Ufa: is there a place to go (other than here in the forums) to keep up on your miner updates?
Thanks
Topic: Ufasoft Miner - Windows/Linux, x86/x64, SSE2/OpenCL, Open Source - page 50. (Read 631117 times)
March 09, 2011, 12:44:17 PM
Excited about the update!
Ufa: is there a place to go (other than here in the forums) to keep up on your miner updates?
Thanks
Ufa: is there a place to go (other than here in the forums) to keep up on your miner updates?
Thanks
March 09, 2011, 12:12:50 PM
0.4 is about 10% faster but affects my GPU mining (by 7Mh/s, from 274 to 267)
March 09, 2011, 12:06:30 PM
There appears to be a small memory leak somewhere in this program. I haven't looked at the code yet to try and find it.
Version 0.4 uploaded:- The Memory leak fixed.
- ATI/AMD GPU adapters now supported, but performance is not better than on other OpenCL miners
March 09, 2011, 11:52:48 AM
After a computer restart I'm getting 20Mh/s lol 0.75 more than before
March 09, 2011, 10:43:06 AM
Went from 3 MH/s to 10 on an AMD Phenom x4. But it only seems to work when I use 127.0.0.1. When I try to send it over a lan to a local server @192.168.0.5 it doesn't work. Also, why do I have connect to port 8332 on 127.0.0.1, doesn't Bitcoin use 8333?
Thanks for the great work!
Thanks for the great work!
March 08, 2011, 10:33:46 AM
I get 19.2~19.3Mh/s constant (while running GPU miner as well) with my i7 920 @ 4ghz. Unfortunately it also lowers my GPU mining by 10Mh/s (260 to 250), and My CPU consumes 200W while my GPU <175W
March 08, 2011, 10:21:40 AM
However, wanted to let people know that my ugly code is now in jgarzik's tree. You can verify it at:
https://github.com/jgarzik/cpuminer/tree/
Cool, just sent you the other 25 BTC. Thanks for your great work https://github.com/jgarzik/cpuminer/tree/
March 07, 2011, 02:04:55 PM
Yea, this code really just runs full bore in my i5. I've got a massive heatsink (Scythe Mugen 2) on mine, so hitting all 4 cores at once isn't that bad. Biggest issue is that sound of my electric meter going WHIRRRRRR over the fans.... So I could see not-so-good colling solutions not working (this code runs the CPU at 100%, and most thermal solutions seem to be designed for a lower % utilization case to save on weight/cost...).
However, wanted to let people know that my ugly code is now in jgarzik's tree. You can verify it at:
https://github.com/jgarzik/cpuminer/tree/
I'll probably be making some minor changes every so often, and I'll put in pull requests when features are ready. The biggest x86 updates I'm looking at are support for the Sandy Bridge hardware, porting the code to NEON (which has the same rough structure as the X86_64 code), and possibly porting the code to x86_32 if anyone has SSE2 and only a 32-bit machine.
However, I have a bunch of other projects I'm interested in I want to get to first...
However, wanted to let people know that my ugly code is now in jgarzik's tree. You can verify it at:
https://github.com/jgarzik/cpuminer/tree/
I'll probably be making some minor changes every so often, and I'll put in pull requests when features are ready. The biggest x86 updates I'm looking at are support for the Sandy Bridge hardware, porting the code to NEON (which has the same rough structure as the X86_64 code), and possibly porting the code to x86_32 if anyone has SSE2 and only a 32-bit machine.
However, I have a bunch of other projects I'm interested in I want to get to first...
March 04, 2011, 03:01:14 PM
Thanks, triple speed up here at intel core 2 T9550 2.66GHz 
Update: weird, Lenovo T500 laptop goes 90 degrees centigrade CPU and switches off, twice in a hour. Getting back to cpu miner, temperature stays at 80 degrees and it worked for weeks.
--
mtve
Update: weird, Lenovo T500 laptop goes 90 degrees centigrade CPU and switches off, twice in a hour. Getting back to cpu miner, temperature stays at 80 degrees and it worked for weeks.
--
mtve
Yeah, this program runs my CPU (T9300 2.50 GHz so similar to yours) far hotter than any other application I've used, so I restrict it to one core which keeps it at about 70 degrees, and give it ample breathing room around the ventilation ports.
March 04, 2011, 10:36:17 AM
Oh. Sandy Bridge...
...if there are willing testers, I could try to port the code to those chips. If I recall, the SB'es have 256 bit wide XMM registers, so we may be able to use those for an 8 way hash. However, don't have the hardware available to me.
...if there are willing testers, I could try to port the code to those chips. If I recall, the SB'es have 256 bit wide XMM registers, so we may be able to use those for an 8 way hash. However, don't have the hardware available to me.
Yup , intel AVX .
I`d be glad to try it out .
March 04, 2011, 10:28:45 AM
Oh. Sandy Bridge...
...if there are willing testers, I could try to port the code to those chips. If I recall, the SB'es have 256 bit wide XMM registers, so we may be able to use those for an 8 way hash. However, don't have the hardware available to me.
...if there are willing testers, I could try to port the code to those chips. If I recall, the SB'es have 256 bit wide XMM registers, so we may be able to use those for an 8 way hash. However, don't have the hardware available to me.
March 04, 2011, 09:56:59 AM
Well , thats a boost - up to 24Mh/s from 8Mh/s with a [email protected] .
March 04, 2011, 06:41:50 AM
Thanks, triple speed up here at intel core 2 T9550 2.66GHz
Update: weird, Lenovo T500 laptop goes 90 degrees centigrade CPU and switches off, twice in a hour. Getting back to cpu miner, temperature stays at 80 degrees and it worked for weeks.
--
mtve
Update: weird, Lenovo T500 laptop goes 90 degrees centigrade CPU and switches off, twice in a hour. Getting back to cpu miner, temperature stays at 80 degrees and it worked for weeks.
--
mtve
March 04, 2011, 06:38:21 AM
I'll give 50 BTC to the person that implements this code into jgarzik's miner. First 25 BTC when working code is released, the other 25 BTC when it's pushed into jgarzik's cpuminer git repository.
Please take a look at https://github.com/chromicant/cpuminer/tree/sse2
[...]
Please, if you like my work, donate at the address in my sig!
March 04, 2011, 02:33:59 AM
Interesting discussion about AMD and Intel differences. But it does not explain why the compiled 4way code is so much faster on AMD than on Intel. If you compare the best code for AMD and the best code for Intel, they are very close in term of MH/s per GHz.
March 04, 2011, 01:57:35 AM
Is there a way of using this software through a proxy such as Tor? I am behind an ISP NAT router that blocks port 8332.
March 03, 2011, 07:29:34 PM
Thanks.
OK, so here are two instruction-level benchmarks
of the ufasoft code, one on core i5, the other on
AMD phenom.
A super quick glance at it seems to indicate that bitwise
integer SSE ops (psrld and friends) are dirt cheap on Intel
chips and rather heavy on AMD
Download the profile here:
profile.tar.bz2
Thanks for the data! The information is enlightening.
It looks like you get a (serious) stall if you access memory then try to use the register on AMD, and the rotates seem to be killer...which is quite a shock. The memory loads impact Intel as well, but it seems to be not as much...
I can definitely clean up some of the memory loads, but something makes me think that there's not much I can do to help the AMD case. Anyone know why you see impacts on the instruction pipeline due to a ps(r/l)ld? Is there a better instruction for this?
March 03, 2011, 06:53:10 PM
A big thank to chromicant ! My hashing speed have done more than x2 !
Here are all the steps i followed one a debian (testing) system to obtain the compiled version :
Here is the compiled binary : http://dl.free.fr/tjbLyHclU
Hope it will be usefull :p
Here are all the steps i followed one a debian (testing) system to obtain the compiled version :
Code:
sudo apt-get install git automake1.7 libc6-dev libcurl4-openssl-dev
#YASM : http://pkgs.org/download/debian-sid/multimedia-main-amd64/yasm_1.1.0-0.0_amd64.deb.html
wget http://ftp.br.debian.org/debian-multimedia/pool/main/y/yasm/yasm_1.1.0-0.0_amd64.deb
sudo dpkg -i yasm_1.1.0-0.0_amd64.deb
git clone https://github.com/chromicant/cpuminer.git
cd cpuminer
git checkout remotes/origin/sse2
cd x86_64/
./build.sh
cd ..
./autogen.sh
./configure
make
Here is the compiled binary : http://dl.free.fr/tjbLyHclU
Hope it will be usefull :p
March 03, 2011, 05:11:41 PM
Thanks.
Someone sent me the output of Intel's compiler on the 4way code. One thing that is different between the sse2_64 and 4way code is that the SSE2 core loop is unrolled. Also, the sse2_64 code is just SHA-256, which means you have to call it twice to get the hash you want.
This may be leading to some overhead.
Someone sent me the output of Intel's compiler on the 4way code. One thing that is different between the sse2_64 and 4way code is that the SSE2 core loop is unrolled. Also, the sse2_64 code is just SHA-256, which means you have to call it twice to get the hash you want.
This may be leading to some overhead.
March 03, 2011, 05:01:33 PM
The chromicant code gives 1.18 Mhash/s per 1 GHz per physical core on Intel Core i5 when utilizing 4 threads (2 physical cores, 2 virtual) as opposed to 0.89 Mhash/s in 4way version. The difference is almost 100% for a single-threaded run but apparently the multithreading can catch up a bit on the 4way ineffciencies.
On K10 AMD, the new code is 0.93 Mhash/s per 1GHz per physical core compared to 1.13 Mhash/s for 4way version of jgarzik's cpuminer compiled with Intel Compiler (icc). It seems that the 4way code compiled with icc is fastest for K10 architecture.
On K10 AMD, the new code is 0.93 Mhash/s per 1GHz per physical core compared to 1.13 Mhash/s for 4way version of jgarzik's cpuminer compiled with Intel Compiler (icc). It seems that the 4way code compiled with icc is fastest for K10 architecture.
Jump to: