Topic: Bitcoin x64 for Windows (Read 37427 times)
ok, posted the installer in its own topic, go ahead and grab it.
I'll grab the SVN tonight and upload it for consumption
Is the x64 version 0.3.8, if not, could we have a new build please?
Sweet, over 5000 khash/sec with the latest release! I believe that's a little more than double over the stock build.
these new builds are not Intel, they're standard VS builds because the SHA optimization is all assembler, there's nothing for the compiler to optimize so performance is identical.
Please expand on that for the non-programmers among us.In particular, will your tweak still be faster than the stock builds? If yes, as much faster?
yes, definitely faster
specifically: the x86 build has to perform a byte reverse before hashing, the x64 build has no such requirement, this saves a fair bit of time. the x64 build also calculates the hash using 64-bit instructions.
these new builds are not Intel, they're standard VS builds because the SHA optimization is all assembler, there's nothing for the compiler to optimize so performance is identical.
Please expand on that for the non-programmers among us.In particular, will your tweak still be faster than the stock builds? If yes, as much faster?
The stock 0.3.6 release seems much faster than 0.3.0 was, overall. With the Linux x64 build on Ubuntu 10.04 x64 machines with four cores, I'm getting over 4,000 Khash/s now, vs 2,200-2,400 Khash/s before.
Even the stock 0.3.6 x86 Windows build is doing over 5,000 Khash/s on a four-CPU VM that gave me up to 5,600 Khash/s with the latest x64 Intel tweak. Given that said tweak tended to crash every few hours, I'm tempted to stick with the stock x86 build (unless it turns out to be unstable).
What's the prognosis on the 0.3.6 x64 tweak -- 8,000 Khash/s for said four-CPU VM? And stability?
Thanks again, Olipro
Even the stock 0.3.6 x86 Windows build is doing over 5,000 Khash/s on a four-CPU VM that gave me up to 5,600 Khash/s with the latest x64 Intel tweak. Given that said tweak tended to crash every few hours, I'm tempted to stick with the stock x86 build (unless it turns out to be unstable).
What's the prognosis on the 0.3.6 x64 tweak -- 8,000 Khash/s for said four-CPU VM? And stability?
Thanks again, Olipro
these new builds are not Intel, they're standard VS builds because the SHA optimization is all assembler, there's nothing for the compiler to optimize so performance is identical.
The stock 0.3.6 release seems much faster than 0.3.0 was, overall. With the Linux x64 build on Ubuntu 10.04 x64 machines with four cores, I'm getting over 4,000 Khash/s now, vs 2,200-2,400 Khash/s before.
Even the stock 0.3.6 x86 Windows build is doing over 5,000 Khash/s on a four-CPU VM that gave me up to 5,600 Khash/s with the latest x64 Intel tweak. Given that said tweak tended to crash every few hours, I'm tempted to stick with the stock x86 build (unless it turns out to be unstable).
What's the prognosis on the 0.3.6 x64 tweak -- 8,000 Khash/s for said four-CPU VM? And stability?
Thanks again, Olipro
Even the stock 0.3.6 x86 Windows build is doing over 5,000 Khash/s on a four-CPU VM that gave me up to 5,600 Khash/s with the latest x64 Intel tweak. Given that said tweak tended to crash every few hours, I'm tempted to stick with the stock x86 build (unless it turns out to be unstable).
What's the prognosis on the 0.3.6 x64 tweak -- 8,000 Khash/s for said four-CPU VM? And stability?
Thanks again, Olipro
here's 0.3.6 offered as a full installer, includes standard GUI plus headless daemon
Get 0.3.6 x64 installer here
Get 0.3.6 x64 installer here
So now they're saying to wait for 0.3.6 and that important changes haven't been committed to SVN yet.
I think changes have been committed now.
So now they're saying to wait for 0.3.6 and that important changes haven't been committed to SVN yet.
Just saw the alert telling us to upgrade to 0.3.5, is the x64 build a build of 0.3.5? I've shut down my generation until I know it is.
well put it this way, I just checked if there have been any new SVN commits since the announcement, and there haven't been any, I don't think satoshi would put an update out before he had it in SVN - not to mention that the SVN I have comes with the improvements he mentioned, so I'm pretty sure this is OK, even though the version number doesn't reflect it.
Just saw the alert telling us to upgrade to 0.3.5, is the x64 build a build of 0.3.5? I've shut down my generation until I know it is.
for anyone that wants it, a headless build of Bitcoin
Code is identical to the GUI version except the GUI isn't linked in obviously.
Code is identical to the GUI version except the GUI isn't linked in obviously.
hitting nearly 2700 khash/s on my Core i7 720QM now. It might actually reach that high if I didn't have so much stuff running.
Well it's good for me, I can change images around as needed. Your optimizations have actually made a bunch of old Celeron machines I had able to do almost 500 khash/s now when they could barely squeak by 100 khash/s before. I need to send some more BTC to you so that it covers the *acceleration* tax now 
[payment sent]
[payment sent]done, grab it, everyone who sees this should redownload, I had to correct an optimization on the result hash reversal so that it correctly rotated the values after doing a 64 bit byteswap (yes, you don't have to byteswap the input but you still have to reverse the result)
Well this certainly blows away my Linux machines now. I hope these kind of optimizations make it into the Linux builds on day. for some reason the x64 asm is only included in MASM form, if someone fancies converting it, good luck to them I say.
I'll be interested to see how this goes stability wise, my bet is that it won't have the crash issues that the Intel builds did; my bet is that was a result of multi-threading being parallelized by the Intel Compiler when it really shouldn't be, race condition etc.
done, grab it, everyone who sees this should redownload, I had to correct an optimization on the result hash reversal so that it correctly rotated the values after doing a 64 bit byteswap (yes, you don't have to byteswap the input but you still have to reverse the result)
Well this certainly blows away my Linux machines now. I hope these kind of optimizations make it into the Linux builds on day.
done, grab it, everyone who sees this should redownload, I had to correct an optimization on the result hash reversal so that it correctly rotated the values after doing a 64 bit byteswap (yes, you don't have to byteswap the input but you still have to reverse the result)
and with that addition, I've broken over 5000khash/s for the first time.
I've updated the link above, so just redownload.
I've updated the link above, so just redownload.
New Link please
just click on the link he provided before.
it's been updated
Jump to: