CCminer(SP-MOD) Modded GPU kernels.

[djm34]

15BTC per copy is $7500~ per copy. How many do you plan on selling? 2?

Quark alone JUST on Nicehash has 60GH. At 2% that's 1.2GH. That's .48BTC per day, for Quark alone and that's a smaller one. That's $18,144 per month at current BTC prices. This doesn't include ANY pools outside of Nicehash and we all know those don't exist.

15BTC= $7,500
0.48BTC/day * 30= 14.4BTC= $18,144 per month
Wow  

  7500$ yeah may-be last year  
 

(wish I knew the exchanges he uses...)
  I guess I won't hire him as accountant (marketing may-be...)

[CapnBDL]

         <<NEWS FLASH>>                                                     <<NEWS FLASH>>                                             <<NEWS FLASH>>

  YAAMP POOL OWNER CALLS IT QUITS                      YAAMP POOL OWNER CALLS IT QUITS                    YAAMP POOL OWNER CALLS IT QUITS

        <<NEWS FLASH>>                                                      <<NEWS FLASH>>                                             <<NEWS FLASH>>


   This has been a special announcement by CapnBDL


                                                                                                                                                                        

[djm34]

         <<NEWS FLASH>>                                                     <<NEWS FLASH>>                                             <<NEWS FLASH>>

  YAAMP POOL OWNER CALLS IT QUITS                      YAAMP POOL OWNER CALLS IT QUITS                    YAAMP POOL OWNER CALLS IT QUITS

        <<NEWS FLASH>>                                                      <<NEWS FLASH>>                                             <<NEWS FLASH>>


   This has been a special announcement by CapnBDL


                                                                                                                                                                                

and we care because ?

[Fuzzbawls]

I think this thread needs a fork of it's own

[CapnBDL]

Sorry djm34...thought some might. Are there any mult-algo trading pools out there I can quark mine on & trade for x11?

[djm34]

Sorry djm34...thought some might. Are there any mult-algo trading pools out there I can quark mine on & trade for x11?

no idea, I am not much into multipool... so I guess I lost my chance to make 18k$ in fee 

[scryptr]

RELEASE 1.53--

I haven't posted in a while, I was disapointed with releases 1.51 and 1.52.  Both would run, but crash over time on both my Linux and Win 7 machines.  Release 1.53 runs smoothly enough on my Win 7 GTX 960, mining Quark.  However, I don't see any significant change in hashing speed since v1.50, my fallback.

I think it was the reduction in default intensity that made things better.  Perhaps SP_ could provide default intensity info for the different algos, and we could each find the top intensity for our own rigs by adjusting from the default.

Another thing, in mining Lyra2 my GTX 960 cards mine more slowly than a 750ti.  I don't understand it.  I think there may have been another post on this 10-20 pages back, but this thread is getting long.                  --scryptr

[joblo]

Another thing, in mining Lyra2 my GTX 960 cards mine more slowly than a 750ti.  I don't understand it.  I think there may have been another post on this 10-20 pages back, but this thread is getting long.                  --scryptr

It seems lyra2 is memory bound. Unless the 960 has more memory I wouldn't expect a big increase, but I
wouldn't expect a decrease either.

[bathrobehero]

Another thing, in mining Lyra2 my GTX 960 cards mine more slowly than a 750ti.  I don't understand it.  I think there may have been another post on this 10-20 pages back, but this thread is getting long.                  --scryptr

It seems lyra2 is memory bound. Unless the 960 has more memory I wouldn't expect a big increase, but I
wouldn't expect a decrease either.

It seems to be memory bandwidth bound, not memory amount bound:



Although a 960 should surely be faster than a 750 Ti.

[scryptr]

Another thing, in mining Lyra2 my GTX 960 cards mine more slowly than a 750ti.  I don't understand it.  I think there may have been another post on this 10-20 pages back, but this thread is getting long.                  --scryptr

It seems lyra2 is memory bound. Unless the 960 has more memory I wouldn't expect a big increase, but I
wouldn't expect a decrease either.

I have a 4GB GTX 960 FTW and a 2GB GTX 960 SSC, both run lame compared to a 750ti when mining Lyra2.       --scryptr

[sp_]

RELEASE 1.53--
I haven't posted in a while, I was disapointed with releases 1.51 and 1.52.  Both would run, but crash over time on both my Linux and Win 7 machines.  Release 1.53 runs smoothly enough on my Win 7 GTX 960, mining Quark.  However, I don't see any significant change in hashing speed since v1.50, my fallback.
I think it was the reduction in default intensity that made things better.  Perhaps SP_ could provide default intensity info for the different algos, and we could each find the top intensity for our own rigs by adjusting from the default.
Another thing, in mining Lyra2 my GTX 960 cards mine more slowly than a 750ti.  I don't understand it.  I think there may have been another post on this 10-20 pages back, but this thread is getting long.                  --scryptr

I have 2 970 test cards. 1 gigabyte windforce oc (8pin + 6pin power), and one zotec

The zotec have 2 6pins power connectors, and only 2 fans. (almost half the size of the gigabyte card.

The gigabyte does 15.8 MHASH with -i 24
the zotec starts at 14.5 MHASH and after a while the gpu is trottleing (to hot) and the speed is reduced to 13.7 MHASH..

in the latest on github I have reduced the default intensity to 22.9 for compute 5.2. Seems to be a mistake, since I am loosing 300kHASH on the gigabyte card. On the zotec card it seems to improve a bit with lower intensity (less trottleing)

[scryptr]

RELEASE 1.53--

I get 10.45Mh/s on my GTX 960 SSC mining quark.  This is with the intensity at 24, and an overclock of +80/240, core/mem. Intensities of 24.1 or higher fail on launch, and most often crash the driver.

I was getting 10.6Mh/s with -i 23.7, and OC of +100/300, core/mem, but it became unstable in the last few versions.

The Lyra2 perfomance really puzzles me, though.  It should run faster than a 750ti.       --scryptr

[djm34]

RELEASE 1.53--

I get 10.45Mh/s on my GTX 960 SSC mining quark.  This is with the intensity at 24, and an overclock of +80/240, core/mem. Intensities of 24.1 or higher fail on launch, and most often crash the driver.

I was getting 10.6Mh/s with -i 23.7, and OC of +100/300, core/mem, but it became unstable in the last few versions.

The Lyra2 perfomance really puzzles me, though.  It should run faster than a 750ti.       --scryptr

you should monitor FB usage (Frame Buffer) with msi ab, if it is too high, then you should decrease the intensity.
A rule of thumb for the intensity is to get it to 99~100% gpu usage (starting with lower value obviously) and stops when you reach 99% there is no point in pushing it higher.

Some progress with neoscrypt:
750ti: 186kh/s (was 150~160)
780ti: 430kh/s (was ~350)
980    735kh/s (was ~650)

[antonio8]

RELEASE 1.53--

I get 10.45Mh/s on my GTX 960 SSC mining quark.  This is with the intensity at 24, and an overclock of +80/240, core/mem. Intensities of 24.1 or higher fail on launch, and most often crash the driver.

I was getting 10.6Mh/s with -i 23.7, and OC of +100/300, core/mem, but it became unstable in the last few versions.

The Lyra2 perfomance really puzzles me, though.  It should run faster than a 750ti.       --scryptr

I have the same issue. I have a Evga GTX960 SSC 4GB and it is slower than my 750ti SC 2GB. I am on Windows 7 64 bit.

[bathrobehero]

RELEASE 1.53--

I get 10.45Mh/s on my GTX 960 SSC mining quark.  This is with the intensity at 24, and an overclock of +80/240, core/mem. Intensities of 24.1 or higher fail on launch, and most often crash the driver.

I was getting 10.6Mh/s with -i 23.7, and OC of +100/300, core/mem, but it became unstable in the last few versions.

The Lyra2 perfomance really puzzles me, though.  It should run faster than a 750ti.       --scryptr

Does the memory overclock help at all?

[scryptr]

RELEASE 1.53--

I get 10.45Mh/s on my GTX 960 SSC mining quark.  This is with the intensity at 24, and an overclock of +80/240, core/mem. Intensities of 24.1 or higher fail on launch, and most often crash the driver.

I was getting 10.6Mh/s with -i 23.7, and OC of +100/300, core/mem, but it became unstable in the last few versions.

The Lyra2 performance really puzzles me, though.  It should run faster than a 750ti.       --scryptr

Does the memory overclock help at all?

Lyra2 and GTX960--

My Lyra2 rig has 5x 750ti SC, and one GTX 960 FTW (4GB).  I can overclock all cards with nVidia Inspector and get bigger numbers, on Win 8.0, but the rig isn't stable.  At stock clocks, the 750ti cards get 725kh/s, and the 960 gets 500-600kh/s.  Memory overclock helped some, but the rig would not run more than a few hours.  The 960 runs like a lame dog either way.

With stock clocks, the rig runs for days on Lyra2.       --scryptr

[MaxDZ8]

I've stopped following NV a couple of years ago so take it with some salt...

As far as I've understood, the ALU-to-memory resource ratios in small maxwell are much more appropriate to efficient memory usage. Approximated: there are more "memory ports" compared to ALU count. What could be happening is that big GPUs ALUs are tripping on each other. Very approximated: every ALU has "more chance" to get to memory.
It is also possible to reproduce this on GCN1.0 (which has quite smaller buffers BTW).

Try measure the amount of memory effectively moved. I'd expect it to be much bigger for big maxwell and in both cases far higher than it should.

As a side note: "memory load" is unlikely to be what you think it is. It's most likely "memory controller active"; it could be active but stalled (~ shut down but burning energy).

[sp_]

RELEASE 1.53--
I get 10.45Mh/s on my GTX 960 SSC mining quark.  This is with the intensity at 24, and an overclock of +80/240, core/mem. Intensities of 24.1 or higher fail on launch, and most often crash the driver.
I was getting 10.6Mh/s with -i 23.7, and OC of +100/300, core/mem, but it became unstable in the last few versions.
The Lyra2 perfomance really puzzles me, though.  It should run faster than a 750ti.       --scryptr

you should monitor FB usage (Frame Buffer) with msi ab, if it is too high, then you should decrease the intensity.
A rule of thumb for the intensity is to get it to 99~100% gpu usage (starting with lower value obviously) and stops when you reach 99% there is no point in pushing it higher.
Some progress with neoscrypt:
750ti: 186kh/s (was 150~160)
780ti: 430kh/s (was ~350)
980    735kh/s (was ~650)

sp-mod 53 is doing 190Kh/s on the 750 (non ti) with overclocking.

my blake rewrite added 15-20KHASH from your first public version.

[djm34]

RELEASE 1.53--
I get 10.45Mh/s on my GTX 960 SSC mining quark.  This is with the intensity at 24, and an overclock of +80/240, core/mem. Intensities of 24.1 or higher fail on launch, and most often crash the driver.
I was getting 10.6Mh/s with -i 23.7, and OC of +100/300, core/mem, but it became unstable in the last few versions.
The Lyra2 perfomance really puzzles me, though.  It should run faster than a 750ti.       --scryptr

you should monitor FB usage (Frame Buffer) with msi ab, if it is too high, then you should decrease the intensity.
A rule of thumb for the intensity is to get it to 99~100% gpu usage (starting with lower value obviously) and stops when you reach 99% there is no point in pushing it higher.
Some progress with neoscrypt:
750ti: 186kh/s (was 150~160)
780ti: 430kh/s (was ~350)
980    735kh/s (was ~650)

sp-mod 53 is doing 190Kh/s on the 750 (non ti) with overclocking.

my blake rewrite added 15-20KHASH from your first public version.

huh ? it is based on your version... (1.5.50) and incorporate that changes already, so whatever is the original speed this one does 30kh/s more

[sp_]

2 sp_
Tried your #49 with neoscrypt

I tried djm's version before and for my poor gtx 750 1gb it runs only with -F 1 and gives 175khs.

Default run with your #49
start /realtime ccminer.exe --cpu-priority 5 -a neoscrypt

it takes almost all gpu memory and pushes load to "Bus interface" as it called in GPU-Z.
With such conditions performance is poor - only 45 khs.
But i tried to play with -i parameter and found that it runs with maximum -i 14.5
With such config it shows 810MB memory usage, 0% bus interface load, 99% gpu load, 30-35% memory controller load.
The numbers are 187khs and this is better then old djm's version. But i didn't check on the pool yet .

So I think you should use not so heavy default launch config for this algo for low-level gpu's like mine.

Could you try release 53 please. Is perfomance lost?