If you check the Read Only thread you will see that Akos has released S5S0003.dat as a "stable" patch. Note the change from "T" to "S" which I assume means "Test" is now regarded as "Stable". It's not immediately clear if the "S" version is identical to the earlier "T" version or whether there were some additional tweaks. My guess is that it should be identical so if you have the "T" version then just keep running it until Akos releases S5T0005.dat :).
Oh, yes.
T -> Test version
S -> Stable version
S5S0003 has only one difference from S5T0003. It write the 'STABLE' to the stderr not the 'TEST' label.
The next test version will be S5T0301.dat. I think it's more logical...
Try to find it why... :-)
Any chance for optimizations for other instruction sets, e. g. 3Dnow?
My old Athlon isn't capable of SSE but was very fast with the latest D41.15 ;)
The optimization of the current app is very difficult, because the validator works with a very low tolerance. The 3DNow! hot loop give a bit different results so they all would be invalid. I would like to do the fastest app at moment. I think it means an about 10-15% faster application than the official, because the official app uses optimized routines too.
S5S0003 has only one difference from S5T0003. It write the 'STABLE' to the stderr not the 'TEST' label.
The next test version will be S5T0301.dat. I think it's more logical...
Try to find it why... :-)
@akosf
Soon be returning another S5T0000 unit and yet another a few hours after. if they also come back as sucessfull and valid would you like me to contunue with the S5T0000 for a while longer or update to your 'stable' S5S0003. Or one the others? i know you want to be thorough with them all so i dont mind any.
Side note, seeing strange (BUT GOOD) behavior on HT machine with S5T0001. Basically looks to be as quick or slightly quicker doing two units than it was doing one on its own (no other projects) with standard app. Know more when they're finished in few hours
A few minutes ago I switched to S5T0301 and in about one hour I will get 2 results with mixed S5T003/S5T0301.
Progress is obviously faster with S5T0301, while S5T003 was about 2% faster, but there is no compareble WU. So the 2% are the difference to to one of the fist S5-test WUs, which got a much lower credit by the server.
Could you please have a look at this post and the problem it points to and let us know if the idea of restricting the new releases to a smaller group of testers is workable for your purposes. In this case your latest release is restricted to SSE3 capable machines so hopefully that will be a smaller potential audience anyway.
RE: If you check the Read
)
Oh, yes.
T -> Test version
S -> Stable version
S5S0003 has only one difference from S5T0003. It write the 'STABLE' to the stderr not the 'TEST' label.
The next test version will be S5T0301.dat. I think it's more logical...
Try to find it why... :-)
Any chance for optimizations
)
Any chance for optimizations for other instruction sets, e. g. 3Dnow?
My old Athlon isn't capable of SSE but was very fast with the latest D41.15 ;)
Aloha, Uli
RE: Any chance for
)
The optimization of the current app is very difficult, because the validator works with a very low tolerance. The 3DNow! hot loop give a bit different results so they all would be invalid. I would like to do the fastest app at moment. I think it means an about 10-15% faster application than the official, because the official app uses optimized routines too.
RE: Oh, yes. T -> Test
)
@akosf
Soon be returning another S5T0000 unit and yet another a few hours after. if they also come back as sucessfull and valid would you like me to contunue with the S5T0000 for a while longer or update to your 'stable' S5S0003. Or one the others? i know you want to be thorough with them all so i dont mind any.
Side note, seeing strange (BUT GOOD) behavior on HT machine with S5T0001. Basically looks to be as quick or slightly quicker doing two units than it was doing one on its own (no other projects) with standard app. Know more when they're finished in few hours
S5T0301 - eliminated
)
S5T0301
- eliminated double jumps
- reduced amount of FPU macro ops
- removed double loads on general purpose registers
- SSE3 truncation
So, use this patch only on SSE3 CPUs!
I think it will give the biggest speed-up ( +5% or more ).
Pentium4 2,66GHz S5S0003 the
)
Pentium4 2,66GHz
S5S0003 the first hybrid result is ready, 50% with S5T0004 the rest with S5S0003
hybrid
other results will follow soon
britta
@Akos When I do patching
)
@Akos
When I do patching in cmd.exe, the patcher.com prompts "Invalid keyboard code specified" in the first line, is it OK?
Best regards,
Yin Gang
Welcome To Team China!
I got 2 valid results with
)
I got 2 valid results with long WUs and S5T003.dat so far.
34581346
34581722
A few minutes ago I switched to S5T0301 and in about one hour I will get 2 results with mixed S5T003/S5T0301.
Progress is obviously faster with S5T0301, while S5T003 was about 2% faster, but there is no compareble WU. So the 2% are the difference to to one of the fist S5-test WUs, which got a much lower credit by the server.
cu,
Michael
first to return so can't tell
)
first to return so can't tell if valid yet. Had no probs.
Done with S5T0000
http://einsteinathome.org/task/34776031
@ Akos, Could you please
)
@ Akos,
Could you please have a look at this post and the problem it points to and let us know if the idea of restricting the new releases to a smaller group of testers is workable for your purposes. In this case your latest release is restricted to SSE3 capable machines so hopefully that will be a smaller potential audience anyway.
Thanks, and thanks very much for your work.
Cheers,
Gary.