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. |
|
|
|
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. |
|
|
|
Well, I'm mining here too now. Good work on the concept.
cgminer version 2.2.3 has some enhancements specifically targeted at working optimally with p2pool so if you're on cgminer, I recommend upgrading.
DiabloMiner had p2pool support first :< |
|
|
|
(no image due to a Google image search of 'woman bitcoin' resulting in too many pictures of Bruce Wagner and that tattooed girl in Vegas) That scares the shit out of me. I am not going to GIS that. |
|
|
|
No no, read the orders, it says TAKE him alive, Lord Vader will have our heads if we fuck that up again Your sad devotion to that ancient Jedi religion has not helped you conjure up the absent BFL singles, or given you enough clairvoyance to find the Rigbox power specs... No no, you got it backwards. The Empire is about GPUs, its the silly Jedis obsessing over FPGAs. An elegant weapon for a more civilized age my ass. |
|
|
|
Really, they could be ANY chip with the identifications sanded off.
Or just 54 heatspreaders with no chips. cost about $26.  (Before the crusaders of the church of BFL impale me it is called a joke). Blasphemer! Burn him alive! No no, read the orders, it says TAKE him alive, Lord Vader will have our heads if we fuck that up again |
|
|
|
I love the fact that it's hashing rate is continuing to increase, and soon it'll be stickied in the forums (meaning it'll be in the Top 10.)
It is in the Top 10 now. I can confirm that I just put it there today. |
|
|
|
Indeed, your notion of "average" reeks more than cow dung.
No need to get personal. btw ... I didn't bring in my definition of average - I only quoted the output of p2pool ("Average time between blocks: 0.42 days"). And that's just not correct. - statistics for 2012 so far (Jan 1st ... now): 36 days / 45 blocks = 0.8 days per block (and not 0.42). - statistics for the last couple of days would be even worse. I suspect the average is since the dawn of p2pool. |
|
|
|
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. |
|
|
|
Or... a hang gliding go-kart! Hell yeah!
Oh, OH, with thin film solar covering the wings and fleetstar/navstar internet and a few BFL boards hashing away while you fly circles around RandyFolds! Very big circles. |
|
|
|
[1/31/12 4:52:59 PM] ERROR: Failed to queue kernel, error -5 -5 == CL_OUT_OF_RESOURCES == driver bug. |
|
|
|
My guess is that the number of possible valid transformations overwhelm the synthesis tools and they blow up either on memory usage or time.
So if I could get my hands on a 4x hex core server with 256GB of RAM, you FPGA guys would love me long time? They may even use lube. |
|
|
|
To the FPGA guys here: Why is it 'rolled' and not 'furled'? It seems way more appropriate.
Because its always been unrolling loops. Ears are unfurled, loops are unrolled. Actually - every time I see comments about unrolling the sha256 code I wonder how you would do it any way but unrolled. The only 'rolled' option I can think of is to make a very small part of the FPGA just be P() and use it 122 times (in 2 loops) yet that would be senseless since I'd imagine it would be a MUCH slower way to do it? ... The "dumb" way is to have one function (in the case of an FPGA, one circuit), and rotate the variables/registers in and out of the function. You have the code compiled/circuit implemented exactly once. This would actually be superior for FPGA _if_ they had enough registers, but they don't. This is extremely slow for basically any implementation, and it also screws over the fact we essentially have 5 or more parallel ops at any given time in the way Bitcoin can optimize* the first, oh, 250 ops (depending on how certain things are implemented, of course). * In OpenCL, due to all the shortcuts calculating stuff in the host, it starts out as two unrelated chains that eventually merge. The ability to pack VLIW5 here is pretty goddamned handy, makes optimization a much easier job. |
|
|
|
To the FPGA guys here: Why is it 'rolled' and not 'furled'? It seems way more appropriate.
Because its always been unrolling loops. Ears are unfurled, loops are unrolled. |
|
|
|
Its not one of those kinds of bugs. Java on Windows is being retarded. Delete the DiabloMiner dir and unzip it again fresh. Weird part is, not all Windows users are hitting it, and no one on any other OS is.
Do you suppose it's a cache updating error? Or a linking error? Either way, my old settings are producing less desktop interactivity now (-w 64 and -f 1000) any suggestions? Again I have an Nvidia 9600 GT. 1GB. Well, its a cache update like error of some sort.... but I'm not aware of Java only on Windows actually having such a cache. As for desktop interactivity, you're screwed. Nvidia hates you, yet they want your money. Such a conundrum. |
|
|
|
But they avoided that issue. Interesting info though. So what's this further unexplained delay?
*yawn* Reported. Tired of all the trolling. We can't even have a conversation in here without you butting in and being a dick. Don't you have a job, a hobby or something? Nice. Did you put "I'm a whiny little bitch" in the comments box? Go fuck yourself, fella. Feel free to report this post as well. Ahh, a fine afternoon in off-topic, indeed. |
|
|
|
What exactly does "hw error" indicate? If I have 1 hw error per 100 accepted blocks, does that indicate that I might be overclocking a little too much, or is some level of hw error normal?
Some level is normal, depends entirely on the card and the machine. |
|
|
|
Upgraded to newest binary (haven't upgraded in months) and here's what I get now. C:\Program Files (x86)\Bitcoin\DiabloMiner>cd "C:\Program Files (x86)\Bitcoin\DiabloMiner"
C:\Program Files (x86)\Bitcoin\DiabloMiner>java -cp target\libs\*;target\DiabloMiner-0.0.1-SNAPSHOT.jar -Djava.library.path=target\libs\natives\windows com.diablominer.DiabloMiner.DiabloMiner -u username -p password -o api2.bitcoin.cz -r 8332 -w 64 -f 1000
Picked up _JAVA_OPTIONS: -Xmx1024M -Xms1024M
[1/29/12 11:51:33 PM] Started
Exception in thread "main" java.lang.UnsatisfiedLinkError: org.lwjgl.opencl.CL.getFunctionAddress(Ljava/lang/String;)J
at org.lwjgl.opencl.CL.getFunctionAddress(Native Method)
at org.lwjgl.opencl.CLCapabilities.<clinit>(CLCapabilities.java:8)
at org.lwjgl.opencl.CL.create(CL.java:123)
at com.diablominer.DiabloMiner.DiabloMiner.execute(DiabloMiner.java:180)
at com.diablominer.DiabloMiner.DiabloMiner.main(DiabloMiner.java:94) It used to work just fine and now this. Interestingly my friend was trying to use the same new binary and he got the old com.diablominer.DiabloMiner.DiabloMiner error. We couldn't fix it at all. But using my old binaries worked for him (but he got about 11Mhash/s versus his old 200Mhash/s on another miner). EDIT: another config C:\Program Files (x86)\Bitcoin\DiabloMiner>cd "C:\Program Files (x86)\Bitcoin\DiabloMiner"
C:\Program Files (x86)\Bitcoin\DiabloMiner>DiabloMiner-Windows.exe --url http://djinfected.gpu:dji12406gpu@api2.bitcoin.
cz:8332/ -w 64 -f 1000
Picked up _JAVA_OPTIONS: -Xmx1024M -Xms1024M
[1/30/12 12:21:17 AM] Started
[1/30/12 12:21:17 AM] Connecting to: http://api2.bitcoin.cz:8332/
Exception in thread "main" java.lang.UnsatisfiedLinkError: org.lwjgl.opencl.CL.getFunctionAddress(Ljava/lang/String;)J
at org.lwjgl.opencl.CL.getFunctionAddress(Native Method)
at org.lwjgl.opencl.CLCapabilities.<clinit>(CLCapabilities.java:8)
at org.lwjgl.opencl.CL.create(CL.java:123)
at com.diablominer.DiabloMiner.DiabloMiner.execute(DiabloMiner.java:610)
at com.diablominer.DiabloMiner.DiabloMiner.main(DiabloMiner.java:142)
Exception in thread "DiabloMiner GetWorkAsync for api2.bitcoin.cz" java.lang.NoSuchMethodError: org.codehaus.jackson.Jso
nNode.asText()Ljava/lang/String;
at com.diablominer.DiabloMiner.DiabloMiner$NetworkState.doJSONRPC(DiabloMiner.java:1048)
at com.diablominer.DiabloMiner.DiabloMiner$NetworkState$GetWorkAsync.run(DiabloMiner.java:1118)
at java.lang.Thread.run(Unknown Source) EDIT 2: fwiw I'm on a 9600 GT and I'm up to date with the latest stable driver for it. Its not one of those kinds of bugs. Java on Windows is being retarded. Delete the DiabloMiner dir and unzip it again fresh. Weird part is, not all Windows users are hitting it, and no one on any other OS is. |
|
|
|
what flags should I use for a dedicated miner? I just switched to Diablo Miner from CGminer
-f 1, usual flags otherwise. |
|
|
|
|
Show Posts
