That isn't problem of one digit. S5T0308 produce totally incompatible results! I don't know whay?
There are only one important difference between S5S0007 and S5T0308, the rounding of the last bits. I think you can find out that how can S5T0308 produces these results. It isn't too difficult...
That isn't problem of one digit. S5T0308 produce totally incompatible results! I don't know whay?
There are only one important difference between S5S0007 and S5T0308, the rounding of the last bits. I think you can find out that how can S5T0308 produces these results. It isn't too difficult...
Maybe for you it is easy, for me is incomprehensible how can rounding of the last bit produce so diferent results.
There is the first row of results from S5T0711:
229.499118078 0.010001 -0.843352 -1.7676e-009 32.6857
It is the same as afficial app, so it seem that S5T0711 produce outcome, which will be valid :)
My results vary a bit, don't know why though. The direction the results are going is pretty clear:
Standard: 29073
S5T0003: 34150 (valid)
S5T0709: 14650 (initial, would make around 50% savings compared to standard app!)
I will post further resuts of S5T0709 and see if that monster-speed-up keeps up.
Many thanks to aksof for his work and effort!!
Hey man what kind of hell computer you use?
Athlon
Some Intel with real 3GHz and more with S5T0709 over 30000 Sec.
My AMD 3500+ @ 2650MHz with S5T0709 under 20000 sec.
Using the same CPU than you. AMD 64 Athlon 3500+ but stock, since I have new memory and haven't had time to do some OC and check the stability. I think the 14650 seconds was a faster WU though. I have checked the other result of that WU and the guy using a P4 3Ghz also has some increase in times, without using optimized app. From 70K down to 45K. So I assume it's the WUs.
different WUs I've seen and had so far:
h1_0436.5_S5R1__80_S5R1a slower (~14500)
h1_0436.5_S5R1__76_S5R1a faster (~26700)
Edit: S5T0709 produces valid results with my machine
OK, grabbing s5to711... Will see how it goes. I'm crunching under Vista Ultimate 64, beta 2. Don't s'pose I could get anymore in a test environment then running a beta operating system and all :D
He would like that we use the 4.10 version number for patched files, so you should rename your files and built an app_info.xml file in the future. The next patches will search einstein_S5R1_4.10_windows_intelx86.exe in their directory.
The other thing is that the validator doesn't drop out those results that have only a small difference in the values, like as S4 validation. So, there is possible to built ( a bit ) faster codes!
The other thing is that the validator doesn't drop out those results that have only a small difference in the values, like as S4 validation. So, there is possible to built ( a bit ) faster codes!
That sounds very good!
Quote:
Okay next week wie arrive the 50%
Athlon
Other Question: How long does it take to crunch a WU without optimised standard App and without optimised Apps here?
Days - Weeks - Month?
He would like that we use the 4.10 version number for patched files, so you should rename your files and built an app_info.xml file in the future. The next patches will search einstein_S5R1_4.10_windows_intelx86.exe in their directory.
Could someone please write, how such an app_info.xml have to look? I don´t know much about such things... ;)
RE: RE: I think if the
)
There are only one important difference between S5S0007 and S5T0308, the rounding of the last bits. I think you can find out that how can S5T0308 produces these results. It isn't too difficult...
RE: RE: RE: I think if
)
Maybe for you it is easy, for me is incomprehensible how can rounding of the last bit produce so diferent results.
There is the first row of results from S5T0711:
229.499118078 0.010001 -0.843352 -1.7676e-009 32.6857
It is the same as afficial app, so it seem that S5T0711 produce outcome, which will be valid :)
RE: standard app / 3140,11
)
Okay next week wie arrive the 50%
Athlon
Stay tuned and keep crunching
RE: RE: My results vary a
)
Using the same CPU than you. AMD 64 Athlon 3500+ but stock, since I have new memory and haven't had time to do some OC and check the stability. I think the 14650 seconds was a faster WU though. I have checked the other result of that WU and the guy using a P4 3Ghz also has some increase in times, without using optimized app. From 70K down to 45K. So I assume it's the WUs.
different WUs I've seen and had so far:
h1_0436.5_S5R1__80_S5R1a slower (~14500)
h1_0436.5_S5R1__76_S5R1a faster (~26700)
Edit: S5T0709 produces valid results with my machine
OK, grabbing s5to711... Will
)
OK, grabbing s5to711... Will see how it goes. I'm crunching under Vista Ultimate 64, beta 2. Don't s'pose I could get anymore in a test environment then running a beta operating system and all :D
Bernd told me some
)
Bernd told me some things.
He would like that we use the 4.10 version number for patched files, so you should rename your files and built an app_info.xml file in the future. The next patches will search einstein_S5R1_4.10_windows_intelx86.exe in their directory.
The other thing is that the validator doesn't drop out those results that have only a small difference in the values, like as S4 validation. So, there is possible to built ( a bit ) faster codes!
RE: The other thing is that
)
That sounds very good!
Athlon
Other Question: How long does it take to crunch a WU without optimised standard App and without optimised Apps here?
Days - Weeks - Month?
Stay tuned and keep crunching
S5T0712.dat - eliminated
)
S5T0712.dat
- eliminated double jumps
- reduced amount of FPU macro ops
- removed double loads on general purpose registers
- better SSE register usage
- reduced memory and integer register usage
- optimized branch structure
- faster FPU comparisons
- SSE3 truncation
- some reordered instructions
- automatic SSE/SSE3 usage
- less FPU-memory-FPU operation
CPU: ALL
app: einstein_S5R1_4.10_windows_intelx86.exe
RE: Bernd told me some
)
Could someone please write, how such an app_info.xml have to look? I don´t know much about such things... ;)
I wrote one but it didn't
)
I wrote one but it didn't work. don't know why. I've done them before for seti and had no prob.