Author

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 114. (Read 2347601 times)

newbie
Activity: 126
Merit: 0
nice project keep it up
hero member
Activity: 773
Merit: 508
Bitcore (BTX) - The Future is Now
Memory is the same across all GPUs except for some weirdo models, so +400 is the same for everyone.
The problem is there's probably 50+% of nvidia users out there that mine in P2 mode and just don't know how to force the miner to P0. So +400 will only be something like "+200 from stock" for a lot of folks. Smiley


How do you force 10 series cards into p0 mode?

I believe nvidia inspector has this option.
newbie
Activity: 42
Merit: 0
Memory is the same across all GPUs except for some weirdo models, so +400 is the same for everyone.
The problem is there's probably 50+% of nvidia users out there that mine in P2 mode and just don't know how to force the miner to P0. So +400 will only be something like "+200 from stock" for a lot of folks. Smiley


How do you force 10 series cards into p0 mode?

https://github.com/nicehash/NiceHashMiner/issues/1000

Here you wil find the application. Run it. In profile "_GLOBAL_DRIVER_PROFILE (Base Profile)", goto section "5 - Common", Change "CUDA - Force P2 state" from ON (0x00000001) to OFF (0x00000000).
full member
Activity: 728
Merit: 106
I can share the link free for you, SP. remember my kindness please!
check the latest Klaus ccminer, I'm sure you know where to find it
 Roll Eyes

and on gtx 1080ti are doing about 1670 kh/s. can you make really better then that?
Now he has the base to improve a little bit with his compilation tricks, if they work  Grin
full member
Activity: 349
Merit: 102
Memory is the same across all GPUs except for some weirdo models, so +400 is the same for everyone.
The problem is there's probably 50+% of nvidia users out there that mine in P2 mode and just don't know how to force the miner to P0. So +400 will only be something like "+200 from stock" for a lot of folks. Smiley


How do you force 10 series cards into p0 mode?
newbie
Activity: 37
Merit: 0
Whos got lyra2z pm me please for exchane, teper u menya vse est,bitcore,xcn,cryptonight,keccak,phi and all i need is lyra2z and i will give all the miners i have for lyra2z.
Thanks
hero member
Activity: 756
Merit: 507
you are saying that current  open source miner on   1080ti   does about 550?
I guess with claims  like that its not so hard to sell mods

I am not selling anything I just tried the 1070ti on different exe files and they all do around 500, but perhaps you can point me in the right direction here. Where can I find the miner that does 1700? I am using cuda 9.0 and the latest driver.

here you are! you can see two gtx1070ti are working on 1360kh/s



I can share the link free for you, SP. remember my kindness please!
check the latest Klaus ccminer, I'm sure you know where to find it
 Roll Eyes

and on gtx 1080ti are doing about 1670 kh/s. can you make really better then that?
legendary
Activity: 1106
Merit: 1014
Without a load on the GPU it shouldn't cause a crash. If you're getting crashes from +200 without a load you're already unstable.
I've no idea what exactly happens when that +200 jump occurs, maybe there is some kind of load in that moment. Maybe some calculations are still being done when the miner is closing (if there was no load at that moment, why would it go to p0 after all and not to some low-power mode?). I've had very stable rigs hashing for 24+ hours but crashing when the miner is closed (and then, after forcing cuda apps to run in p0 on those systems, they're stable at those clocks for days/weeks). I've seen quite a few similar reports from other users, afaik it's a relatively common issue.
full member
Activity: 728
Merit: 106
The problem is there's probably 50+% of nvidia users out there that mine in P2 mode and just don't know how to force the miner to P0. So +400 will only be something like "+200 from stock" for a lot of folks. Smiley

That's basically a OC. Also not sure why you'd go through the trouble of forcing p0 when you could just OC the memory right out of the gate. Less messing around. Maybe makes your OCD feel better? It doesn't seem to increase voltage to the memory and it doesn't allow you to OC the memory further.
Of cause you don't know. One of the reasons you are doing 2.5$ with 1070  Grin
legendary
Activity: 1764
Merit: 1024
Also not sure why you'd go through the trouble of forcing p0 when you could just OC the memory right out of the gate. Less messing around. Maybe makes your OCD feel better? It doesn't seem to increase voltage to the memory and it doesn't allow you to OC the memory further.
Because there's a bug (or is that a feature?) in nvidia drivers or mining software that makes the memory clock jump ~ 200 MHz up when you close the miner's window. I didn't spend too much time investigating it, so maybe it's not all driver versions and/or not all miners that do that, but every single one I tried behaved that way. Let's say you've got gtx 1080 and your stock memory clock is supposed to be 5000. When you set OC to +400 (in MSI AB, for example), and run the miner, your actual memclock will be ~ 5200 or smth like that, and not 5400 (because cuda apps run in p2 instead of p0). Which would be ok, you could just add +600 instead to reach the clock you wanted (5400)... except when you close the miner, the memclock will jump to 5600 for a brief moment. And that is a problem - if you're pushing you memory overclock close to its max, then you'll get the driver crash in that short moment when the clock jumps to +200 more.

If you force cuda apps to run in p0 mode, then that jump is gone, the memory clock stays at whatever you set it to even when you close the miner window. And yeah, it does allow you to OC memory further, because without that weird jump you don't have to leave so much headroom for that brief clock increase and can set higher memclock.

Without a load on the GPU it shouldn't cause a crash. If you're getting crashes from +200 without a load you're already unstable. For instance you can do +800 on the desktop for pretty much any GPU and it wont crash. I would say +1000, but that's very close to absolute maximum and may crash just by sneezing at it.

+500 for instance is a pretty universally accepted memory OC across all my gpus when mining Ethereum (I haven't OC'd memory while mining Cryptonote or NeoS yet, so that might change). I was running them at p2. They don't crash even though they went +700 upon exiting the miner. However. If I ran +700, almost none of them could run the miner at that, not just when closing the miner.

Maybe you're running super bleeding edge, so I guess, but for all intents and purposes most people aren't going to run into that.
legendary
Activity: 1106
Merit: 1014
Also not sure why you'd go through the trouble of forcing p0 when you could just OC the memory right out of the gate. Less messing around. Maybe makes your OCD feel better? It doesn't seem to increase voltage to the memory and it doesn't allow you to OC the memory further.
Because there's a bug (or is that a feature?) in nvidia drivers or mining software that makes the memory clock jump ~ 200 MHz up when you close the miner's window. I didn't spend too much time investigating it, so maybe it's not all driver versions and/or not all miners that do that, but every single one I tried behaved that way. Let's say you've got gtx 1080 and your stock memory clock is supposed to be 5000. When you set OC to +400 (in MSI AB, for example), and run the miner, your actual memclock will be ~ 5200 or smth like that, and not 5400 (because cuda apps run in p2 instead of p0). Which would be ok, you could just add +600 instead to reach the clock you wanted (5400)... except when you close the miner, the memclock will jump to 5600 for a brief moment. And that is a problem - if you're pushing you memory overclock close to its max, then you'll get the driver crash in that short moment when the clock jumps to +200 more.

If you force cuda apps to run in p0 mode, then that jump is gone, the memory clock stays at whatever you set it to even when you close the miner window. And yeah, it does allow you to OC memory further, because without that weird jump you don't have to leave so much headroom for that brief clock increase and can set higher memclock.
legendary
Activity: 1764
Merit: 1024
The problem is there's probably 50+% of nvidia users out there that mine in P2 mode and just don't know how to force the miner to P0. So +400 will only be something like "+200 from stock" for a lot of folks. Smiley

That's basically a OC. Also not sure why you'd go through the trouble of forcing p0 when you could just OC the memory right out of the gate. Less messing around. Maybe makes your OCD feel better? It doesn't seem to increase voltage to the memory and it doesn't allow you to OC the memory further.
legendary
Activity: 1106
Merit: 1014
Memory is the same across all GPUs except for some weirdo models, so +400 is the same for everyone.
The problem is there's probably 50+% of nvidia users out there that mine in P2 mode and just don't know how to force the miner to P0. So +400 will only be something like "+200 from stock" for a lot of folks. Smiley

legendary
Activity: 1764
Merit: 1024
GTX 1070 gives 1350 kh/s with free alexis miner. What is the purpose of that picture with slow-hashing thing?

That would be incorrect (maybe with hardcore OCing). Also Alexis isn't the fastest branch for NeoS.


You guys REALLLLLY need to list the clocks you're operating at. +1000 for memory is unrealistic for like 99.9999% of the GPUs. Memory is the same across all GPUs except for some weirdo models, so +400 is the same for everyone. Bench's should be done at stock clocks. OC's increase hashrate for every miner ever made, no one cares about your super unstable clocks you mine with for 15 minutes before your system hardlocks and you don't notice for two hours.

That's why I've mentioned a few different times SP needs to list fastest available pub miner next to his, so you can weed out a clocks difference. Apparently you guys are being dumbasses and listing OC results against stock clock results.

...but i presume it should also yield good result since neoscrypt isn't very memory dependent.

NeoS is EXTREMELY memory dependent. You'll get a little less hashrate with a 1080 then a 1070, a 1080ti will be about 20% faster. As I mentioned multiple times NeoS is one of the algos that GDDR5X does not operate well on.

full member
Activity: 224
Merit: 100
CryptoLearner
you are saying that current  open source miner on   1080ti   does about 550?
I guess with claims  like that its not so hard to sell mods

I am not selling anything I just tried the 1070ti on different exe files and they all do around 500, but perhaps you can point me in the right direction here. Where can I find the miner that does 1700? I am using cuda 9.0 and the latest driver.

Have you tried KlausT ccminer ? I do about 1100-1150 on 1070 with 60% tdp (112W) and good OC and about 1400/1450 on 1070Ti with about the same TDP/OC, don't have any 1080+ though, but i presume it should also yield good result since neoscrypt isn't very memory dependent.
full member
Activity: 308
Merit: 100
you are saying that current  open source miner on   1080ti   does about 550?
I guess with claims  like that its not so hard to sell mods

I am not selling anything I just tried the 1070ti on different exe files and they all do around 500, but perhaps you can point me in the right direction here. Where can I find the miner that does 1700? I am using cuda 9.0 and the latest driver.

" but perhaps you can point me in the right direction here. "

Donate 0.01 btc I send you directions to a free miner tha does 1300-1550 on 1080ti


"Where can I find the miner that does 1700?"   I said 1070ti "SHOULD" go there   I did not said there is a miner available yet
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
you are saying that current  open source miner on   1080ti   does about 550?
I guess with claims  like that its not so hard to sell mods

I am not selling anything I just tried the 1070ti on different exe files and they all do around 500, but perhaps you can point me in the right direction here. Where can I find the miner that does 1700? I am using cuda 9.0 and the latest driver.
full member
Activity: 308
Merit: 100
the Price must reflect the result  of a miner

Is someone willing to pay for a neoscrypt that does 1700 on the 1080ti and 1300 on the 1070ti? Thats about 300% faster than the legacy code from alexis.

you are saying that current  open source miner on   1080ti   does about 550?



I guess with claims  like that its not so hard to sell mods


ps. Rednow is right the 1070ti should go to 1500-1700 
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
the Price must reflect the result  of a miner

Is someone willing to pay for a neoscrypt that does 1700 on the 1080ti and 1300 on the 1070ti? Thats about 300% faster than the legacy code from alexis.
full member
Activity: 308
Merit: 100
The alexis miner is doing 450 KHASH on the 1070ti, so you need a dev to change the code to do the expected 1100-1400 Khash. I am reaching 1300+ on the gtx 1070 with my cuda 9.0 mod and only 550 on the 1070ti.
$750 dollars is nothing. Devs need money too.

the Price must reflect the result  of a miner

I remember you wanted to sell Polytimos miner for 0.2btc?    yeah  TOO Much for nothing
Jump to: