Only akosf to post here, please, if he has any other links to download the new gear....

For any discussion, please start a new thread or add to an existing one. :-) :-)

The following was provided by akosf. I have tested by successfully downloading from the link given. It points to a zip file ( 1,044,480 bytes packed down to 412,787 bytes ) containing his optimised version of 'albert_4.37_windows_intel86.exe'.

C37 - optimised x86 compatible windows executable

Install:

1, download file
2, stop BOINC
3, unzip file to ../BOINC/projects/einstein.phys.uwm.edu/ directory
4, restart BOINC

S38 - optimised x86-SSE compatible executable

S39 - optimised x86-SSE compatible executable (for windows)

S39L - optimised SSE compatible windows executable

Change: S39L needs less L1 cache (~11kB) than S39 (~33kB) for its important datas.

C40 - optimised x86 compatible windows executable

Changes: sin/cos interpolator, AGI optimisations
Speedup: ~10% faster than C37

D40 - 3DNow! optimised windows executable

Comment: 3DNow! supported on these cpus:
K6-2,K6-III,Athlon,Duron,AthlonXP,Sempron,Athlon64,C3,Samuel-II,Ezra,Nehemiah,Winchip2

S40 - SSE optimised windows executable

Changes: better prefetching, AGI optimisations

S40 is not perfect! Don't use it!
Sometimes gives back an "access violation" on one of my Durons.
(I will check the code and the PC.)

S40.03 - SSE optimised windows executable

Comment: S40 was good, just one of my Durons is in poor health.

S40.04 - SSE optimised windows executable

Comment: bugfixed version of S40.03 (zero credit problem)

S40.12 - SSE optimised windows executable

- double size sin/cos look-up table ( 2 more valuable bits but worse speed )
- 4 cache-lines are freed up ;-)

C41.00 - 386 compatible windows executable

- some tricks from S40.xx
- 2kB size sin/cos look-up table

C41.01 - 386 compatible windows executable

- better look-up table handling

D41.12 - 3DNow! compatible windows executable

- 3DNow! based trigonometry (2kB look-up table)
- reorganized address generation
- newton-raphson iteration
- mathematical reductions

D41.13 - 3DNow! compatible windows executable

- increased accuracy
- FPU based trigonometry with 4kB data
- common denominator for 4 pairs

D41.13 is refreshed (status bit bug is fixed).

Why did you give negative markings for these messages?

S41.06 - SSE compatible windows executable

- FPU based trigonomety with 4kB data
- common denominator for 4 pairs
- better FPU/SSE overlapping

U41.01 - SSE3 compatible windows executable

- SSE3 truncation

Hello! Don't use U41.01 because it doesn't work well. Sorry...

edit: I will look after and correct it.

C41.02 - 386 compatible windows executable (386,486,Pentium,Pentium2,K5,K6,...)
D41.14 - 3DNow! compatible windows executable (K6-2,K6-3,Spitfire,Thunderbird,Cyrix3,Samuel-II,Ezra,...)
S41.07 - SSE compatible windows executable (Pentium-III,Pentium4,AthlonXP,Duron(1GHz+),...)
U41.04 - SSE3 compatible windows executable (AMD: from E3 stepping, Intel: Prescott? and newer)

- special rounding method (~2% speedup)

comments:
1,use at your own risk
2,use the observation threads for bug report and for your questions

C41.03
D41.15
S41.08
U41.05

- fixed memory prefetching fault

Now already distributed, official applications are optimized. Don't worry about some other. Official one are the right one.