Thanks for all your replies!
so if I summarize 0.31f is the best on all cards but the rx550?
you didn't report about rx560 but i've those cards and i reach 360 on cn heavy on f, more than previous. versions.
high priority is now:
Better autoconfig
fix no-gpu bug on win7 + betablockchain
fan speed and temperature
watchdog
Topic: [XMR] JCE Miner Cryptonight/forks, now with GPU! - page 77. (Read 90842 times)
Wow, great speed! What settings in miner and overclock settings for 580 card you're using?
{ "mode" : "GPU", "worksize" : 8, "alpha" : 64, "beta" : 16, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta":4, "index" : 0, "multi_hash":864 },
{ "mode" : "GPU", "worksize" : 8, "alpha" : 64, "beta" : 16, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta":4, "index" : 0, "multi_hash":864 },
This is the divine micron 1250/2200
What Vegas you have? My vegas (all references) have nominal hashrate same or betterthan srbminer, but effective hashrate is like 10-15% more, there is no even comparison. 6x vega 64 = 12700H/s, 7x vega 56 14500H/s. You say you have better hashrate with f version? I am still on C. I am 8000km away from home, I don´t know if I dare to try to update
0.31f is brilliant! got my 8 vega 56 to 15.06k h/s; slightly besting srbminer.
FWIW the max multi_hash setting that could be used is 1872. Anymore and i would get random thread failures after compilation
FWIW the max multi_hash setting that could be used is 1872. Anymore and i would get random thread failures after compilation
Wow, great speed! What settings in miner and overclock settings for 580 card you're using?
{ "mode" : "GPU", "worksize" : 8, "alpha" : 64, "beta" : 16, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta":4, "index" : 0, "multi_hash":864 },
{ "mode" : "GPU", "worksize" : 8, "alpha" : 64, "beta" : 16, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta":4, "index" : 0, "multi_hash":864 },
This is the divine micron 1250/2200
In version 31f the overall performance on my rig with 12 552rx fell to 200+ hashes (from one card 347-> 330) The settings were exactly the same, the trick I used, maybe this version does not work now.) 
I return to the royal 31e version of the miner
Developer, what drivers do you use for testing? I'm using 18.3.4.
PS: Just in case, I bring the parameters of the video card on which the rig is assembled.
Rx550 2gb Baffin 1150/1800 (1071/1500) 768sp/12cu (unlock 640/10), mem Elpida
On the test card 588 on the new version, on the contrary, growth is seen on the Heavy Algo from 1100 to 1120 hashes!
Wow, great speed! What settings in miner and overclock settings for 580 card you're using? I return to the royal 31e version of the miner
Developer, what drivers do you use for testing? I'm using 18.3.4.
PS: Just in case, I bring the parameters of the video card on which the rig is assembled.
Rx550 2gb Baffin 1150/1800 (1071/1500) 768sp/12cu (unlock 640/10), mem Elpida
On the test card 588 on the new version, on the contrary, growth is seen on the Heavy Algo from 1100 to 1120 hashes!
Just FYI, Latest 31f version has the same performance for me on RX580/580/550 mixed rig with 18.6.1 driver on windows 10
In version 31f the overall performance on my rig with 12 552rx fell to 200+ hashes (from one card 347-> 330) The settings were exactly the same, the trick I used, maybe this version does not work now.)
I return to the royal 31e version of the miner
Developer, what drivers do you use for testing? I'm using 18.3.4.
PS: Just in case, I bring the parameters of the video card on which the rig is assembled.
Rx550 2gb Baffin 1150/1800 (1071/1500) 768sp/12cu (unlock 640/10), mem Elpida
On the test card 588 on the new version, on the contrary, growth is seen on the Heavy Algo from 1100 to 1120 hashes!
I return to the royal 31e version of the miner
Developer, what drivers do you use for testing? I'm using 18.3.4.
PS: Just in case, I bring the parameters of the video card on which the rig is assembled.
Rx550 2gb Baffin 1150/1800 (1071/1500) 768sp/12cu (unlock 640/10), mem Elpida
On the test card 588 on the new version, on the contrary, growth is seen on the Heavy Algo from 1100 to 1120 hashes!
ok, at least that's a clear error telling it doesn't find your OpenCL
i'll need to turn one rig into a Win7 + betablockchain to reproduce, i cannot continue to fail at blind attemps every time
i'll need to turn one rig into a Win7 + betablockchain to reproduce, i cannot continue to fail at blind attemps every time
Hello,
Are they 2Gb cards ? It sounds that's some RX550 or 560 with those speed.
With 2Gb i've a stable hashrate with multi-hash (intensity) of 432, and 448 or 464 when no screen is plugged.
You've a rig of 13 cards, that's big under windows, maybe you reach some drivers limits somewhere. JCE is a pure OpenCL miner, i let OpenCL and the drivers allocate resources, so if your driver makes your hashrate drop, JCE tries nothing to prevent it. Safety first.
I'm building 0.31f right now, with:
edit:0.31f online
Are they 2Gb cards ? It sounds that's some RX550 or 560 with those speed.
With 2Gb i've a stable hashrate with multi-hash (intensity) of 432, and 448 or 464 when no screen is plugged.
You've a rig of 13 cards, that's big under windows, maybe you reach some drivers limits somewhere. JCE is a pure OpenCL miner, i let OpenCL and the drivers allocate resources, so if your driver makes your hashrate drop, JCE tries nothing to prevent it. Safety first.
I'm building 0.31f right now, with:
Code:
Linked against official AMD SDK 3.0
Invalid shares detected by CPU now logs the Thread and Lane (good for memory diagnostic)
Still new OpenCL code, it's good on all my cards (HD7000, RX550, 560) but blind-experimental on RX570, 580 and vega. Reads: perfs may be worse than 0.31c
edit:0.31f online
Unfortunately, it still won't work for my win 7-64, blockchain drivers, RX 470-8GB with integrated graphics on MB. But the output is different this time, now it says:
Code:
Detecting OpenCL-capable GPUs...
Calling clGetPlatformIDs
Returned CL_UNKNOWN_ERROR -1001
Found 0 OpenCL PlatformIDs
Version f litterally crashes all my RX470s lol
EDIT: stupid me was launching 2 instances of the miner...
EDIT: stupid me was launching 2 instances of the miner...
Autoconfig is still in early stage, at least it produces a config that is valid
i'll improve it for next version.
fine i made a code that is fast on all cards. version e was a regression.
any feedback about the no-GPU found bug?
Heavy : f is faster on RX560 and 550, i peak at 360 against 347 before. but i giveup optimizing on vega for now, i need to get a real vega. blind optims was just regressions on versions d and e.
i also stop changing my netcode, it's a good balance between Stale share aggresivity and rejects. JCE has more rejects than others, but a better pool hashrate. By design.
next big change : gpu fan and temperature. And watchdog.
i'll improve it for next version.
fine i made a code that is fast on all cards. version e was a regression.
any feedback about the no-GPU found bug?
Heavy : f is faster on RX560 and 550, i peak at 360 against 347 before. but i giveup optimizing on vega for now, i need to get a real vega. blind optims was just regressions on versions d and e.
i also stop changing my netcode, it's a good balance between Stale share aggresivity and rejects. JCE has more rejects than others, but a better pool hashrate. By design.
next big change : gpu fan and temperature. And watchdog.
Code:
Linked against official AMD SDK 3.0
Invalid shares detected by CPU now logs the Thread and Lane (good for memory diagnostic)
Still new OpenCL code, it's good on all my cards (HD7000, RX550, 560) but blind-experimental on RX570, 580 and vega. Reads: perfs may be worse than 0.31c
Seems to have same performance as 0.31C for Vega. Good job. The --auto is still not finding the best configuration unfortunately for Vega.
Hello,
Are they 2Gb cards ? It sounds that's some RX550 or 560 with those speed.
With 2Gb i've a stable hashrate with multi-hash (intensity) of 432, and 448 or 464 when no screen is plugged.
You've a rig of 13 cards, that's big under windows, maybe you reach some drivers limits somewhere. JCE is a pure OpenCL miner, i let OpenCL and the drivers allocate resources, so if your driver makes your hashrate drop, JCE tries nothing to prevent it. Safety first.
I'm building 0.31f right now, with:
edit:0.31f online
Heavy algo is still the same as before? Are they 2Gb cards ? It sounds that's some RX550 or 560 with those speed.
With 2Gb i've a stable hashrate with multi-hash (intensity) of 432, and 448 or 464 when no screen is plugged.
You've a rig of 13 cards, that's big under windows, maybe you reach some drivers limits somewhere. JCE is a pure OpenCL miner, i let OpenCL and the drivers allocate resources, so if your driver makes your hashrate drop, JCE tries nothing to prevent it. Safety first.
I'm building 0.31f right now, with:
Code:
Linked against official AMD SDK 3.0
Invalid shares detected by CPU now logs the Thread and Lane (good for memory diagnostic)
Still new OpenCL code, it's good on all my cards (HD7000, RX550, 560) but blind-experimental on RX570, 580 and vega. Reads: perfs may be worse than 0.31c
edit:0.31f online
Hello,
Are they 2Gb cards ? It sounds that's some RX550 or 560 with those speed.
With 2Gb i've a stable hashrate with multi-hash (intensity) of 432, and 448 or 464 when no screen is plugged.
You've a rig of 13 cards, that's big under windows, maybe you reach some drivers limits somewhere. JCE is a pure OpenCL miner, i let OpenCL and the drivers allocate resources, so if your driver makes your hashrate drop, JCE tries nothing to prevent it. Safety first.
I'm building 0.31f right now, with:
edit:0.31f online
Are they 2Gb cards ? It sounds that's some RX550 or 560 with those speed.
With 2Gb i've a stable hashrate with multi-hash (intensity) of 432, and 448 or 464 when no screen is plugged.
You've a rig of 13 cards, that's big under windows, maybe you reach some drivers limits somewhere. JCE is a pure OpenCL miner, i let OpenCL and the drivers allocate resources, so if your driver makes your hashrate drop, JCE tries nothing to prevent it. Safety first.
I'm building 0.31f right now, with:
Code:
Linked against official AMD SDK 3.0
Invalid shares detected by CPU now logs the Thread and Lane (good for memory diagnostic)
Still new OpenCL code, it's good on all my cards (HD7000, RX550, 560) but blind-experimental on RX570, 580 and vega. Reads: perfs may be worse than 0.31c
edit:0.31f online
Yes it is RX552. The same situation is evident in other miners of the latest versions, but only on lexa cards. I'll try the new version.
Thank you.
Your miner I really like, good speed on the pool and the main stability without falling into the blue screen
Hello,
Are they 2Gb cards ? It sounds that's some RX550 or 560 with those speed.
With 2Gb i've a stable hashrate with multi-hash (intensity) of 432, and 448 or 464 when no screen is plugged.
You've a rig of 13 cards, that's big under windows, maybe you reach some drivers limits somewhere. JCE is a pure OpenCL miner, i let OpenCL and the drivers allocate resources, so if your driver makes your hashrate drop, JCE tries nothing to prevent it. Safety first.
I'm building 0.31f right now, with:
edit:0.31f online
Are they 2Gb cards ? It sounds that's some RX550 or 560 with those speed.
With 2Gb i've a stable hashrate with multi-hash (intensity) of 432, and 448 or 464 when no screen is plugged.
You've a rig of 13 cards, that's big under windows, maybe you reach some drivers limits somewhere. JCE is a pure OpenCL miner, i let OpenCL and the drivers allocate resources, so if your driver makes your hashrate drop, JCE tries nothing to prevent it. Safety first.
I'm building 0.31f right now, with:
Code:
Linked against official AMD SDK 3.0
Invalid shares detected by CPU now logs the Thread and Lane (good for memory diagnostic)
Still new OpenCL code, it's good on all my cards (HD7000, RX550, 560) but blind-experimental on RX570, 580 and vega. Reads: perfs may be worse than 0.31c
edit:0.31f online
Good evening. Can anyone come across a situation, there is a rig for 13 cards. 10 cards on the buffin work stably at 500 h/s, and 3 cards on the Lexa give out 515 h/s but regularly drop to 440 h/s and run at this speed from 2 minutes to 15 minutes, then the speed is again restored to 515 h/s. After a while everything repeats itself. The intensity reduced, alpha and beta, too, the same result.
now everything works with such settings:
{ "mode" : "GPU", "worksize" : 8, "alpha" : 64, "beta" : 8, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta" : 4, "index" : 3, "multi_hash":432 },
{ "mode" : "GPU", "worksize" : 8, "alpha" : 64, "beta" : 8, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta" : 4, "index" : 3, "multi_hash":432 },
I tried with these, the same result
{ "mode" : "GPU", "worksize" : 8, "alpha" : 128, "beta" : 16, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta" : 4, "index" : 3, "multi_hash":432 },
{ "mode" : "GPU", "worksize" : 8, "alpha" : 128, "beta" : 16, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta" : 4, "index" : 3, "multi_hash":432 },
Drivers 18.5.1
the frequency does not decrease in these moments
now everything works with such settings:
{ "mode" : "GPU", "worksize" : 8, "alpha" : 64, "beta" : 8, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta" : 4, "index" : 3, "multi_hash":432 },
{ "mode" : "GPU", "worksize" : 8, "alpha" : 64, "beta" : 8, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta" : 4, "index" : 3, "multi_hash":432 },
I tried with these, the same result
{ "mode" : "GPU", "worksize" : 8, "alpha" : 128, "beta" : 16, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta" : 4, "index" : 3, "multi_hash":432 },
{ "mode" : "GPU", "worksize" : 8, "alpha" : 128, "beta" : 16, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta" : 4, "index" : 3, "multi_hash":432 },
Drivers 18.5.1
the frequency does not decrease in these moments
Thanks !
It's royal... on the small RX. On Vega it looks like it's slower than the 0.31c, and since i've no vega to test, i couldn't know. I'll split my code into a RX- and a Vega- version, the vega being closer to 0.31c code.
About OpenCL: maybe the topic is old but i've the exact same problem in the exact same context. I link JCE with a small OpenCL lib for mingw, not the official AMD SDK. I tought the OpenCL lib was just a link wrapper over system OpenCL.dll but obviously, not.
I'll make a 0.31f build over full AMD SDK and, I hope, a fix for Vega speed.
It's royal... on the small RX. On Vega it looks like it's slower than the 0.31c, and since i've no vega to test, i couldn't know. I'll split my code into a RX- and a Vega- version, the vega being closer to 0.31c code.
About OpenCL: maybe the topic is old but i've the exact same problem in the exact same context. I link JCE with a small OpenCL lib for mingw, not the official AMD SDK. I tought the OpenCL lib was just a link wrapper over system OpenCL.dll but obviously, not.
I'll make a 0.31f build over full AMD SDK and, I hope, a fix for Vega speed.
Developer! The last version of the miner is just royal, thank you! I look forward to supporting backup pools and temperature monitoring. This is the most useful, in my opinion, functions. Here is the result of the latest version on the bittube, rx552x12, although the hash is a bit unstable (perhaps because of the small amount of RAM for 12 cards).
http://prntscr.com/k71i2b
Thank you again for your work!
http://prntscr.com/k71i2b
Thank you again for your work!
good, at least you get an error message. I can investigate, thanks.
I'm reworking my openCL, i hope I can make a version that is fast on all platforms. I can reach 360 on my little rx560 on CN-Heavy, which is better than before, but maybe those optims are bad on Vega...
edit:
https://community.amd.com/thread/129433
Looks like OpenCL + Mingw (my compiler under Win64) don't work well together. Other miners use Visual Studio, so why they work fine.
Theme on link is 8 years old...I'm reworking my openCL, i hope I can make a version that is fast on all platforms. I can reach 360 on my little rx560 on CN-Heavy, which is better than before, but maybe those optims are bad on Vega...
edit:
https://community.amd.com/thread/129433
Looks like OpenCL + Mingw (my compiler under Win64) don't work well together. Other miners use Visual Studio, so why they work fine.
I think it's not a problem. XMRig AMD and Stak-XMR compiles with mingw64 and works fine. Problem maybe in AMD ADL wich need to compile...
Jump to: