It should be possible for a WCG science program to determine whether the processor on which it is running supports the various SSE instructions that it can use, and automatically switch between using using the SSE hardware or emulating it in software. Then the available CPU features could be used without needing separate program versions.

However, several years back, I read here in the WCG forum that the WCG sciences did not use any of the SSE instructions because some CPUs might give slightly different results, and the WU validations might fail. (I can't find the post so can't provide a link).
BUT ... during the Beta test for HCMD2 v6.14, in the 2nd post in thread Re: Beta Test for Help Cure Muscular Dystrophy - Phase 2 , knreed said: "The beta test determined that the new version will require processors without SSE2 instructions to be in a different homogeneous class then processors with SSE2 instructions so we have modified HCMD2 to do that. There may be some AMD processors that we aren't breaking out correctly ... "
So, it seems that some projects now use some SSE instructions, but it is not without some problems.

If WCG scientists can use them, it will be interesting to see whether the AVX [1],[2] instructions in the forthcoming Intel "Sandy Bridge" and AMD "Bulldozer" CPUs can provide even more acceleration than the current SSE instructions.

[1]. From Intel to Ramp Up Sandy Bridge Aggressively on Desktop Market.
"The key feature of Sandy Bridge Intel AVX (Advanced Vector Extensions) which, when used by software programmers, will increase performance in floating point, media, and processor intensive software, according to the Intel. "
[2]: AMD to Disclose Details About Bulldozer Micro-Architecture in August.

Sekerob has now said so often in various threads that there won't be a "beyond basics" CPU support that I started to believe it. It is just the same as with the translations to other languages or 64 bit support (which brings sse with it). And - since we users are not paying for it - we probably just need to accept it. I can understand this, since it is our spare resources that are not used efficiently and not IBM's. And from IBM's marketing effect, the number of eyeballs does not change with increased efficiency. This would be very different if some processor maker would be part of the team. What I never understood is why all applications would need to be ported. The porting (aka recompilation for the low-hanging fruits) should start with those apps to which it matters most and the other apps should just continue shipping the 32bit / non-see / non-GPU variants, just like BOINC is offering it.

The three routes around that I see in:
* moving to other projects in BOINC that best support your hardware (Einstein@Home or Milkyway@Home come to mind) but I don't want to do that
* talking to the scientists behind your favorite project to start a WCG-independent project that is using BOINC directly and that OpenSources everything for platform-specific optimisations but I would not want to do that either, though I am tempted
* collecting money for IBM to pay half a position for the supporting to other platforms ... did I just suggest to collect money for IBM? Since we are so many, I think this should be the easiest and also help IBM to decide where to increase their efforts. Open a paypal account and count 50 Euros of mine in. With another 999 donors this should pay half an IBM position or a full position at some University.

Greetings

Steffen

i have disassembled pretty much all of the active projects on windows. they either use x87/x86. i have reason to believe that linux HCC uses SSE as my Q6600 on linux beats my core i7 windows machine by 2x:(

what also concerns me is that the project's wont support 64bit windows. why lose extra registers and up to date compilers?