Topic: [XMR] JCE Miner Cryptonight/forks, now with GPU! - page 118. (Read 90815 times)
When I run the default "start_as_admin" in one thread (-t 1), cryptonight I got small improvement 1-2h/s vesus xmrig (1 minute average h/s). AMD A8-5600K. Should I increase the thread? or avoid the cryptonight algo?
Can please somebody help me out to extract the best performance from my rig?
E-5 2660 v4 @2GHz x2
I can clearly see that there's a lot room for performance as the CPU tab on the Windows task manager only Top 60%
I ran the process twice
Monero
jce_cn_cpu_miner64.1.exe -o stratum+tcp://Pool -u Address -p x --elevate --variation 3 -t 27 --auto
jce_cn_cpu_miner64.1.exe -o stratum+tcp://Pool -u Address -p x --elevate --variation 3 -t 27 --auto
The Affinity jce_cn_cpu_miner64.1.exe run on the CPU 0 only and jce_cn_cpu_miner64.2.exe run on the CPU 1
Definitely I'm getting better performance but if possible would like to extract more.
Thank you
E-5 2660 v4 @2GHz x2
I can clearly see that there's a lot room for performance as the CPU tab on the Windows task manager only Top 60%
I ran the process twice
Monero
jce_cn_cpu_miner64.1.exe -o stratum+tcp://Pool -u Address -p x --elevate --variation 3 -t 27 --auto
jce_cn_cpu_miner64.1.exe -o stratum+tcp://Pool -u Address -p x --elevate --variation 3 -t 27 --auto
The Affinity jce_cn_cpu_miner64.1.exe run on the CPU 0 only and jce_cn_cpu_miner64.2.exe run on the CPU 1
Definitely I'm getting better performance but if possible would like to extract more.
Thank you
Monero
-t 7
amd fx 8350@4100MHZ 405h/s
amd fx 8320e@3200MHZ 343h/s
-t 7
amd fx 8350@4100MHZ 405h/s
amd fx 8320e@3200MHZ 343h/s
Ok, Intense to be added too.
I'm writing assembly to add double-hash feature, starting with Aes 64 bits.
I see a huge gain on 32 bits core2, never tested on i3. Is the configuration the same (number of threads, Huge pages...) ? there should be a gap of at least 25%, however the gap is smaller than on 64 bits.
I'm writing assembly to add double-hash feature, starting with Aes 64 bits.
I see a huge gain on 32 bits core2, never tested on i3. Is the configuration the same (number of threads, Huge pages...) ? there should be a gap of at least 25%, however the gap is smaller than on 64 bits.
Ok ultranote, in next release.
I found a way to make doublehash but in 64 bits only, in 32 bits i run out of registers very fast, i don't see how to do it well.
I'll give a try.
Please, add Intense coin (ITNS) support.I found a way to make doublehash but in 64 bits only, in 32 bits i run out of registers very fast, i don't see how to do it well.
I'll give a try.
P.S. I don't see any substantial difference in speed from XMRig on 32-bit nonAES CPU Core i3.
Ok ultranote, in next release.
I found a way to make doublehash but in 64 bits only, in 32 bits i run out of registers very fast, i don't see how to do it well.
I'll give a try.
I found a way to make doublehash but in 64 bits only, in 32 bits i run out of registers very fast, i don't see how to do it well.
I'll give a try.
Thanks man. I wasn't kidding about the speed, it's really all written in assembly, instruction after instruction. For every Cryptonight variation, for 32 and 64 bits, with or without AES, with or withour dual-mining... Dozens of assembly codes, hence the large binary size.
The hardest to beat was the Original Cryptonight + AES : it's so simple, a big loop that does AES and 64-bits multiply, that there were no room for improvement. I beat a compiler by a small 1%
The new, more complex variations let JCE have a real gain over C-based miners.
I'm making 0.18 right now, with support of TRTL version Cryptolight-v7. If you need support of more coins and/or pools, just ask
I know i'm a new member here, but no danger, the miner is not a virus not a trojan, and all hashrate values are real.
The hardest to beat was the Original Cryptonight + AES : it's so simple, a big loop that does AES and 64-bits multiply, that there were no room for improvement. I beat a compiler by a small 1%
The new, more complex variations let JCE have a real gain over C-based miners.I'm making 0.18 right now, with support of TRTL version Cryptolight-v7. If you need support of more coins and/or pools, just ask
I know i'm a new member here, but no danger, the miner is not a virus not a trojan, and all hashrate values are real.
Carry on the good work buddy, anyway.. would you include Ultranote (XUN) with the miner? Appreciate it..!!!
It doesn't work yet, my netcode needs direct connection to the pool. I could fix, but again, it won't be my top priority.
Hello,
I would need support for PROXY.
Classic WEB proxy or XMRIG PROXY.
I have about 100 PCs behind a proxy
Would that be possible?
i7-3370 - 168 h / s
i5-2430M - 65 h / s
I would need support for PROXY.
Classic WEB proxy or XMRIG PROXY.
I have about 100 PCs behind a proxy
Would that be possible?
i7-3370 - 168 h / s
i5-2430M - 65 h / s
I can do it, it's just about filling the cpu pipelines, but Linux version and asm improvements are higher priority.
I hope it won't flood the binaries, i already have about 50 versions of asm in the exe, for every combination of cpu, aes, fork...
PROCESSOR_ARCHITECTURE is supposed to be AMD64 on x64
Do you use a default 32 bits console ?
Ok in next version i reverse the logic and default to 64
I hope it won't flood the binaries, i already have about 50 versions of asm in the exe, for every combination of cpu, aes, fork...
PROCESSOR_ARCHITECTURE is supposed to be AMD64 on x64
Do you use a default 32 bits console ?
Ok in next version i reverse the logic and default to 64
Yeah, turbo on AMD... i got surprises with that when testing the ryzen assembly. I had a version that was faster than compiled C on one thread and slower on several threads, guess why... At least current asm is always faster 
Version 0.18 done:
https://github.com/jceminer/cn_cpu_miner/raw/master/jce_cn_cpu_miner.018.zip
Just add double threads, please. I can't run your miner on 1 thread faster other miners with 1 thread.Version 0.18 done:
https://github.com/jceminer/cn_cpu_miner/raw/master/jce_cn_cpu_miner.018.zip
Code:
* Parameter -a for compatibility (ignored)
* Better handling of TRTL
* Elevated console forwards the title
* Better autoconfig
* Better handling of TRTL
* Elevated console forwards the title
* Better autoconfig
And another 1 thing: part of your bat-file didn't work
set BITS=32
if x%PROCESSOR_ARCHITECTURE%==xAMD64 set BITS=64
On my system it always run 32-bit.
Yeah, turbo on AMD... i got surprises with that when testing the ryzen assembly. I had a version that was faster than compiled C on one thread and slower on several threads, guess why... At least current asm is always faster
Version 0.18 done:
https://github.com/jceminer/cn_cpu_miner/raw/master/jce_cn_cpu_miner.018.zip
Version 0.18 done:
https://github.com/jceminer/cn_cpu_miner/raw/master/jce_cn_cpu_miner.018.zip
Code:
* Parameter -a for compatibility (ignored)
* Better handling of TRTL
* Elevated console forwards the title
* Better autoconfig
* Better handling of TRTL
* Elevated console forwards the title
* Better autoconfig
From 60 to 100h, wow, i've a Excavator Amd cpu, do you think i could use it to test that double-hash mode ?
Maybe... I think that high hasgrate on 1 or 2 double threads linked with enabled Turbo mode on my CPU and no affinity. Therefore 1 double mining thread calculates on several cpu cores with turbo cores.And high hasrate results can be archived only if computer is idle. If I continue to work on it, hashrate is only 80-90.
From 60 to 100h, wow, i've a Excavator Amd cpu, do you think i could use it to test that double-hash mode ?
Since everything is assembly, making a double thread version would double the amount of asm to write, because double thread is not just one thread repeated two times.
Do you really have better perf on double thread than on two dedicated threads ?
Also try manual config with config file (parameter -c) and set the two mining threads on the same CPU, example :
Explanation here on Linux:
https://wiki.debian.org/Hugepages
Speed :
* Have or not AES (factor 2, factor 3 on CryptonightHeavy)
* Have a lot of Cache
* A lot of Cores
* Huge Pages enabled
* DRAM speed or amount is negligible
Yes. Double thread really better for me. 1 double thread uses 16-17% of my 8 core CPU. Gives 90-100 h/s.Do you really have better perf on double thread than on two dedicated threads ?
Also try manual config with config file (parameter -c) and set the two mining threads on the same CPU, example :
Code:
"cpu_threads_conf" :
[
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true },
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true }
]
[
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true },
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true }
]
Explanation here on Linux:
https://wiki.debian.org/Hugepages
Speed :
* Have or not AES (factor 2, factor 3 on CryptonightHeavy)
* Have a lot of Cache
* A lot of Cores
* Huge Pages enabled
* DRAM speed or amount is negligible
2 threads on your miner uses 30-31% of CPU and gives 120 h/s. 60 for each thread.
Variant with
Code:
"cpu_threads_conf" :
[
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true },
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true }
]
gives only 65-70 h/s.[
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true },
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true }
]
Single threads effective only when use many cores/threads. But if use only 1-2 core, double threads more effective.
Since everything is assembly, making a double thread version would double the amount of asm to write, because double thread is not just one thread repeated two times.
Do you really have better perf on double thread than on two dedicated threads ?
Also try manual config with config file (parameter -c) and set the two mining threads on the same CPU, example :
Explanation here on Linux:
https://wiki.debian.org/Hugepages
Speed :
* Have or not AES (factor 2, factor 3 on CryptonightHeavy)
* Have a lot of Cache
* A lot of Cores
* Huge Pages enabled
* DRAM speed or amount is negligible
Do you really have better perf on double thread than on two dedicated threads ?
Also try manual config with config file (parameter -c) and set the two mining threads on the same CPU, example :
Code:
"cpu_threads_conf" :
[
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true },
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true }
]
[
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true },
{ "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true }
]
Explanation here on Linux:
https://wiki.debian.org/Hugepages
Speed :
* Have or not AES (factor 2, factor 3 on CryptonightHeavy)
* Have a lot of Cache
* A lot of Cores
* Huge Pages enabled
* DRAM speed or amount is negligible
No need for real hardware access, but better your VM offers real access to Huge Page (that's an hardware feature) and AES if you have.
Linux version planned, all my code is standard C++ and GNU Assembly, compiling under Linux GCC and GNU-Asm should be easy. Just need time to do it. I make it more priority than the http server for stats
Linux version planned, all my code is standard C++ and GNU Assembly, compiling under Linux GCC and GNU-Asm should be easy. Just need time to do it. I make it more priority than the http server for stats
I'm not familiar with what "Huge Page" means. can you explain?
the specs of the processor is here: https://ark.intel.com/products/75277/Intel-Xeon-Processor-E5-2680-v2-25M-Cache-2_80-GHz
do you mean "extended page"?
what's the big determination of hashrates for CPU mining? CPU speed alone? CPU cache size? DRAM size? DRAM speed?
Please add double threads, when 1 thread do 2 hashes. For me important to use 1 or 2 double threads than use 2 or 4 single threads.
For me 1 double thread do about 90-100 h/s on my FX8320.
On your miner 1 thread do only 60 h/s.
For me 1 double thread do about 90-100 h/s on my FX8320.
On your miner 1 thread do only 60 h/s.
No need for real hardware access, but better your VM offers real access to Huge Page (that's an hardware feature) and AES if you have.
Linux version planned, all my code is standard C++ and GNU Assembly, compiling under Linux GCC and GNU-Asm should be easy. Just need time to do it. I make it more priority than the http server for stats
Linux version planned, all my code is standard C++ and GNU Assembly, compiling under Linux GCC and GNU-Asm should be easy. Just need time to do it. I make it more priority than the http server for stats
Any idea when a Linux version will be available? I could try it but I am Windows-free and have nowhere to run it. Your miner sounds promising and I would like to try it on an old non-AES comp, but again, it's Linux machine.
Jump to: