CCminer(SP-MOD) Modded GPU kernels.

[myagui]

Most important recommendation I have, is to benchmark at a low'ish & fixed GPU clock setting (stuck in performance mode), along with a high'ish & fixed fan speed. This should eliminate most/all thermal & clock stability variations.

For significant code changes, and particularly when the changes are not strictly related to the hashing algorithm in particular - so more about backend/generally efficiency - definitely need to test with a pool, Nicehash is usually great at that, with a minimum interval of 4 hours, but ideally, a full 24 hour cycle.

For specific kernel changes, benchmark should be best (and also the practical approach). I tend to benchmark for whatever duration is necessary until I see that I get about 5 minutes of stable output speed & stable temperature reading. Some algorithms take a while ramping up, others not so much, some algorithms start faster, but slow down once temperatures get high enough.

[bathrobehero]

What's the most accurate way to benchmark kernel modifications?
Using --benchmark or connecting to a pool? What time interval? Any other reccomendation?

Poolmining is never completely reliable, especially if there's vardiff which will always fluctuate. But even with fixed diff you'll have to run it for a long time to be sure.
--benchmark had plenty of issues in the past so I've stopped using it so I'm not sure how it is recently.

For checking speed only, I personally like to just solomine to any local wallet, doesn't even matter if the algo is different and just check the reported hashrate.
Stable clocks and fix fanspeed recommended.

Regarding aluminum cases, I'm also curious how much people who own them paid for them.

[chrysophylax]

What's the most accurate way to benchmark kernel modifications?
Using --benchmark or connecting to a pool? What time interval? Any other reccomendation?

Poolmining is never completely reliable, especially if there's vardiff which will always fluctuate. But even with fixed diff you'll have to run it for a long time to be sure.
--benchmark had plenty of issues in the past so I've stopped using it so I'm not sure how it is recently.

For checking speed only, I personally like to just solomine to any local wallet, doesn't even matter if the algo is different and just check the reported hashrate.
Stable clocks and fix fanspeed recommended.

Regarding aluminum cases, I'm also curious how much people who own them paid for them.

frames? ... i build them custom ...

the frame ( 1.5mm square aluminium tubing ) - edge connectors - screws and lugs - angle iron ( 90 degree edging ) ... all this comes to approx $70.00aud ...

but initially i had to buy the cutting and measuring equipment as well as the files and mallets ( yes plastic head mallets - not hammers ) ...

parts alone - not counting labour - if you are looking at adding teh countless hours to design the thing in the first place ...

prototype 1 - prototype 2 and now prototype 3 is probably the design we are going with ...

hope that helps bathrobehero ...

#crysx

[pallas]

I must say that I didn't have any problem benchmarking amd cards: if the room was hot, I'd put the fans at high speed, run the miner and wait a couple minute to stabilize, and that's it. I could make 100 changes to a kernel in a day and check them all, accurately.

On nvidia I have throttling problems I can't easily fix (the cards reduce clock speed in a number of situations I just can't predict), overclocking/downclocking is more difficult as the cards tend to change clocks by themselves, and the hashrates fluctuates wildly, and even changes between ccminer runs.
The rig is headless so I only have nvidia-smi to work with, and it can't set the fan speed.
So when I make a little kernel speedup, I spend more time benchmarking it (to be sure it's indeed an improvement), than making the improvement itself :-/
Maybe there are some nvidia-smi settings to make it more stable?
Or maybe on windows it's different...
Finally I may need a workstation with a nvidia as main card, and work on it.

[myagui]

Or maybe on windows it's different...

Yeps. On Windows it is extremely simple to set fixed clocks, fans, overclocking, so you can easily have a "benchmark platform".

With a headless Linux system I don't think there's any solution for fixed fans yet, others might know differently. I previously had fixed fans and clocks on Linux, but I perfectly recall that I specifically had to configure/attach a monitor in order to get that working at the time.

[sp_]

I must say that I didn't have any problem benchmarking amd cards: if the room was hot, I'd put the fans at high speed, run the miner and wait a couple minute to stabilize, and that's it. I could make 100 changes to a kernel in a day and check them all, accurately.

On nvidia I have throttling problems I can't easily fix (the cards reduce clock speed in a number of situations I just can't predict), overclocking/downclocking is more difficult as the cards tend to change clocks by themselves, and the hashrates fluctuates wildly, and even changes between ccminer runs.
The rig is headless so I only have nvidia-smi to work with, and it can't set the fan speed.
So when I make a little kernel speedup, I spend more time benchmarking it (to be sure it's indeed an improvement), than making the improvement itself :-/
Maybe there are some nvidia-smi settings to make it more stable?
Or maybe on windows it's different...
Finally I may need a workstation with a nvidia as main card, and work on it.

Buy another 970 card. The gigabyte windforce oc never trottle and mines on a stable clockrate. easy to verify speedups. For very small changes, taka a look at the generated PTX assembly code, less code lines is bether but not always..
You can also test your chances on a big rig with many cards, If you have many cards small speedups of 0-1 KHASH per card can be visible.

To finance the cards, you can hope that you will ROI the cards in 1 year by increasing the speed of the kernals

[Slava_K]

TPB52 7 is best for mining. My gtx 980 make 11.2-11.6 Mh. 960 gtx make 6 Mh. Super!
For rig (2x980+1x960) hash is 28500!
PS. It is LYRA2v2! +100 git 1052 works!

[sp_]

These settinge should be taken out of the hardcoded kernal and be adjustable in the commandline. Like in scrypt.   -l 7x19 (7threads per block with -X intensity 19).
Because on the 970 9 seems to be bether.

[Slava_K]

Ok! Thanks.

[bensam1231]

I must say that I didn't have any problem benchmarking amd cards: if the room was hot, I'd put the fans at high speed, run the miner and wait a couple minute to stabilize, and that's it. I could make 100 changes to a kernel in a day and check them all, accurately.

On nvidia I have throttling problems I can't easily fix (the cards reduce clock speed in a number of situations I just can't predict), overclocking/downclocking is more difficult as the cards tend to change clocks by themselves, and the hashrates fluctuates wildly, and even changes between ccminer runs.
The rig is headless so I only have nvidia-smi to work with, and it can't set the fan speed.
So when I make a little kernel speedup, I spend more time benchmarking it (to be sure it's indeed an improvement), than making the improvement itself :-/
Maybe there are some nvidia-smi settings to make it more stable?
Or maybe on windows it's different...
Finally I may need a workstation with a nvidia as main card, and work on it.

Buy another 970 card. The gigabyte windforce oc never trottle and mines on a stable clockrate. easy to verify speedups. For very small changes, taka a look at the generated PTX assembly code, less code lines is bether but not always..
You can also test your chances on a big rig with many cards, If you have many cards small speedups of 0-1 KHASH per card can be visible.

To finance the cards, you can hope that you will ROI the cards in 1 year by increasing the speed of the kernals

Yeah the one plus side to the Gigabyte cards is they never seem to throttle and clock the highest. Of course they use like 10-15% more power then all the other cards, so...

[sp_]

Yeah the one plus side to the Gigabyte cards is they never seem to throttle and clock the highest. Of course they use like 10-15% more power then all the other cards, so...

As I said, my Zotac card start to mine quark at 15.5 mhash, and then it trottles and fall down to 13.5-14. My gigabyte is stable at 17MHASH. 3,5 MHASH bether.

http://www.zotac.com/products/graphics-cards/geforce-900-series/product/geforce-900-series/detail/geforce-gtx-970.html

The zotac card is much smaller and heats up faster.

[pallas]

Yeah the one plus side to the Gigabyte cards is they never seem to throttle and clock the highest. Of course they use like 10-15% more power then all the other cards, so...

As I said, my Zotac card start to mine quark at 15.5 mhash, and then it trottles and fall down to 13.5-14. My gigabyte is stable at 17MHASH. 3,5 MHASH bether.

http://www.zotac.com/products/graphics-cards/geforce-900-series/product/geforce-900-series/detail/geforce-gtx-970.html

The zotac card is much smaller and heats up faster.

I have an msi 970 that is more stable than the others, but still it starts at 16.5 and goes down to 16 after a while.
Well, I may get a gigabyte or just wait for the winter ;-)

[Slava_K]

Asus Strix 980GTX 20.8 Mh Quark 1380 MHz.

[myagui]

These settinge should be taken out of the hardcoded kernal and be adjustable in the commandline. Like in scrypt.   -l 7x19 (7threads per block with -X intensity 19).
Because on the 970 9 seems to be bether.

@sp_:
Shouldn't the optimal blocks/threads values always be relative to the number of SMM/SMXs in any given card? It might be possible to have ccminer automatically detect - and adjust - to the number of SMM/SMXs on the available cards, and then just use the intensity parameter for the fine tuning.

I lost this bit of information somewhere along the thread history, but seeing as you now have -X , is -i still doing anything on your fork? If both parameters can be used, what the F#%@ is each one doing? Thanks for clearing this up 

It's starting to look a bit mad to find the best settings for one's card, if there are 3 distinct parameters to tune (-X, -i, -l).

[badgoes]

Or maybe on windows it's different...

Yeps. On Windows it is extremely simple to set fixed clocks, fans, overclocking, so you can easily have a "benchmark platform".

With a headless Linux system I don't think there's any solution for fixed fans yet, others might know differently. I previously had fixed fans and clocks on Linux, but I perfectly recall that I specifically had to configure/attach a monitor in order to get that working at the time.

to set the fan on linux just test this

https://gist.github.com/squadbox/e5b5f7bcd86259d627ed

[t-nelson]

These settinge should be taken out of the hardcoded kernal and be adjustable in the commandline. Like in scrypt.   -l 7x19 (7threads per block with -X intensity 19).
Because on the 970 9 seems to be bether.

@sp_:
Shouldn't the optimal blocks/threads values always be relative to the number of SMM/SMXs in any given card? It might be possible to have ccminer automatically detect - and adjust - to the number of SMM/SMXs on the available cards, and then just use the intensity parameter for the fine tuning.

Seems reasonable.  I don't think we can directly query the physical configuration of the chipset.  But it wouldn't be much to maintain a LUT.  The major hurdle would probably be reliably parsing the compilation output for the kernel.

[myagui]

I remember getting the number of SMM/SMXs in the CryptoNight miner... can't remember how, though.

Yep. TSIV's Cryptonight ccminer complains whenever the blocks/threads (don't recall which one now) is not a multiple of the SMX/SMM on the card. IIRC, it does this also accurately for cards that did not exist when it was released, so the approach is not LUT based.

[pallas]

Or maybe on windows it's different...

Yeps. On Windows it is extremely simple to set fixed clocks, fans, overclocking, so you can easily have a "benchmark platform".

With a headless Linux system I don't think there's any solution for fixed fans yet, others might know differently. I previously had fixed fans and clocks on Linux, but I perfectly recall that I specifically had to configure/attach a monitor in order to get that working at the time.

to set the fan on linux just test this

https://gist.github.com/squadbox/e5b5f7bcd86259d627ed

Thanks, but I fear it needs a monitor or the x session will not start... Or will it?

[joblo]

Or maybe on windows it's different...

Yeps. On Windows it is extremely simple to set fixed clocks, fans, overclocking, so you can easily have a "benchmark platform".

With a headless Linux system I don't think there's any solution for fixed fans yet, others might know differently. I previously had fixed fans and clocks on Linux, but I perfectly recall that I specifically had to configure/attach a monitor in order to get that working at the time.

to set the fan on linux just test this

https://gist.github.com/squadbox/e5b5f7bcd86259d627ed

Thanks, but I fear it needs a monitor or the x session will not start... Or will it?

Look back about 10 pages in this thread, I got an xsession, OC and fan control on a headless card. Hashbrown
pointed to another thread with a detailed procedure for doing this. Coincidentally it only offers fixed fan settings.
The OC is a fixed offset from the base clock which can throttle.

[badgoes]

Or maybe on windows it's different...

Yeps. On Windows it is extremely simple to set fixed clocks, fans, overclocking, so you can easily have a "benchmark platform".

With a headless Linux system I don't think there's any solution for fixed fans yet, others might know differently. I previously had fixed fans and clocks on Linux, but I perfectly recall that I specifically had to configure/attach a monitor in order to get that working at the time.

to set the fan on linux just test this

https://gist.github.com/squadbox/e5b5f7bcd86259d627ed

Thanks, but I fear it needs a monitor or the x session will not start... Or will it?

i have no monitor connected
but i have a linux 14.04 desktop version