CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1197.

Bombadil

hero member

Activity: 644

Merit: 500

Quote from: scryptr on December 30, 2014, 08:56:10 AM

Quote from: K1773R on December 29, 2014, 04:01:29 PM

Quote from: Bombadil on December 29, 2014, 04:00:55 PM

Quote from: K1773R on December 29, 2014, 03:34:55 PM

does it work with mono?

No idea, and I won't get my hopes up for it ^^" But it works over the network, so another windoze pc or virtualbox with windoze should work.

winblows is no option Wink

KBOMBA-CCMONITOR-WINBLOWS-MONO-

Ok, before I even try to read the documentation, would I be able to run ccmonitor on a Win 7-64 computer and monitor a linux rig remotely? Or would I have to run ccmonitor with Mono on the Linux rig, and view it on the Win 7 computer?

Mono appears to be a big project. I'll be looking at ccmonitor today. --scryptr

Lol, no, normally it should work on your windoze and monitor the linux one from there. I haven't tested it, so please tell me if something goes wrong ^^"
You can also find me on the IRC Wink

scryptr

legendary

Activity: 1797

Merit: 1028

Quote from: K1773R on December 29, 2014, 04:01:29 PM

Quote from: Bombadil on December 29, 2014, 04:00:55 PM

Quote from: K1773R on December 29, 2014, 03:34:55 PM

does it work with mono?

No idea, and I won't get my hopes up for it ^^" But it works over the network, so another windoze pc or virtualbox with windoze should work.

winblows is no option Wink

KBOMBA-CCMONITOR-WINBLOWS-MONO-

Ok, before I even try to read the documentation, would I be able to run ccmonitor on a Win 7-64 computer and monitor a linux rig remotely? Or would I have to run ccmonitor with Mono on the Linux rig, and view it on the Win 7 computer?

Mono appears to be a big project. I'll be looking at ccmonitor today. --scryptr

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Looks like a memory leak somewhere, but seems to work fine on Windows. Klaus_t has removed a leak of 4 bytes per throughput in his latest checkin on github. Might help.

scryptr

legendary

Activity: 1797

Merit: 1028

Quote from: tbearhere on December 30, 2014, 07:19:43 AM

Quote from: scryptr on December 30, 2014, 06:10:30 AM

Quote from: sp_ on December 30, 2014, 03:12:45 AM

Quote from: djm34 on December 29, 2014, 07:05:04 PM

Quote from: scryptr on December 29, 2014, 06:46:31 PM

DJM34 --
Please explain what you mean by scaling? Should it reach the same ratio between card models as scrypt?

I was expecting something different... but actually it looks like sp_ x11 in terms of scaling

The problem with the 750ti cards is that they use so little power. When you optimize the code, the card sometimes downclock itself automaticly and the hashrate will drop.
This is why I have reduced the number of threads in many of the hashing functions. In BMW f.ex I only use 32 threads. On 970/980 on the other hand, more threads increase the performance.

Hmmmmm... that may explain why my slower rig slows down to a crawl and then dies. I found it barely hashing after restarting it at good speed 5 hours ago. Should I not use the intensity flag? --scryptr

My 750ti rig is still going strong since yesterday .......NO intensity setting.
EDIT: windows 8 and #23 release.

I am still not exactly clear on what is meant by the term "scaling". Is it refering to operations within the code execution, or with overall hash-rate performance in various GPU models/makes?

Whatever. I have the same commandline for both my 750ti rigs. My intensity flag is set at 20.5, the rest is address/user. However, the faster rig (750ti/AMD FX cpu/8 gig) rocks steady with +/- 10kh/s rate of 18,425kh/s. The slower rig (750ti/AMD Sempron 145/8 gig) starts as previously posted at 17,965kh/s and then slows to far less, maybe 1000kh/s. It was pretty steady on 1.5.2-git, and before I started tweaking intensity. Both have identical motherboards.

Maybe I should reload the OS? Or maybe I should relax the intensity to less than default? --scryptr

tbearhere

legendary

Activity: 3164

Merit: 1003

Quote from: scryptr on December 30, 2014, 06:10:30 AM

Quote from: sp_ on December 30, 2014, 03:12:45 AM

Quote from: djm34 on December 29, 2014, 07:05:04 PM

Quote from: scryptr on December 29, 2014, 06:46:31 PM

DJM34 --
Please explain what you mean by scaling? Should it reach the same ratio between card models as scrypt?

I was expecting something different... but actually it looks like sp_ x11 in terms of scaling

The problem with the 750ti cards is that they use so little power. When you optimize the code, the card sometimes downclock itself automaticly and the hashrate will drop.
This is why I have reduced the number of threads in many of the hashing functions. In BMW f.ex I only use 32 threads. On 970/980 on the other hand, more threads increase the performance.

Hmmmmm... that may explain why my slower rig slows down to a crawl and then dies. I found it barely hashing after restarting it at good speed 5 hours ago. Should I not use the intensity flag? --scryptr

My 750ti rig is still going strong since yesterday .......NO intensity setting.
EDIT: windows 8 and #23 release.

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: djm34 on December 30, 2014, 07:04:25 AM

Quote from: sp_ on December 30, 2014, 06:52:17 AM

Quote from: djm34 on December 30, 2014, 06:26:53 AM

The miner will be publicly released once 5btc has been reached.
It looks like a lot, but it isn't much if everyone participate. Now if only a bunch of people allows to reach that amount or if only a few donates, then the miner will be released privately only to those who pledge.
I think it is fair this way, it is a crowd funding project, which means everyone has to participate and give a tiny bit, it may also protect coin value...
(I'll make a new thread when I was a minute)

But with only 100KHASH on the 750ti this miner is not profitable. If you want donations, you want to make sure that the miners get's a profit.

That includes support and update... and it is already more profitable than many things around... (also tdp usage is only 80%)

Looks like the best opensource miner is doing 50KHASH with overclock on the 750ti (opencl code) (Neo scrypt)

http://cryptomining-blog.com/3682-mining-feathercoin-after-the-neoscrypt-fork-with-amd-and-nvidia-gpus/

and from the user

scryptr: 3.7.8 neoscrypt on a 550ti at a rate of 25kh/s (Win64).

So your kernal is more than 100% faster than the opensource. Pretty good.

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: maxim000 on December 30, 2014, 12:19:34 AM

good day
23-th version automaticly change -f multiplier to 0.67 even thouse it is set in 1.0 in bat file - it`s normal ?
so my hash rate in pool significantly lose
also one of two my farms on 750Ti crashe after half hour working

I think I have a fix. I will remove the code that reduce the diff in the next build. (build 24)

Difficulty is reduced in the miner, but not on the pool. This meens that the pool think you are mining on higher diff, but you only submitt low diff shares.

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: sp_ on December 30, 2014, 06:52:17 AM

Quote from: djm34 on December 30, 2014, 06:26:53 AM

The miner will be publicly released once 5btc has been reached.

It looks like a lot, but it isn't much if everyone participate. Now if only a bunch of people allows to reach that amount or if only a few donates, then the miner will be released privately only to those who pledge.
I think it is fair this way, it is a crowd funding project, which means everyone has to participate and give a tiny bit, it may also protect coin value...
(I'll make a new thread when I was a minute)

But with only 100KHASH on the 750ti this miner is not profitable. If you want donations, you want to make sure that the miners get's a profit.

That includes support and update... and it is already more profitable than many things around... (also tdp usage is only 80%)

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: djm34 on December 30, 2014, 06:26:53 AM

The miner will be publicly released once 5btc has been reached.

It looks like a lot, but it isn't much if everyone participate. Now if only a bunch of people allows to reach that amount or if only a few donates, then the miner will be released privately only to those who pledge.
I think it is fair this way, it is a crowd funding project, which means everyone has to participate and give a tiny bit, it may also protect coin value...
(I'll make a new thread when I was a minute)

But with only 100KHASH on the 750ti this miner is not profitable. If you want donations, you want to make sure that the miners get's a profit.

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: scryptr on December 30, 2014, 06:10:30 AM

Hmmmmm... that may explain why my slower rig slows down to a crawl and then dies. I found it barely hashing after restarting it at good speed 5 hours ago. Should I not use the intensity flag? --scryptr

No this must be something else. My miners run without the intensity flag. The head on github is untested on linux, so there might be some issues.

Amph

legendary

Activity: 3248

Merit: 1070

Quote from: ?? on ??

Quote from: djm34 on December 29, 2014, 05:57:01 PM

Quote from: djm34 on December 28, 2014, 12:39:01 PM

Quote from: djm34 on December 28, 2014, 11:20:45 AM

some progress on neoscrypt:
750ti: ~~81kh/s~~ ~~110kh/s~~ 107kh/s
780ti: ~~165kh/s~~ ~~182kh/s~~ 193kh/s
980: ~~290kh/s~~ ~~299kh/s~~ 335kh/s
(still doesn't scale right Huh

)

Let me know when the 980 hits 600kh/s. Grin

you beat amd with your own miner on neoscrypt?

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: ?? on ??

Quote from: djm34 on December 29, 2014, 05:57:01 PM

Quote from: djm34 on December 28, 2014, 12:39:01 PM

Quote from: djm34 on December 28, 2014, 11:20:45 AM

some progress on neoscrypt:
750ti: ~~81kh/s~~ ~~110kh/s~~ 107kh/s
780ti: ~~165kh/s~~ ~~182kh/s~~ 193kh/s
980: ~~290kh/s~~ ~~299kh/s~~ 335kh/s
(still doesn't scale right Huh

)

Let me know when the 980 hits 600kh/s. Grin

I haven't finished yet to clean the code... Is this what you get with your private kernel ?

Ok here what I am thinking regarding the public release of the nvidia miner for neoscrypt:
Pledge for the bounty will have to be sent to that BTC address: 18UZRrybgjmR4M53Aca4Q8YBguqErZ5qaP

The miner will be publicly released once 5btc has been reached.

It looks like a lot, but it isn't much if everyone participate. Now if only a bunch of people allows to reach that amount or if only a few donates, then the miner will be released privately only to those who pledge.
I think it is fair this way, it is a crowd funding project, which means everyone has to participate and give a tiny bit, it may also protect coin value...
(I'll make a new thread when I was a minute)

scryptr

legendary

Activity: 1797

Merit: 1028

Quote from: sp_ on December 30, 2014, 03:12:45 AM

Quote from: djm34 on December 29, 2014, 07:05:04 PM

Quote from: scryptr on December 29, 2014, 06:46:31 PM

DJM34 --
Please explain what you mean by scaling? Should it reach the same ratio between card models as scrypt?

I was expecting something different... but actually it looks like sp_ x11 in terms of scaling

The problem with the 750ti cards is that they use so little power. When you optimize the code, the card sometimes downclock itself automaticly and the hashrate will drop.
This is why I have reduced the number of threads in many of the hashing functions. In BMW f.ex I only use 32 threads. On 970/980 on the other hand, more threads increase the performance.

Hmmmmm... that may explain why my slower rig slows down to a crawl and then dies. I found it barely hashing after restarting it at good speed 5 hours ago. Should I not use the intensity flag? --scryptr

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: djm34 on December 29, 2014, 07:05:04 PM

Quote from: scryptr on December 29, 2014, 06:46:31 PM

DJM34 --
Please explain what you mean by scaling? Should it reach the same ratio between card models as scrypt?

I was expecting something different... but actually it looks like sp_ x11 in terms of scaling

The problem with the 750ti cards is that they use so little power. When you optimize the code, the card sometimes downclock itself automaticly and the hashrate will drop.
This is why I have reduced the number of threads in many of the hashing functions. In BMW f.ex I only use 32 threads. On 970/980 on the other hand, more threads increase the performance.

maxim000

full member

Activity: 304

Merit: 100

Quote from: sp_ on December 30, 2014, 02:10:28 AM

The crashes are known, and can be fixed by a failover script like this one:

https://github.com/KBomba/failover-ccminer-bat

thanks !

Quote from: sp_ on December 30, 2014, 02:10:28 AM

Thanks for finding this, this might be the bug we are looking for. Some pools report a drop in performance.

it happens often when pool stratum difficulty increase too mach and too quickly, for examle from 0.03 to 4 for one step.
I`m saw it frequently on stratum+tcp://src.coinmine.pl:6020

i think good idea to fix this value if it strict appear in bat file

upd. also this bag appear in previous versions, compiled by http://cryptomining-blog.com/ , for example from http://cryptomining-blog.com/4085-updated-windows-binary-of-the-ccminer-1-5-2-git-fork-by-sp-for-maxwell/

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: maxim000 on December 30, 2014, 12:19:34 AM

good day
23-th version automaticly change -f multiplier to 0.67 even thouse it is set in 1.0 in bat file - it`s normal ?
so my hash rate in pool significantly lose
also one of two my farms on 750Ti crashe after half hour working

Thanks for finding this, this might be the bug we are looking for. Some pools report a drop in performance.

The crashes are known, and can be fixed by a failover script like this one:

https://github.com/KBomba/failover-ccminer-bat

maxim000

full member

Activity: 304

Merit: 100

good day

23-th version automaticly change -f multiplier to 0.67 even thouse it is set in 1.0 in bat file - it`s normal ?
so my hash rate in pool significantly lose

also one of two my farms on 750Ti crashe after half hour working

scryptr

legendary

Activity: 1797

Merit: 1028

SP_ Build 23 --

I compiled and ran sp_ build 23 today, starting a run at about 1615hrs. At 2115hrs I noticed it had crashed on my slower rig. However, my faster rig is running build 23 steadily at the same improved hashing rate as yesterday with build 22, and as earlier today with build 23.

I have since restarted build 23 on my slower rig. It is running ok, but both rigs are headless. I guess I need to start looking at ccmonitor by KBomba. --scryptr

scryptr

legendary

Activity: 1797

Merit: 1028

Quote from: djm34 on December 29, 2014, 10:36:11 PM

Quote from: chrysophylax on December 29, 2014, 08:59:48 PM

Quote from: djm34 on December 29, 2014, 08:55:22 PM

Quote from: scryptr on December 29, 2014, 08:47:11 PM

Quote from: djm34 on December 29, 2014, 08:26:58 PM

Quote from: scryptr on December 29, 2014, 07:11:10 PM

DJM34 --

Perhaps the original author of neoscrypt code meant for it to be difficult to port to nVidia.

? still not sure why you are saying that...
and actually doesn't seem to be the case, actually the 980 already beats the 290x/public kernel... (which is around 300kh/s)

Wolf0 answered for you. I thought that the code for neoscrypt may have been obfuscated deliberately in order to hinder its porting to Cuda code, perhaps in response to CBuchner's release of difficult-to-port "Killer Groestl". I'm wondering why development for nVidia is behind, but I am not slighting your efforts.

because there is no bounty Grin

(and paid project or project which are profitable have the priority... )

in that case - what would it take to 'work' on this project to make it worth its weight in gold?

i would be happily pointing my miners to 'donate' for such a project ...

or is there something that could be done in private that would benefit the contributors 'first' before a public release of such optimized code?

#crysx

I am wondering actually... how I will proceed.
1st solution one bounty address for public release, when the amount requested is reached the miner is released for everybody
2st a bounty address for private release once the amount is reached those who pledge for the private miner, gets the miner.
well it is always possible to contact me and propose something...

My two-bits worth --

Public miner! I would contribute, by hamster-wheel mining, or by dipping into my wallet. There are hot-rod coders, and there are careful coders, but if the code is not shared, it doesn't get the best wax job. --scryptr

EDIT: We could perhaps agree on how long my nVidia rigs should mine with your code, pre-release, pointed at your address. I could agree to keep the code confidential. --scryptr

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: chrysophylax on December 29, 2014, 08:59:48 PM

Quote from: djm34 on December 29, 2014, 08:55:22 PM

Quote from: scryptr on December 29, 2014, 08:47:11 PM

Quote from: djm34 on December 29, 2014, 08:26:58 PM

Quote from: scryptr on December 29, 2014, 07:11:10 PM

DJM34 --

Perhaps the original author of neoscrypt code meant for it to be difficult to port to nVidia.

? still not sure why you are saying that...
and actually doesn't seem to be the case, actually the 980 already beats the 290x/public kernel... (which is around 300kh/s)

Wolf0 answered for you. I thought that the code for neoscrypt may have been obfuscated deliberately in order to hinder its porting to Cuda code, perhaps in response to CBuchner's release of difficult-to-port "Killer Groestl". I'm wondering why development for nVidia is behind, but I am not slighting your efforts.

because there is no bounty Grin

(and paid project or project which are profitable have the priority... )

in that case - what would it take to 'work' on this project to make it worth its weight in gold?

i would be happily pointing my miners to 'donate' for such a project ...

or is there something that could be done in private that would benefit the contributors 'first' before a public release of such optimized code?

#crysx

I am wondering actually... how I will proceed.
1st solution one bounty address for public release, when the amount requested is reached the miner is released for everybody
2st a bounty address for private release once the amount is reached those who pledge for the private miner, gets the miner.
well it is always possible to contact me and propose something...

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1197. (Read 2347597 times)