With Akosf being the Chuck Yeager of E@H its a good thing he is crunching away openly, i for one like to watch his apps pushing the envelope of optimization. And its not like he is getting those credits for free, i wouldnt think for a second that this stuff is easy or was done in a few hours time.
...and hopefully Bernd is feeling more and more comfortable with these optimizations, so that the rest of us will at some point be able to take advantage of them! (And the project too of course will benefit greatly, it is like getting a big "free" boost of computing power! ;)
With Akosf being the Chuck Yeager of E@H its a good thing he is crunching away openly, i for one like to watch his apps pushing the envelope of optimization. And its not like he is getting those credits for free, i wouldnt think for a second that this stuff is easy or was done in a few hours time.
...and hopefully Bernd is feeling more and more comfortable with these optimizations, so that the rest of us will at some point be able to take advantage of them! (And the project too of course will benefit greatly, it is like getting a big "free" boost of computing power! ;)
It's already happening, as the latest beta and "power user" apps contain stuff that was originally authored/co-authored by Akos (the SSE hotloop and linear sin/cos code).
Akos deserves all the credit and credits for his work, IMHO. It would be a waste if all the crunching would go into /dev/null .
Akos deserves all the credit and credits for his work, IMHO. It would be a waste if all the crunching would go into /dev/null .
Aside from the waste, it seems to me that validating a large number of diverse results against those returned by a large number of diverse quorum partners is a very important step in checking out these algorithms, both for correctness and for good behavior with the project's chosen checking schemes.
And if just one host is running, then one does not have the risk of false validations against another running the same ap.
Akos ran his Einstein credit total past mine a few weeks ago, with a RAC vastly over mine. I cheered to myself--so here is a cheer in public.
....It's already happening, as the latest beta and "power user" apps contain stuff that was originally authored/co-authored by Akos (the SSE hotloop and linear sin/cos code).
CU
Bikeman
Actually, as a "selfish" Windows user, I was hoping the Windows app would start to implement the SSE hotloop code, which to my understanding has not yet been implemented into the Windows application. ;)
....It's already happening, as the latest beta and "power user" apps contain stuff that was originally authored/co-authored by Akos (the SSE hotloop and linear sin/cos code).
CU
Bikeman
Actually, as a "selfish" Windows user, I was hoping the Windows app would start to implement the SSE hotloop code, which to my understanding has not yet been implemented into the Windows application. ;)
Have you tried the Windows "Power users" app? It does make a difference.
....It's already happening, as the latest beta and "power user" apps contain stuff that was originally authored/co-authored by Akos (the SSE hotloop and linear sin/cos code).
CU
Bikeman
Actually, as a "selfish" Windows user, I was hoping the Windows app would start to implement the SSE hotloop code, which to my understanding has not yet been implemented into the Windows application. ;)
I'd guess it's the next thing in the pipeline. Stay tuned ;-)
....It's already happening, as the latest beta and "power user" apps contain stuff that was originally authored/co-authored by Akos (the SSE hotloop and linear sin/cos code).
CU
Bikeman
Actually, as a "selfish" Windows user, I was hoping the Windows app would start to implement the SSE hotloop code, which to my understanding has not yet been implemented into the Windows application. ;)
Have you tried the Windows "Power users" app? It does make a difference.
The "power users" app is now the official app... It finally puts us "selfish Windows users" roughly on par with Linux users with the same hardware... (4.26 Windows is roughly the same speed as 4.20 Linux)...
Edit: I should clarify that is at least for AMD K8-class and higher, which is obviously all I am concerned with. There may be discrepancies still with Intel platforms or with other AMD platforms, but at least I'm happy with the performance vs. Linux on my system right now. I think Linux still has a little bit of an edge, but only perhaps 3-5% vs. the 15-20% when comparing 4.15 vs. 4.20...
...I'd guess it's the next thing in the pipeline. Stay tuned ;-)
CU
Bikeman
Let's hope you are a good guesser! ;D
I certainly will stay tuned, especially since I just purchased a new home computer with the E8500 dual-core 3.16 GHz processor, which is really cranking out the EAH work units now, but sure could do a lot more I imagine with SSE! ;)
RE: With Akosf being the
)
...and hopefully Bernd is feeling more and more comfortable with these optimizations, so that the rest of us will at some point be able to take advantage of them! (And the project too of course will benefit greatly, it is like getting a big "free" boost of computing power! ;)
RE: RE: With Akosf being
)
It's already happening, as the latest beta and "power user" apps contain stuff that was originally authored/co-authored by Akos (the SSE hotloop and linear sin/cos code).
Akos deserves all the credit and credits for his work, IMHO. It would be a waste if all the crunching would go into /dev/null .
CU
Bikeman
RE: Akos deserves all the
)
Aside from the waste, it seems to me that validating a large number of diverse results against those returned by a large number of diverse quorum partners is a very important step in checking out these algorithms, both for correctness and for good behavior with the project's chosen checking schemes.
And if just one host is running, then one does not have the risk of false validations against another running the same ap.
Akos ran his Einstein credit total past mine a few weeks ago, with a RAC vastly over mine. I cheered to myself--so here is a cheer in public.
RE: ....It's already
)
Actually, as a "selfish" Windows user, I was hoping the Windows app would start to implement the SSE hotloop code, which to my understanding has not yet been implemented into the Windows application. ;)
RE: RE: ....It's already
)
Have you tried the Windows "Power users" app? It does make a difference.
RE: RE: ....It's already
)
I'd guess it's the next thing in the pipeline. Stay tuned ;-)
CU
Bikeman
RE: RE: RE: ....It's
)
The "power users" app is now the official app... It finally puts us "selfish Windows users" roughly on par with Linux users with the same hardware... (4.26 Windows is roughly the same speed as 4.20 Linux)...
Edit: I should clarify that is at least for AMD K8-class and higher, which is obviously all I am concerned with. There may be discrepancies still with Intel platforms or with other AMD platforms, but at least I'm happy with the performance vs. Linux on my system right now. I think Linux still has a little bit of an edge, but only perhaps 3-5% vs. the 15-20% when comparing 4.15 vs. 4.20...
RE: ...I'd guess it's the
)
Let's hope you are a good guesser! ;D
I certainly will stay tuned, especially since I just purchased a new home computer with the E8500 dual-core 3.16 GHz processor, which is really cranking out the EAH work units now, but sure could do a lot more I imagine with SSE! ;)