restless
Legendary
Offline
Activity: 1151
Merit: 1001
|
|
May 30, 2014, 11:12:30 AM |
|
I'll say it once and only once, from personal experience.
Thread concurrency for Hawaii chipsets (290/290X) should be 14080 not 8192.
You'll be surprised, but thread-concurrency (and lookup gap too) is affected only for scrypt kernels ) ocl.c sprintf(CompilerOptions, "-I \"%s\" -I \"%s\" -I \"%skernel\" -I \".\" -D LOOKUP_GAP=%d -D CONCURRENT_THREADS=%d -D WORKSIZE=%d", opt_kernel_path, sgminer_path, sgminer_path, cgpu->lookup_gap, (unsigned int)cgpu->thread_concurrency, (int)clState->wsize);
Neither x11, nor x13 are not using external definition of CONCURRENT_THREADS. Then why ship your version with predefined number of threads in conf files?! And changing -tc new kernels are generated - sometimes with different size - so number of threads DOES CHANGE something during opencl compilation. Can you explain this?!
|
|
|
|
pallas
Legendary
Offline
Activity: 2716
Merit: 1094
Black Belt Developer
|
|
May 30, 2014, 11:21:14 AM |
|
I'll say it once and only once, from personal experience.
Thread concurrency for Hawaii chipsets (290/290X) should be 14080 not 8192.
You'll be surprised, but thread-concurrency (and lookup gap too) is affected only for scrypt kernels ) ocl.c sprintf(CompilerOptions, "-I \"%s\" -I \"%s\" -I \"%skernel\" -I \".\" -D LOOKUP_GAP=%d -D CONCURRENT_THREADS=%d -D WORKSIZE=%d", opt_kernel_path, sgminer_path, sgminer_path, cgpu->lookup_gap, (unsigned int)cgpu->thread_concurrency, (int)clState->wsize);
Neither x11, nor x13 are not using external definition of CONCURRENT_THREADS. Then why ship your version with predefined number of threads in conf files?! And changing -tc new kernels are generated - sometimes with different size - so number of threads DOES CHANGE something during opencl compilation. Can you explain this?! no changes, you can use file names without tc inside pretty well. and without lookup gap and vectors for the same reason.
|
|
|
|
Elun
Member
Offline
Activity: 117
Merit: 10
|
|
May 30, 2014, 11:26:48 AM |
|
although watt meter says all cards working after 7-8 hurs miner freeze and pool hashrate is 0anybody having the same issue?
Yes. It runs well for a couple of hours and suddenly hashrate goes ballistic (1Gh/sec) but nothing gets sent to the pool anymore. No idea what causes it. EDIT: it would seem that the card is going sick. Can i see your miner log? my cards arent sick they continue after restarting miner and where is the miner log i can post Described here: https://bitcointalk.org/index.php?topic=623409.msg7006129#msg7006129But it's not acceptable for you, because it works fine in hours. Is there something interesting in miner windows, when it hangs? nothing happens miner window open but kh/s number and other numers freeze btw my cards are 290 Looks like multithread deadlock in code. But i'm not sure. Currently i don't understand why it happens. If someone have issue like this, please post your card model.
|
|
|
|
JPS2K5
|
|
May 30, 2014, 11:32:34 AM |
|
Oddness.... I had the b2 running for over 24 hours. I decided to fiddle with some settings. All I got was HW errors, reversed back to old settings, still HW errors. Deleted .bin files, still HW errors. Went for latest version, same thing..... Now all that seems to be running 'normally' is the Original release with I=15 I am still trying stuff with this... thing. The only thing I notice is that once, just once I ran setx GPU_MAX_ALLOC_PERCENT 100 and setx GPU_USE_SYNC_OBJECTS 1. I do not get the message anymore that they are not set. Not even after a full reboot. Is there any way to 'unset' this ?
|
|
|
|
t00tie
Newbie
Offline
Activity: 18
Merit: 0
|
|
May 30, 2014, 11:40:22 AM |
|
...
I am still trying stuff with this... thing. The only thing I notice is that once, just once I ran setx GPU_MAX_ALLOC_PERCENT 100 and setx GPU_USE_SYNC_OBJECTS 1. I do not get the message anymore that they are not set. Not even after a full reboot. Is there any way to 'unset' this ? http://support.microsoft.com/kb/195050
|
|
|
|
JPS2K5
|
|
May 30, 2014, 11:44:15 AM |
|
...
I am still trying stuff with this... thing. The only thing I notice is that once, just once I ran setx GPU_MAX_ALLOC_PERCENT 100 and setx GPU_USE_SYNC_OBJECTS 1. I do not get the message anymore that they are not set. Not even after a full reboot. Is there any way to 'unset' this ? http://support.microsoft.com/kb/195050ty. I did figure something out. If I take the x11modTahitiglg2tc8192w256l4 from the Original working miner and put it into the b2 folder, it works... So at least I'm up to 2.83 Mh/s again.
|
|
|
|
tr0ubles0me
Newbie
Offline
Activity: 51
Merit: 0
|
|
May 30, 2014, 11:47:07 AM |
|
Finally! Optimal performance with zero stability issues on 2 x MSI R9 280x (GAMING 3G) mining DRK: https://i.imgur.com/xaRoV83.jpg?1{ "pools" : [ { "name" : "Suchpool, EU", "url" : "stratum+tcp://stratum-eu.suchpool.pw:3335", "user" : "tr0ubles0me.h81pro", "pass" : "xxxx", "pool-priority" : "0" }, { "name" : "Suchpool, Stratum-1", "url" : "stratum+tcp://stratum1.suchpool.pw:3335", "user" : "tr0ubles0me.h81pro", "pass" : "xxxx", "pool-priority" : "1" }, { "name" : "Suchpool, Stratum-2", "url" : "stratum+tcp://stratum2.suchpool.pw:3335", "user" : "tr0ubles0me.h81pro", "pass" : "xxxx", "pool-priority" : "2" }, { "name" : "Darkpool", "url" : "stratum+tcp://drkpool.com:3333", "user" : "tr0ubles0me.h81pro", "pass" : "xxxx", "pool-priority" : "3" } ] , "intensity" : "18,18", "xintensity" : "0,0", "rawintensity" : "0,0",
"kernel" : "x11mod,x11mod",
"lookup-gap" : "2,2", "vectors" : "1,1", "worksize" : "128,128",
"thread-concurrency" : "8193,8193", "shaders" : "2048,2048",
"gpu-threads" : "4,4", "gpu-dyninterval" : "7", "gpu-platform" : "0",
"gpu-engine" : "1160-1160,1160-1160", "gpu-memclock" : "1500,1500", "gpu-memdiff" : "0,0", "gpu-fan" : "0-86,0-86", "gpu-powertune" : "20,20", "gpu-vddc" : "0.000,0.000",
"temp-cutoff" : "95,95", "temp-overheat" : "85,85", "temp-target" : "70,70",
"temp-hysteresis" : "4",
"auto-fan" : true, "auto-gpu" : true,
"api-allow" : "W:127.0.0.1", "api-listen" : true, "api-mcast-port" : "4028", "api-port" : "4028",
"expiry" : "1", "queue" : "0", "scan-time" : "1",
"failover-only" : true, "failover-switch-delay" : "40",
"tcp-keepalive" : "20",
"no-pool-disable" : true, "no-client-reconnect" : true,
"shares" : "0",
"log" : "4", "log-dateformat" : "0",
"worktime" : true,
"kernel-path" : "/usr/local/bin" }
|
|
|
|
emdje
|
|
May 30, 2014, 11:54:18 AM |
|
Wow this is some good improvement. My HD7950 hashes around >3500 kH/s edit: I see I am lying it is more like 3150 - 3500 kH/s
|
|
|
|
jjiimm_64
Legendary
Offline
Activity: 1876
Merit: 1000
|
|
May 30, 2014, 11:54:47 AM |
|
Finally! Optimal performance with zero stability issues on 2 x MSI R9 280x (GAMING 3G) mining DRK:
I think the key is intensity. I was dropping one gpu on each rig until I lowered to 18
|
1jimbitm6hAKTjKX4qurCNQubbnk2YsFw
|
|
|
Mister1k
|
|
May 30, 2014, 12:00:10 PM Last edit: May 30, 2014, 12:11:22 PM by Mister1k |
|
Chello all. Great thread,great people,good times. I just want to thank lasybear and Co for their incredible work. I am using x11 mod b2. I havent tried x13 at all. I Just started using Sgminer two days ago. Two machines. Not rigs. Sapphire reference Hd7950s (Stilt bios)win7 64 i5 Sapphire DualX Hd 7950 (Stilt bios) win7 32 Athlon X2(more memory on the card lol) Username Stilt lite coin forum. He tweaked the memory timings. Amazing work. If you can find your bios for your cards in that thread,you'll gain huge. When mining I use Stilts bios. 1500 memory all cards. Hd7950s Intensity 15 Thread concurrency 24000 always. Look up gap 2 Threads 2 Shaders 1792 Work 256( i tried 128 and lost 150 kh) I never use set max gpu or sync, ever!! Bat files are for after its tweaked . Use the comman prompt and run it manual till its tweaked. Sapphire reference card drivers 14.3 beta 64 Sapphire Dualx card drivers 13.12 32 I find x11 voltage sensitive. I recommend settings voltage to stock, The one reference I have is 1.112 undervolted stock. It needed a boost to 1.118. I would get one or two HW in three four hours. Gone now. The other stock voltage 1.1250 all day long zero HW. I always set voltage and clocks With Trixx , then close Trixx or Msi. If they run in the background you will lose Kh. Then I start Sgminer. I run them all 15 intensity. No sense mashing the gas till its tweaked. I will say I find it strange that the peeps with higher end cards 280/280x 290/290x 7970s etc are down compared to my 7950s. Even some that are Running good numbers, I feel good get more. The Stilt bios could be a huge factor in my opinion. Anyway I better get on with it. My first post an all. Lol. Heres the numbers. Sapphire references 1150 clock 1500 memory 2.895 Mh/s +- 5 khSapphire DualX 1100 clock 1500 memory 1.187 volts. 2.770 Mh/s +- 5 khHooooooo $@&#%* Rawww baby Absolutely awesome work peeps. Runs cool. 99.5 plus valid at the pool. In closing , I think I can get more. I wanna try this xintensity thing. Pg 20 Cheers folks. The x11mod is killer. Thankx again to all involved. I hope this helps some peeps.
|
|
|
|
BitSmacker
Member
Offline
Activity: 84
Merit: 10
|
|
May 30, 2014, 12:05:52 PM |
|
I'll say it once and only once, from personal experience.
Thread concurrency for Hawaii chipsets (290/290X) should be 14080 not 8192.
You'll be surprised, but thread-concurrency (and lookup gap too) is affected only for scrypt kernels ) ocl.c sprintf(CompilerOptions, "-I \"%s\" -I \"%s\" -I \"%skernel\" -I \".\" -D LOOKUP_GAP=%d -D CONCURRENT_THREADS=%d -D WORKSIZE=%d", opt_kernel_path, sgminer_path, sgminer_path, cgpu->lookup_gap, (unsigned int)cgpu->thread_concurrency, (int)clState->wsize);
Neither x11, nor x13 are not using external definition of CONCURRENT_THREADS. So tc of 1 vs 999999999 will show no difference with x11? What about shaders? Thanks!
|
|
|
|
Mister1k
|
|
May 30, 2014, 12:07:05 PM |
|
Wow this is some good improvement. My HD7950 hashes around >3500 kH/s edit: I see I am lying it is more like 3150 - 3500 kH/s Wow whats your intensity.. Hell i thought i had it good.
|
|
|
|
tr0ubles0me
Newbie
Offline
Activity: 51
Merit: 0
|
|
May 30, 2014, 12:08:55 PM |
|
Wow this is some good improvement. My HD7950 hashes around >3500 kH/s edit: I see I am lying it is more like 3150 - 3500 kH/s I had the same "issue" But then I run it with 4 gpu-threads which helped stabilize it a little. Then worksize=128 and tc=8193 pinned it to +/- 0.005 Mh/s all the time.
|
|
|
|
Elun
Member
Offline
Activity: 117
Merit: 10
|
|
May 30, 2014, 12:10:39 PM |
|
New release. No speed updates in this release. Github sources as always here: https://github.com/lasybear/sph-sgminer_x11modChanges:- Total uptime in first string
- Gpu sets idle, if all pools dead. Tried to unplug internet cable and gpu set to idle state, clocks dropped. Share your results about this feature.
- show-coindiff option from cgminer-3.7.2-kalroth added. Description: Option to show coin/network difficulty instead of hash value. --show-coindiff or "show-coindiff" : true, can now be used to show the coin/network difficulty in accepted or rejected messages. This is only relevant for coin switching pools and it is off by default.
Link for windows binaries: https://www.dropbox.com/s/vtj5z5e56jc8wh7/sgminer_x11x13mod_30_05_2014.zipThx platinum4 for his ideas, and lasybear for his free miner.
|
|
|
|
oktay50000
|
|
May 30, 2014, 12:19:12 PM |
|
New release. No speed updates in this release. Github sources as always here: https://github.com/lasybear/sph-sgminer_x11modChanges:- Total uptime in first string
- Gpu sets idle, if all pools dead. Tried to unplug internet cable and gpu set to idle state, clocks dropped. Share your results about this feature.
- show-coindiff option from cgminer-3.7.2-kalroth added. Description: Option to show coin/network difficulty instead of hash value. --show-coindiff or "show-coindiff" : true, can now be used to show the coin/network difficulty in accepted or rejected messages. This is only relevant for coin switching pools and it is off by default.
Link for windows binaries: https://www.dropbox.com/s/vtj5z5e56jc8wh7/sgminer_x11x13mod_30_05_2014.zipThx platinum4 for his ideas, and lasybear for his free miner. tnx elun what about bamt? reinstall?
|
BTC : bc1qqz9hvv806w2zs42mx4rn576whxmr202yxp00e9
feel free to buy me a bear
|
|
|
lasybear (OP)
Member
Offline
Activity: 113
Merit: 14
|
|
May 30, 2014, 12:21:38 PM |
|
I'll say it once and only once, from personal experience.
Thread concurrency for Hawaii chipsets (290/290X) should be 14080 not 8192.
You'll be surprised, but thread-concurrency (and lookup gap too) is affected only for scrypt kernels ) ocl.c sprintf(CompilerOptions, "-I \"%s\" -I \"%s\" -I \"%skernel\" -I \".\" -D LOOKUP_GAP=%d -D CONCURRENT_THREADS=%d -D WORKSIZE=%d", opt_kernel_path, sgminer_path, sgminer_path, cgpu->lookup_gap, (unsigned int)cgpu->thread_concurrency, (int)clState->wsize);
Neither x11, nor x13 are not using external definition of CONCURRENT_THREADS. So tc of 1 vs 999999999 will show no difference with x11? What about shaders? Thanks! only intensity/xintensity/rawintensity does matter. Threads = 1 << intensity or xintensity * shaders or rawintensity
|
|
|
|
Elun
Member
Offline
Activity: 117
Merit: 10
|
|
May 30, 2014, 12:22:06 PM |
|
New release. No speed updates in this release. Github sources as always here: https://github.com/lasybear/sph-sgminer_x11modChanges:- Total uptime in first string
- Gpu sets idle, if all pools dead. Tried to unplug internet cable and gpu set to idle state, clocks dropped. Share your results about this feature.
- show-coindiff option from cgminer-3.7.2-kalroth added. Description: Option to show coin/network difficulty instead of hash value. --show-coindiff or "show-coindiff" : true, can now be used to show the coin/network difficulty in accepted or rejected messages. This is only relevant for coin switching pools and it is off by default.
Link for windows binaries: https://www.dropbox.com/s/vtj5z5e56jc8wh7/sgminer_x11x13mod_30_05_2014.zipThx platinum4 for his ideas, and lasybear for his free miner. tnx elun what about bamt? reinstall? Sorry, can't help with bamt. Have only windows machine.
|
|
|
|
emdje
|
|
May 30, 2014, 12:22:39 PM |
|
Wow this is some good improvement. My HD7950 hashes around >3500 kH/s edit: I see I am lying it is more like 3150 - 3500 kH/s Wow whats your intensity.. Hell i thought i had it good. The config file: "intensity" : "20", "vector" : "1", "worksize" : "256", "kernel" : "x11mod", "thread-concurrency" : "24000", "lookup-gap" : "2", "gpu-threads" : "2" GPU clock: 1135 MHz MEM Clock: 1350 MHz But it fluctuated a lot. I will try the 4 gpu threads when I get home. Maybe that works. Powertune: 10%
|
|
|
|
bullus
|
|
May 30, 2014, 12:24:34 PM |
|
Chello all. Great thread,great people,good times. I just want to thank lasybear and Co for their incredible work. I am using x11 mod b2. I havent tried x13 at all. I Just started using Sgminer two days ago. Two machines. Not rigs. Sapphire reference Hd7950s (Stilt bios)win7 64 i5 Sapphire DualX Hd 7950 (Stilt bios) win7 32 Athlon X2(more memory on the card lol) Username Stilt lite coin forum. He tweaked the memory timings. Amazing work. If you can find your bios for your cards in that thread,you'll gain huge. When mining I use Stilts bios. 1500 memory all cards. Hd7950s Intensity 15 Thread concurrency 24000 always. Look up gap 2 Threads 2 Shaders 1792 Work 256( i tried 128 and lost 150 kh) I never use set max gpu or sync, ever!! Bat files are for after its tweaked . Use the comman prompt and run it manual till its tweaked. Sapphire reference card drivers 14.3 beta 64 Sapphire Dualx card drivers 13.12 32 I find x11 voltage sensitive. I recommend settings voltage to stock, The one reference I have is 1.112 undervolted stock. It needed a boost to 1.118. I would get one or two HW in three four hours. Gone now. The other stock voltage 1.1250 all day long zero HW. I always set voltage and clocks With Trixx , then close Trixx or Msi. If they run in the background you will lose Kh. Then I start Sgminer. I run them all 15 intensity. No sense mashing the gas till its tweaked. I will say I find it strange that the peeps with higher end cards 280/280x 290/290x 7970s etc are down compared to my 7950s. Even some that are Running good numbers, I feel good get more. The Stilt bios could be a huge factor in my opinion. Anyway I better get on with it. My first post an all. Lol. Heres the numbers. Sapphire references 1150 clock 1500 memory 2.895 Mh/s +- 5 khSapphire DualX 1100 clock 1500 memory 1.187 volts. 2.770 Mh/s +- 5 khHooooooo $@&#%* Rawww baby Absolutely awesome work peeps. Runs cool. 99.5 plus valid at the pool. In closing , I think I can get more. I wanna try this xintensity thing. Pg 20 Cheers folks. The x11mod is killer. Thankx again to all involved. I hope this helps some peeps. stilt bios? With reference bios my 7950's runs at 2940kh/s @ 1100/1500 x11modb2 . -I 18
|
btc 16LWhms487bzCxQWq5oeW8SMDPmU2rvQjf
|
|
|
lasybear (OP)
Member
Offline
Activity: 113
Merit: 14
|
|
May 30, 2014, 12:28:10 PM |
|
New release. No speed updates in this release. Github sources as always here: https://github.com/lasybear/sph-sgminer_x11modChanges:- Total uptime in first string
- Gpu sets idle, if all pools dead. Tried to unplug internet cable and gpu set to idle state, clocks dropped. Share your results about this feature.
- show-coindiff option from cgminer-3.7.2-kalroth added. Description: Option to show coin/network difficulty instead of hash value. --show-coindiff or "show-coindiff" : true, can now be used to show the coin/network difficulty in accepted or rejected messages. This is only relevant for coin switching pools and it is off by default.
Link for windows binaries: https://www.dropbox.com/s/vtj5z5e56jc8wh7/sgminer_x11x13mod_30_05_2014.zipThx platinum4 for his ideas, and lasybear for his free miner. tnx elun what about bamt? reinstall? go to miner folder restart miner
|
|
|
|
|