Pages:
Author

Topic: DiabloMiner GPU Miner - page 31. (Read 866596 times)

legendary
Activity: 1162
Merit: 1000
DiabloMiner author
February 08, 2012, 03:06:16 PM
"-v 2 -w 128" results in the best performance, but also about 15% hw errors on my 6970.

Current snapshot: accept: 79 | reject: 2 | hw error: 11

Question: What exactly does this hw error mean?

It means your hardware or drivers may not be functioning correctly. Are you overclocking?
hero member
Activity: 1162
Merit: 500
February 08, 2012, 03:05:01 PM
"-v 2 -w 128" results in the best performance, but also about 15% hw errors on my 6970.

Current snapshot: accept: 79 | reject: 2 | hw error: 11

Question: What exactly does this hw error mean?
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
February 08, 2012, 02:48:15 PM

anyone having problems using diablo with gpumax.com ?   

syptoms:
  start.  works OK.  hashrate good.
  during:  start to get connection issues reported by diablo, but it reconnects and works.
  after a while:  the connection issues increase to the point that no shares are accepted.  but it reports a full hashrate.

fix:  stop and start diablo


I thought gpumax was a scam?

apparently it is the Long Polling that is causing it with gpumax..  can I turn off LP with diablo?


No. Tell gpumax to fix their software, no other pool is having this issue.
legendary
Activity: 1876
Merit: 1000
February 08, 2012, 02:20:54 PM

anyone having problems using diablo with gpumax.com ?   

syptoms:
  start.  works OK.  hashrate good.
  during:  start to get connection issues reported by diablo, but it reconnects and works.
  after a while:  the connection issues increase to the point that no shares are accepted.  but it reports a full hashrate.

fix:  stop and start diablo


I thought gpumax was a scam?

apparently it is the Long Polling that is causing it with gpumax..  can I turn off LP with diablo?
rjk
sr. member
Activity: 448
Merit: 250
1ngldh
February 08, 2012, 01:33:42 PM
I thought gpumax was a scam?
Nope, working fine here on cgminer currently. Payout AOK.
hero member
Activity: 497
Merit: 500
February 08, 2012, 11:52:47 AM
Diablo I have 3 7970's let me know if you need testing done.  Right now I am using Diablo no flags getting 680MHash/s 1144core/685mem.

Don't need external testing. The community donated enough that I can go buy a 7970 of my own now (thanks everyone!), but my 5850 testing just isn't done yet. I think after this current round of optimization hunting I'll finally switch.

I have 3 5850's let me know if that can help. Currently running Diablo on two of them.
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
February 08, 2012, 10:24:29 AM
Diablo I have 3 7970's let me know if you need testing done.  Right now I am using Diablo no flags getting 680MHash/s 1144core/685mem.

Don't need external testing. The community donated enough that I can go buy a 7970 of my own now (thanks everyone!), but my 5850 testing just isn't done yet. I think after this current round of optimization hunting I'll finally switch.
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
February 08, 2012, 10:22:25 AM

anyone having problems using diablo with gpumax.com ?  

syptoms:
  start.  works OK.  hashrate good.
  during:  start to get connection issues reported by diablo, but it reconnects and works.
  after a while:  the connection issues increase to the point that no shares are accepted.  but it reports a full hashrate.

fix:  stop and start diablo


I thought gpumax was a scam?

Edit: Apparently its not, others on the forum are just morons
legendary
Activity: 1876
Merit: 1000
February 08, 2012, 09:48:07 AM

anyone having problems using diablo with gpumax.com ?   

syptoms:
  start.  works OK.  hashrate good.
  during:  start to get connection issues reported by diablo, but it reconnects and works.
  after a while:  the connection issues increase to the point that no shares are accepted.  but it reports a full hashrate.

fix:  stop and start diablo
hero member
Activity: 497
Merit: 500
February 08, 2012, 09:42:16 AM
Diablo I have 3 7970's let me know if you need testing done.  Right now I am using Diablo no flags getting 680MHash/s 1144core/685mem.
newbie
Activity: 7
Merit: 0
February 08, 2012, 08:35:04 AM
Awesome job Diablo...I can't wait to see what the end results are.

I just moved over to your miner after GUIMiner went down and cgminer simply wasn't performing.  I'm happy with the performance on my current setup, but am closely following your progress as I have plans for GCN miners in the near future.
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
February 08, 2012, 03:38:52 AM
Done stage 1:

Old kernel:
-v 1 sdk 2.6 = 883 ops 13 registers, 2.1 = 803 ops 17 registers
-v 2 sdk 2.6 = 1503 ops 21 registers, 2.1 = 1362 ops 21 registers

2.6 is 10% slower than 2.1.

New kernel:
-v 1 sdk 2.6 = 886 ops 23 registers, 887 ops 21 registers
-v 2 sdk 2.6 = 1395 39 registers, 2.1 = 1396 ops 36 registers

2.6 is tied with 2.1.

New vs old:
-v 1 =~ 0% faster 2.6, 10% slower 2.1, 2.6 is 30% away from minimum target of 681 ops
-v 2 =~ 8% faster 2.6, 2% slower 2.1, 2.6 is 2% away from minimum target of 1362 ops

If I committed this right now, 2.1 users could switch to 2.6 and not see a real difference in speed.

Now I begin stage 2: optimum ALU clause packing and register use reduction.

Edit: Somehow I got the new numbers wrong, lets try that again.
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
February 07, 2012, 11:56:41 PM
After looking at the ops generated in the .isa output, I'm clearly starting to force 2.6 to act more like the 2.1 results, so I seem to be getting somewhere. The hard part is going to be ordering instructions to properly saturate ALU clauses on vliw5 and GCN both.

Wouldn't it be easier and more efficient just to have a separate kernel for GCN? It's a totally different architecture so it would make sense to, right?

Nope, that makes no sense. Its still the same compiler, and OpenCL is a generic language meant for wildly different hardware. This kernel runs on everything from x86 to ARM to Sparc to POWER and PPC to Cell SPUs to Intel GPUs (if/when Intel actually fixes their drivers) to Nvidia to even certain DSPs.

A compiler is free to completely ignore the specific ordering of instructions I am using in the kernel and optimize it correctly for it's target hardware. AMD's doesn't, and takes significant hinting from the code to allow developers to influence the output. This is arguably both right and wrong, given how much code effects/is effected by bandwidth usage and latency and other things.

The difference between 2.6 and 2.1 is that 2.6 is tuned differently, and 2.6 allows much more hinting from the kernel layout. The compiler could freely reorder everything safely to give full performance, its just difficult on complex hardware like this; 2.1 went farther into this than 2.6 does.

I'm slightly surprised I actually need to do this, however, because AMD uses LLVM in their compiler, which is very good at SSA tree optimizations. Arguably, its rather good at what I'm doing by hand. Wink
full member
Activity: 131
Merit: 100
February 07, 2012, 10:08:57 PM
After looking at the ops generated in the .isa output, I'm clearly starting to force 2.6 to act more like the 2.1 results, so I seem to be getting somewhere. The hard part is going to be ordering instructions to properly saturate ALU clauses on vliw5 and GCN both.

Wouldn't it be easier and more efficient just to have a separate kernel for GCN? It's a totally different architecture so it would make sense to, right?
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
February 07, 2012, 09:22:41 PM
Which flags do I have to use to get the best out of my 7970?

ATM -v 1, but who knows once I finish this more optimized kernel.

Any faster so far?

This is a two stage project. I'm almost done stage 1.

Old kernel:
-v 1 sdk 2.6 = 883 ops 13 registers, 2.1 = 803 ops 17 registers
-v 2 sdk 2.6 = 1503 ops 21 registers, 2.1 = 1362 ops 21 registers

2.6 is 10% slower than 2.1.

New kernel:
-v 1 sdk 2.6 = 827 ops 21 registers, 828 ops 20 registers
-v 2 sdk 2.6 = 1464 36 registers, 2.1 = 1402 ops 37 registers

2.6 is 4% slower than 2.1.

New vs old:
-v 1 =~ 7% faster 2.6, 3% slower 2.1, 2.6 is 21% away from minimum target of 681 ops
-v 2 =~ 3% slower 2.6, 3% slower 2.1, 2.6 is 8% away from minimum target of 1362 ops

After looking at the ops generated in the .isa output, I'm clearly starting to force 2.6 to act more like the 2.1 results, so I seem to be getting somewhere. The hard part is going to be ordering instructions to properly saturate ALU clauses on vliw5 and GCN both.
full member
Activity: 131
Merit: 100
February 07, 2012, 03:46:03 PM
Which flags do I have to use to get the best out of my 7970?

ATM -v 1, but who knows once I finish this more optimized kernel.

Any faster so far?
legendary
Activity: 1022
Merit: 1000
Freelance videographer
February 07, 2012, 01:33:50 PM
Is there a windows GUI version of Diablo miner? Like GUIMiner but without the constant connection dropouts.Can diablo be launched from GUIMiner?
legendary
Activity: 1162
Merit: 1000
DiabloMiner author
February 07, 2012, 04:44:11 AM
Which flags do I have to use to get the best out of my 7970?

ATM -v 1, but who knows once I finish this more optimized kernel.
hero member
Activity: 1162
Merit: 500
February 07, 2012, 03:24:11 AM
Which flags do I have to use to get the best out of my 7970?
rjk
sr. member
Activity: 448
Merit: 250
1ngldh
February 03, 2012, 10:38:45 PM
7970 tweaks?

I've been on a 3 "day" almost interrupted hackfest trying to make SDK 2.6 suck less dick.

My brain is starting to melt after 24-hours-straight sessions.

Good news is: I seem to be winning, and I think I'm almost done.

Bad news is: the kernel is almost unrecognizable, and its known to the state of California to cause cancer in rats.

More bad news: I have no clue what its going to do to GCN.
I can't wait.

See? Give a dev some specs, and he'll code for a day, but buy him a fucking awesome card and he will pour out his heart and soul into making a better miner. Kudos.
Pages:
Jump to: