Pages:
Author

Topic: DiabloMiner GPU Miner - page 39. (Read 866596 times)

legendary
Activity: 1162
Merit: 1000
DiabloMiner author
October 17, 2011, 01:40:07 AM
I have a 6990 and Debian unstable, which includes fglrx 11.9.1. I have tried AMD SDK 2.4 and 2.5. Either way, DiabloMiner uses about 90% of a CPU core when mining on one GPU, and about 180% (nearly 2 CPU cores) when mining on 2 GPUs.  It also causes lag when trying to play games on the primary GPU while mining on the second, which worked fine on fglrx 11.4.  I understand this is a driver bug, but how can I fix it?  Which driver can I install and how?  There's no fglrx driver at all in Debian testing (presumably due to bugs in 11.9).  Debian stable has no support for AMDOverdriveCtrl so I couldn't underclock my RAM.  I tried to downgrade my deb packages, but then got dependency problems with libmesa and couldn't install anything new.

Would AMD's installer for an intermediate version, like 11.7 work better?  Or can we just fix or report the actual bug in the driver?


Never try to use AMD's installer on a distro that provides packages. It usually just causes a mess.

Your only choice is to wait for AMD to fix it, but I suggest you keep trying to report the bug.

Games are probably lagging due to the CPU use bug, just turn DiabloMiner off when gaming.
full member
Activity: 213
Merit: 100
October 16, 2011, 10:00:18 PM
I have a 6990 and Debian unstable, which includes fglrx 11.9.1. I have tried AMD SDK 2.4 and 2.5. Either way, DiabloMiner uses about 90% of a CPU core when mining on one GPU, and about 180% (nearly 2 CPU cores) when mining on 2 GPUs.  It also causes lag when trying to play games on the primary GPU while mining on the second, which worked fine on fglrx 11.4.  I understand this is a driver bug, but how can I fix it?  Which driver can I install and how?  There's no fglrx driver at all in Debian testing (presumably due to bugs in 11.9).  Debian stable has no support for AMDOverdriveCtrl so I couldn't underclock my RAM.  I tried to downgrade my deb packages, but then got dependency problems with libmesa and couldn't install anything new.

Would AMD's installer for an intermediate version, like 11.7 work better?  Or can we just fix or report the actual bug in the driver?
hero member
Activity: 774
Merit: 500
Lazy Lurker Reads Alot
October 15, 2011, 09:09:34 AM
me bad lol you where right i forgot which i downloaded xD
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
October 08, 2011, 11:25:54 PM
I tried to get this running but constant crashes on Exception in thread "main"java.lang.UnsatisfiedLinkError: no lwjgl in java.library.path


Sounds like you tried to download it from git and didn't run mvn package first. Use the binary zip if you don't intend on building it.
hero member
Activity: 774
Merit: 500
Lazy Lurker Reads Alot
October 08, 2011, 04:58:45 PM
I tried to get this running but constant crashes on Exception in thread "main"java.lang.UnsatisfiedLinkError: no lwjgl in java.library.path
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
October 07, 2011, 12:43:50 AM
Update: Improve FPS timing to handle slow devices better
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
October 06, 2011, 06:07:23 AM
So I finally did a 10k share test of the new code. 10349 shares, 13 rejects, so 0.12% reject rate down from around 0.48%, testing on BTCGuild.

So yeah, I think I can say my new code does what it was supposed to.
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
October 04, 2011, 11:14:00 AM
Hi Diablo,
I have a Radeon 6670 running on Win7-64bit and after i upgraded the Catalyst to the last version (11.8 from 11.6) my hardware errors reported are between 20% and 25% and the CPU usage is 40% to 50% (before was under 10%).

Any idea what should i do or where i should look for some info?

Thanks.
edit: forgot to mention that i use the default configuration (launched the .exe using only with the -o, -r, -u and -p options).

Try adding -v 2 to see if it decreases HW errors.

Also, newer versions of Catalyst have a CPU use bug that effects all OpenCL apps. It cannot be fixed from within the app.
Thanks, Diablo.
The -v 2 option reduced the HW errors from 20-25% to 1-2% and increased the Mhs by 12%!

With 11.9 driver I started to have 10% HW errors but the CPU usage disappeared.
I added to the previous configuration -w 128 and now it all works perfectly Smiley (HW errors <0.5%, gained another 4% MHs and CPU usage is 1%)
Thanks



Huh, 10% you say? I wonder if that could be a driver bug.
member
Activity: 70
Merit: 10
October 04, 2011, 03:58:54 AM
Hi Diablo,
I have a Radeon 6670 running on Win7-64bit and after i upgraded the Catalyst to the last version (11.8 from 11.6) my hardware errors reported are between 20% and 25% and the CPU usage is 40% to 50% (before was under 10%).

Any idea what should i do or where i should look for some info?

Thanks.
edit: forgot to mention that i use the default configuration (launched the .exe using only with the -o, -r, -u and -p options).

Try adding -v 2 to see if it decreases HW errors.

Also, newer versions of Catalyst have a CPU use bug that effects all OpenCL apps. It cannot be fixed from within the app.
Thanks, Diablo.
The -v 2 option reduced the HW errors from 20-25% to 1-2% and increased the Mhs by 12%!

With 11.9 driver I started to have 10% HW errors but the CPU usage disappeared.
I added to the previous configuration -w 128 and now it all works perfectly Smiley (HW errors <0.5%, gained another 4% MHs and CPU usage is 1%)
Thanks

legendary
Activity: 1162
Merit: 1000
DiabloMiner author
October 03, 2011, 10:28:38 PM
Update: Try to drive the reject average further below 0.25%
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
September 17, 2011, 01:12:04 AM
No, it isn't a guideline. 1/3rd core clock for memory clock sits in a zone that on most Radeon 5xxxes it hits the stock memory timings correctly and incurs no speed loss for applications that don't rely on memory bandwidth.

If you're too low or too high, you incur a speed loss or sometimes the card just locks up.

Some kernels require better compliance with this than others.

except it is a guideline, because my 5750 is not stable with memory at 233 mhz
my 5850 card is faster as slightly more than 1/3, its core clock is is 725 and 275 is faster than both 242 and 300

you can blame the kernel, but phatk 2.2 is the fastest kernel on both cards and those timings are the fastest timings in practice
update: got a new card
at 725 core speed, 275 mem clock is faster than 250, and 240 gives artifacts

this is higher than 1/3
between 270 and 280 gives the best results

Hrm, if the timing is off (ie, for 1200 instead of 1000), it should be closer to 290 is the best.
hero member
Activity: 658
Merit: 500
September 17, 2011, 12:36:15 AM
No, it isn't a guideline. 1/3rd core clock for memory clock sits in a zone that on most Radeon 5xxxes it hits the stock memory timings correctly and incurs no speed loss for applications that don't rely on memory bandwidth.

If you're too low or too high, you incur a speed loss or sometimes the card just locks up.

Some kernels require better compliance with this than others.

except it is a guideline, because my 5750 is not stable with memory at 233 mhz
my 5850 card is faster as slightly more than 1/3, its core clock is is 725 and 275 is faster than both 242 and 300

you can blame the kernel, but phatk 2.2 is the fastest kernel on both cards and those timings are the fastest timings in practice
update: got a new card
at 725 core speed, 275 mem clock is faster than 250, and 240 gives artifacts

this is higher than 1/3
between 270 and 280 gives the best results
member
Activity: 78
Merit: 10
September 13, 2011, 04:39:00 AM
That was never really true. I test on many of the large pools. But if I can't reproduce it, its not a bug.
Ah, well I could agree with that.
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
September 13, 2011, 03:07:11 AM
And for future note, I'm going to treat all future bugs like this: If you're not using Eligius, it is not my problem.
Wait so does this still apply?

That was never really true. I test on many of the large pools. But if I can't reproduce it, its not a bug.
member
Activity: 78
Merit: 10
September 13, 2011, 02:37:51 AM
And for future note, I'm going to treat all future bugs like this: If you're not using Eligius, it is not my problem.
Wait so does this still apply?
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
September 12, 2011, 07:13:52 AM
Please note: Eligius is intentionally disabling rollntime for DiabloMiner users and tripling reject rates in the process. Use a different pool, such as btcguild which maintains a reject rate below 0.5%.
hero member
Activity: 658
Merit: 500
September 12, 2011, 04:20:07 AM
Phateus posted this graph:

does that look like 316 is the fastest? no, I'm pretty sure 410 is faster (vectors 2, worksize 256) right after the dip in speeds

and it obviously doesn't matter that much whether you're running 300ish or 400ish clocks according to the graph

Huh, I wonder what hes using for vectors, I assume he means uint4 = V4, etc. That graph is very interesting, it highlights the register spillover problem in the phatk design quite nicely.

I also wonder what card that is.

5870 overclocked, and v4 is indeed uint4

Those numbers might not be entirely valid then. (Some?) 1200mhz cards do not seem to have the same timing as 1000mhz cards, so 1/4th might work better. On my 5850, the peak seems to be around 1/3rd instead, and on some 5870s from what I've heard its still 1/3rd.
on MY 5850 275 is faster than 250 at 725 clock so the peak is higher than 1/3
I had to RMA it due to artifacts, so when I get a new one I'll test again and see if my new card differs
member
Activity: 70
Merit: 10
September 12, 2011, 04:08:21 AM
Hi Diablo,
I have a Radeon 6670 running on Win7-64bit and after i upgraded the Catalyst to the last version (11.8 from 11.6) my hardware errors reported are between 20% and 25% and the CPU usage is 40% to 50% (before was under 10%).

Any idea what should i do or where i should look for some info?

Thanks.
edit: forgot to mention that i use the default configuration (launched the .exe using only with the -o, -r, -u and -p options).

Try adding -v 2 to see if it decreases HW errors.

Also, newer versions of Catalyst have a CPU use bug that effects all OpenCL apps. It cannot be fixed from within the app.
Thanks, Diablo.
The -v 2 option reduced the HW errors from 20-25% to 1-2% and increased the Mhs by 12%!
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
September 11, 2011, 02:39:10 AM
Update: Cut network failure sleep in half, move execution threads from 2 to 3 to increase performance until AMD fixes CPU usage bug
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
September 08, 2011, 11:44:05 AM
Hi Diablo,
I have a Radeon 6670 running on Win7-64bit and after i upgraded the Catalyst to the last version (11.8 from 11.6) my hardware errors reported are between 20% and 25% and the CPU usage is 40% to 50% (before was under 10%).

Any idea what should i do or where i should look for some info?

Thanks.
edit: forgot to mention that i use the default configuration (launched the .exe using only with the -o, -r, -u and -p options).

Try adding -v 2 to see if it decreases HW errors.

Also, newer versions of Catalyst have a CPU use bug that effects all OpenCL apps. It cannot be fixed from within the app.
Pages:
Jump to: