0.3 almost ready -- please test the Mac version!

[satoshi]

Here's RC1 for linux for testing:
(link removed, see below)

It contains both 32-bit and 64-bit binaries.

Recent changes:

build-unix.txt:
- Added instructions for building wxBase, which is needed to compile bitcoind.
- The package libboost-dev doesn't install anything anymore, you need to get libboost-all-dev.
- Updated version numbers.

makefile.unix:
- The libboost libraries have removed the "-mt" from their filenames in 1.40.  If you're compiling with Boost 1.38 or lower, like on Ubuntu Karmic, you would need to change it back to boost_system-mt and boost_filesystem-mt.

[virtualcoin]

Nice. :-)

64-bit version works fine on Ubuntu 10.04, of course it's (much) faster than through WINE emulation (and because it's 64-bit!?)

4 cores: 2850 khash/s
3 cores: 2130 khash/s
2 cores: 1420 khash/s
1 core: 700 khash/s

WINE (32-bit):

4 cores - 2300 khash/s
3 cores - 1740 khash/s
2 cores - 1150 khash/s
1 core - 580 khash/s

Thank you - looks pretty nice, too.

http://i50.tinypic.com/2ih1h0p.jpg

edit: What's the difference between "Sent/Received" and "All transactions"?

[HostFat]

will it change something if you release a x64 bit version for Windows?

[Vasiliev]

Ubuntu 9.04 x86:

vlad@vlad:~/bitcoin/bin/32$ ./bitcoind
./bitcoind: /usr/lib/libstdc++.so.6: version `GLIBCXX_3.4.11' not found (required by ./bitcoind)

Anybody know what Bitcoin needs?

[satoshi]

I don't know.  Maybe someone with more Linux experience knows how to install the library it needs.

I built it on Ubuntu 10.04.  I hope that wasn't a mistake.  Maybe it should have been built on an older version for more backward compatibility.  Is this a problem on Linux, that if you build on the latest version, then it has trouble working on older versions?  Is there any way I can downgrade to an older version of GCC on 10.04?

The 64-bit version shouldn't be any faster than the 32-bit version, but it would be great if someone could do a side-by-side comparison of the two linux versions and check.  SHA-256 is a 32-bit algorithm and nothing in BitcoinMiner uses 64-bit at all.

We don't need to bother with a 64-bit version for Windows.  32-bit programs work on all versions of Windows.  It's not like Linux where the 64-bit OS wants 64-bit programs.

I'm also curious if it's a little faster on linux than windows.

Do you think I should make the directories:
/bin32/
/bin64/
instead of
/bin/32/
/bin/64/

[HostFat]

Windows 7 x64
4 GB RAM
Intel i7 860

4 cores ( 8 virtuals ): 2200 khash/s

It's lower than Ubuntu :|

[theymos]

Linux on an overclocked Pentium D:
- 1 core: 565 khash/s, 1.01 kW h/day
- 2 cores: 1100 khash/s, 1.78 kW h/day

The new version seems to give me an improvement of 20-30 khash/s (with two cores) over SVN 80 with laszlo's patch.

Does anyone know what I have to change in the makefile so that I don't have to add LD_LIBRARY_PATH pointing to the dependency libraries when executing BitCoin? I already have -I and -L pointing to the single, correct location. (This isn't caused by the new version; I've always had this problem.)

[virtualcoin]

The 64-bit version shouldn't be any faster than the 32-bit version, but it would be great if someone could do a side-by-side comparison of the two linux versions and check.  SHA-256 is a 32-bit algorithm and nothing in BitcoinMiner uses 64-bit at all.

But look here:

32-bit Linux version on Ubuntu 10.04

4 cores: 2500 khash/s
3 cores: 1900 khash/s
2 cores: 1260 khash/s
1 core: 630 khash/s

64-bit Linux version on Ubuntu 10.04 (new measure)

4 cores: 2880 khash/s
3 cores: 2150 khash/s
2 cores: 1450 khash/s
1 core: 720 khash/s

(Though I never created one coin, yet - but I don't let bitcoin run on 4 cores the whole day, not even on 1 core...)

@Joozero - The Intel i7 860 is @ 2.8 Ghz, isn't it? My Phenom II is @ 3 Ghz, that's another important fact I think. I will try bitcoin on Windows 7 at the same machine later.

/edit

32-bit Win version on Windows 7 64-bit

4 cores: 2310 khash/s
3 cores: 1740 khash/s
2 cores: 1160 khash/s
1 core: 580 khash/s

[satoshi]

Thanks virtualcoin, that's a perfect comparison.

The 8% speedup from 32-bit Windows (2310k) to 32-bit Linux (2500k) is probably from the newer version of GCC on Linux (4.4.3 vs 3.4.5).

The 15% speedup from 32-bit to 64-bit Linux is more of a mystery.  The code is completely 32-bit.

Hmm, I think the 8 extra registers added by x86-64 must be what's helping.  That would make a significant difference to SHA if it could hold most of the 16 state variables in registers.

[satoshi]

Lets try using Laszlo's irc.lfnet.org instead of freenode.  Here's RC2, that's the only change in it:

(see below for download links)

[NewLibertyStandard]

It looks good! The number of hashes per second intermittently disappears for a few seconds at a time. I think it might be triggered by clicking on the file menu multiple times. I'm using the Linux 64-bit build.

[satoshi]

The first panel of the status bar is shared with the help description of menu items as you hover over them.  Since all our menu item descriptions are blank, it replaces it with blank when you're hovering in a menu.

[satoshi]

Changed the version number to 1.3 and removed "Beta".

(links removed, see below)

Uses irc.lfnet.org.

[laszlo]

I'm not sure why but I noticed that as well - my Win32 build was a lot slower than the linux/mac builds I did.  I used MinGW and all that but maybe I made a mistake somewhere..  I probably used a different version of Berkeley DB than Satoshi too so that's why the logs weren't binary compatible.

[satoshi]

MinGW still only has good old stable 3.4.5.  There's not much reason for them to update it.

When I looked at the 3.4.5 compiled SHA disassembly, I couldn't see any room for improvement at all.  I can't imagine how 8% more could be squeezed out of it.  Is it possible Windows could have 8% more overhead?  Not making system calls or anything, just plain busy computational code, could task switching and other housekeeping operations take away that much?

[dkaparis]

MinGW still only has good old stable 3.4.5.  There's not much reason for them to update it.

When I looked at the 3.4.5 compiled SHA disassembly, I couldn't see any room for improvement at all.  I can't imagine how 8% more could be squeezed out of it.  Is it possible Windows could have 8% more overhead?  Not making system calls or anything, just plain busy computational code, could task switching and other housekeeping operations take away that much?

Maybe best to look at the 4.4.3 compiled disassembly as well and see if there are any differences.

On a related note, is the thing compilable by Visual C++? I'm inclined to give it a try when I get around to it.

Dimitri

[satoshi]

On a related note, is the thing compilable by Visual C++? I'm inclined to give it a try when I get around to it.

It is, but generating is more than twice as slow.

[satoshi]

(reverted to rc2)

Links removed, 0.3 is now released, so go to http://www.bitcoin.org to download it.

[laszlo]

Mac OS version - requires Intel processor and 10.5 (10.4 not supported)

http://heliacal.net/~solar/bitcoin/builds/MacOSX-Intel-0.3/bitcoin-0.3.0.zip

[satoshi]

Laszlo's build is going to be our first Mac release so please test it!