
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
1080 g1 2050 core -i 27 405 mh, I think this is stable. 400 mh is stable though with -i 26. 1080 doesnt scale well with overclock. windows 8.1.  -i 28 has nonce problems

I'd love to see a screenshot of the 1080 above 400. Smiley did you remember to reduce the memclock? Use NVIDIA inspector -1000 mhz. Then you have more power to overclock the core..
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
At exact same setting my version is around

5-7% faster on the gtx 1070 and 1080.( +10% with oc and tuning).
maxwell compute 5.2 l cards are 10-25% faster
compute 5.0  750ti cards are 5-7% faster

OK then send me your kernel and I will down clock your kernel to the open source to compare the speed. That way it is speed for speed.

I can send you a copy if you donate 0.1BTC
Jump to: