Pages:
Author

Topic: [ANN] Yet another GPU miner release! [YAC] - page 9. (Read 53369 times)

sr. member
Activity: 441
Merit: 250
My 7870LE w/Tahiti chip 180khash
Code:
--worksize 256 -g 1 --lookup-gap 2 -I 15 --thread-concurrency 21000
hero member
Activity: 812
Merit: 1000


Those are the settings I use to run it, nothing left out. I do have 8 gigs of system ram cause it's cheap. I leave s, q and e alone. Half the time the pool communications crap out, so it's a problem on their end...

Oh and it's Win7.

Thanks man, I guess it's my ram holding me back. Same for scrypt in general, everything BUT sha256 which nets me 750kh/s
sr. member
Activity: 275
Merit: 250
I'm stuck at 223kh/s with my 7970...not too bad though. My pentium d does 1kh/s

I'm getting that on my 7950's. Shouldn't the 7970 get a bit more? Just curious...

Care to share your settings ? I can't go beyond 150 with my 7950.
Also what are your readings for WU in cgminer ? (mine avg 160)

Code:
yacminer --scrypt -o http://pool -u x -p x -w 256 -I 16 --thread-concurrency 21712 --gpu-memclock 1040 --gpu-engine 0-1040

I set memclock low because it does not have an effect on performance which is very strange since scrypt is memory hard, guess that only a part of the algo is gpu optimized ATM.
I have also noticed that lowering the memory speed boosts my WU from 100/110 at 1490 Mhz to 140/150 at 1040 Mhz.



Here's my yacminer stuff:

Code:
yacminer --scrypt -w 256 -I 18 --lookup-gap 2 -o http://yac.coinmine.pl:8088 -u x -p x --thread-concurrency 41000

Using Visiontek 7950s, core at 1000 and mem at 1500, voltage at stock, fans at 80%. WU will be rough estimate since I had to restart yacminer... damn pool died on me, again.... things should level out in a couple hours:

WU:  362/m

tried I 19 and it didn't like it  Wink

How much system memory and is that win or linux?
I'm at 225-250 with my 7970, i 13, tc 18368, lookup gap 3 g 2. 2gb system memory won't allow me to set higher TC.

Quick question: what is the best s, q and e setting? I get pool is not providing work fast enough from time to time.

Those are the settings I use to run it, nothing left out. I do have 8 gigs of system ram cause it's cheap. I leave s, q and e alone. Half the time the pool communications crap out, so it's a problem on their end...

Oh and it's Win7.
hero member
Activity: 812
Merit: 1000
I'm stuck at 223kh/s with my 7970...not too bad though. My pentium d does 1kh/s

I'm getting that on my 7950's. Shouldn't the 7970 get a bit more? Just curious...

Care to share your settings ? I can't go beyond 150 with my 7950.
Also what are your readings for WU in cgminer ? (mine avg 160)

Code:
yacminer --scrypt -o http://pool -u x -p x -w 256 -I 16 --thread-concurrency 21712 --gpu-memclock 1040 --gpu-engine 0-1040

I set memclock low because it does not have an effect on performance which is very strange since scrypt is memory hard, guess that only a part of the algo is gpu optimized ATM.
I have also noticed that lowering the memory speed boosts my WU from 100/110 at 1490 Mhz to 140/150 at 1040 Mhz.



Here's my yacminer stuff:

Code:
yacminer --scrypt -w 256 -I 18 --lookup-gap 2 -o http://yac.coinmine.pl:8088 -u x -p x --thread-concurrency 41000

Using Visiontek 7950s, core at 1000 and mem at 1500, voltage at stock, fans at 80%. WU will be rough estimate since I had to restart yacminer... damn pool died on me, again.... things should level out in a couple hours:

WU:  362/m

tried I 19 and it didn't like it  Wink

How much system memory and is that win or linux?
I'm at 225-250 with my 7970, i 13, tc 18368, lookup gap 3 g 2. 2gb system memory won't allow me to set higher TC.

Quick question: what is the best s, q and e setting? I get pool is not providing work fast enough from time to time.
sr. member
Activity: 275
Merit: 250
I'm stuck at 223kh/s with my 7970...not too bad though. My pentium d does 1kh/s

I'm getting that on my 7950's. Shouldn't the 7970 get a bit more? Just curious...

Care to share your settings ? I can't go beyond 150 with my 7950.
Also what are your readings for WU in cgminer ? (mine avg 160)

Code:
yacminer --scrypt -o http://pool -u x -p x -w 256 -I 16 --thread-concurrency 21712 --gpu-memclock 1040 --gpu-engine 0-1040

I set memclock low because it does not have an effect on performance which is very strange since scrypt is memory hard, guess that only a part of the algo is gpu optimized ATM.
I have also noticed that lowering the memory speed boosts my WU from 100/110 at 1490 Mhz to 140/150 at 1040 Mhz.



Here's my yacminer stuff:

Code:
yacminer --scrypt -w 256 -I 18 --lookup-gap 2 -o http://yac.coinmine.pl:8088 -u x -p x --thread-concurrency 41000

Using Visiontek 7950s, core at 1000 and mem at 1500, voltage at stock, fans at 80%. WU will be rough estimate since I had to restart yacminer... damn pool died on me, again.... things should level out in a couple hours:

WU:  362/m

tried I 19 and it didn't like it  Wink
hero member
Activity: 637
Merit: 500
I'm stuck at 223kh/s with my 7970...not too bad though. My pentium d does 1kh/s

I'm getting that on my 7950's. Shouldn't the 7970 get a bit more? Just curious...

Care to share your settings ? I can't go beyond 150 with my 7950.
Also what are your readings for WU in cgminer ? (mine avg 160)

Code:
yacminer --scrypt -o http://pool -u x -p x -w 256 -I 16 --thread-concurrency 21712 --gpu-memclock 1040 --gpu-engine 0-1040

I set memclock low because it does not have an effect on performance which is very strange since scrypt is memory hard, guess that only a part of the algo is gpu optimized ATM.
I have also noticed that lowering the memory speed boosts my WU from 100/110 at 1490 Mhz to 140/150 at 1040 Mhz.

legendary
Activity: 1242
Merit: 1020
No surrender, no retreat, no regret.
Code:
C:\DOCUME~1\GHO~1.TH~\LOCALS~1\Temp\OCL34E.tmp.cl(135): error: bad argument
          type to opencl vload/vstore: expected pointer to scalar with
          addrSpace global/local/private/constant(vloa [2013-06-26 09:42:44] Failed to init GPU thread 0, disabling device 0
 [2013-06-26 09:42:44] Restarting the GPU from the menu will not fix this.
 [2013-06-26 09:42:44] Try restarting cgminer.

What's your driver version? I think your OpenCL compiler may be too old. It's also possible to get rid of vstore but I'm not sure how that would affect performance.

That's HD4870 on Windows XP, so the drivers and APP SDK are pretty old. This OpenCL 1.0 HW delivers ~100MH/s SHA-256 at 800MHz engine and maybe 120W power consumption. Not worth mining anymore unless you have free electricity. I think it can be used very well for scrypt-jane mining as it offers over 100GB/s of memory bandwidth.

Here's a quick version that doesn't use vstore. Only recommended for people who cannot compile v6.

scrypt-chacha-v6-no-vstore
https://mega.co.nz/#!p9l20QLa!YbT2A0i0WqJS1moB6hGGaU4vRgJ7nChIKgs-2o1jB8k

Now getting from 20KH/s on HD4850 @ 700MHz engine & 2200MHz GDDR3 memory to 35KH/s on HD4890 @ 900MHz engine & 4000MHz GDDR5 memory. GDDR5 vs. GDDR3 makes big difference as I've expected. Maybe the code or settings can be optimised a bit more. HD4890 is supposed to deliver 1/3rd of HD7970 performance.

Code:
yacminer --scrypt --thread-concurrency 4000 --lookup-gap 4 -w 64 -I 16 --no-submit-stale --retries 10
efx
sr. member
Activity: 378
Merit: 250
The extra stream processors are often under utilized.
sr. member
Activity: 275
Merit: 250
Thanks for your continued development with this Mikaelh.

 Smiley
+1

N=10 settings for 560ti with no HW errors and v6 cl, 24kh
--scrypt --worksize 128 --lookup-gap 1 -I 11 --thread-concurrency 4192

Try higher thread concurrency and lookup-gap and then you'll be able to increase intensity.
You need a cudaminer :p

I'm stuck at 223kh/s with my 7970...not too bad though. My pentium d does 1kh/s

I'm getting that on my 7950's. Shouldn't the 7970 get a bit more? Just curious...
sr. member
Activity: 252
Merit: 250
Thanks for your continued development with this Mikaelh.

 Smiley
+1

N=10 settings for 560ti with no HW errors and v6 cl, 24kh
--scrypt --worksize 128 --lookup-gap 1 -I 11 --thread-concurrency 4192

Try higher thread concurrency and lookup-gap and then you'll be able to increase intensity.
You need a cudaminer :p

I'm stuck at 223kh/s with my 7970...not too bad though. My pentium d does 1kh/s
hehe yeah

was able to get I up to 15 with lookup-gap 2 and tc 8448 (crashed on 16) but the hashrate was still lower than lg 1 i 11
hero member
Activity: 812
Merit: 1000
Thanks for your continued development with this Mikaelh.

 Smiley
+1

N=10 settings for 560ti with no HW errors and v6 cl, 24kh
--scrypt --worksize 128 --lookup-gap 1 -I 11 --thread-concurrency 4192

Try higher thread concurrency and lookup-gap and then you'll be able to increase intensity.
You need a cudaminer :p

I'm stuck at 223kh/s with my 7970...not too bad though. My pentium d does 1kh/s
sr. member
Activity: 252
Merit: 250
Thanks for your continued development with this Mikaelh.

 Smiley
+1

N=10 settings for 560ti with no HW errors and v6 cl, 24kh
--scrypt --worksize 128 --lookup-gap 1 -I 11 --thread-concurrency 4192
sr. member
Activity: 266
Merit: 250
Thanks for your continued development with this Mikaelh.

 Smiley
sr. member
Activity: 301
Merit: 250
Code:
C:\DOCUME~1\GHO~1.TH~\LOCALS~1\Temp\OCL34E.tmp.cl(135): error: bad argument
          type to opencl vload/vstore: expected pointer to scalar with
          addrSpace global/local/private/constant(vloa [2013-06-26 09:42:44] Failed to init GPU thread 0, disabling device 0
 [2013-06-26 09:42:44] Restarting the GPU from the menu will not fix this.
 [2013-06-26 09:42:44] Try restarting cgminer.

What's your driver version? I think your OpenCL compiler may be too old. It's also possible to get rid of vstore but I'm not sure how that would affect performance.

That's HD4870 on Windows XP, so the drivers and APP SDK are pretty old. This OpenCL 1.0 HW delivers ~100MH/s SHA-256 at 800MHz engine and maybe 120W power consumption. Not worth mining anymore unless you have free electricity. I think it can be used very well for scrypt-jane mining as it offers over 100GB/s of memory bandwidth.

Here's a quick version that doesn't use vstore. Only recommended for people who cannot compile v6.

scrypt-chacha-v6-no-vstore
https://mega.co.nz/#!p9l20QLa!YbT2A0i0WqJS1moB6hGGaU4vRgJ7nChIKgs-2o1jB8k
sr. member
Activity: 301
Merit: 250
7790 users!

Please post here new N=10 settings. I try many setups, always hw error.

MY N9 settings:
hashing avg.: 126.3kh/s
Parameters:
Code:
 --device 0 --scrypt --worksize 256 --lookup-gap 2 -I 11 --thread-concurrency 8192 -o adress -u name -p pass

My current settings for a 7790 are:
Code:
--scrypt -w 128 --lookup-gap 2 -I 17 --thread-concurrency 12000

This produces about 109 kH/s after the latest N change.
legendary
Activity: 1242
Merit: 1020
No surrender, no retreat, no regret.
Code:
C:\DOCUME~1\GHO~1.TH~\LOCALS~1\Temp\OCL34E.tmp.cl(135): error: bad argument
          type to opencl vload/vstore: expected pointer to scalar with
          addrSpace global/local/private/constant(vloa [2013-06-26 09:42:44] Failed to init GPU thread 0, disabling device 0
 [2013-06-26 09:42:44] Restarting the GPU from the menu will not fix this.
 [2013-06-26 09:42:44] Try restarting cgminer.

What's your driver version? I think your OpenCL compiler may be too old. It's also possible to get rid of vstore but I'm not sure how that would affect performance.

That's HD4870 on Windows XP, so the drivers and APP SDK are pretty old. This OpenCL 1.0 HW delivers ~100MH/s SHA-256 at 800MHz engine and maybe 120W power consumption. Not worth mining anymore unless you have free electricity. I think it can be used very well for scrypt-jane mining as it offers over 100GB/s of memory bandwidth.
hero member
Activity: 820
Merit: 1000
It should literally be half the rate with all other settings being the same
sr. member
Activity: 441
Merit: 250
If one had 360 khash before what should they expect now? Trying to tune my miner probably to maximize my hash rate. Best I can do before getting swamped with HW errors is about 160khash on a 7870LE Tahiti.
newbie
Activity: 11
Merit: 0
7790 users!

Please post here new N=10 settings. I try many setups, always hw error.

MY N9 settings:
hashing avg.: 126.3kh/s
Parameters:
Code:
 --device 0 --scrypt --worksize 256 --lookup-gap 2 -I 11 --thread-concurrency 8192 -o adress -u name -p pass
sr. member
Activity: 301
Merit: 250
Code:
C:\DOCUME~1\GHO~1.TH~\LOCALS~1\Temp\OCL34E.tmp.cl(135): error: bad argument
          type to opencl vload/vstore: expected pointer to scalar with
          addrSpace global/local/private/constant(vloa [2013-06-26 09:42:44] Failed to init GPU thread 0, disabling device 0
 [2013-06-26 09:42:44] Restarting the GPU from the menu will not fix this.
 [2013-06-26 09:42:44] Try restarting cgminer.

What's your driver version? I think your OpenCL compiler may be too old. It's also possible to get rid of vstore but I'm not sure how that would affect performance.
Pages:
Jump to: