Hi all
Some words about seti_K78 for SETI.
I upgraded 4 comps to this verion - 2 Athlon 64 with SSE2/SSE3 support - it seems seti_K78 faster than Crunch3r's dedicated verions for SSE2/SSE3 - very cool :)
But another 2 comps are Athlon XP with only SSE support. I thought it's OK cause seti_K78 claimed to be optimisation of Crunch3r's SSE version and it has the same size indeed, but... both PC returned computation error immediately on all WUs in cache :(.
Installation procedure and BOINC version are the same on all 4 PCs. So maybe seti_K78 contains some instructions Athlon XP cant understand?
Or what? Please give me some ideas/support in this
P.S.
from stderr:
***UNHANDLED EXCEPTION****
Reason: Access Violation (0xc0000005) at address 0x00509404 read attempt to address 0x01F0FFC0
from stderr:
***UNHANDLED EXCEPTION****
Reason: Access Violation (0xc0000005) at address 0x00509404 read attempt to address 0x01F0FFC0
There are small cycle with two load instruction to prefetch datas into L1 cache. I did it long time ago, so I can't remember exactly.
I see that there is one free byte, so you can exchange one of them (the critical one in your case) with a prefetchnta (mmx) instruction.
Try to modify:
00109404: 8B 4C DE 80 83 EB 10 75 F3 89 EA 90 ->
00109404: 0F 18 44 DE 80 83 EB 10 75 F2 89 EA
00109404: 8B 4C DE 80 83 EB 10 75 F3 89 EA 90 ->
00109404: 0F 18 44 DE 80 83 EB 10 75 F2 89 EA
SETI_K78b
Well, I modified, but I'm not interested in SETI.
Hmm... will this run on a SSE-capable Pentium III-M?
[edit] Just tried, it definitely runs not on SSE-capable Pentium III-M:
***UNHANDLED EXCEPTION****
Reason: Access Violation (0xc0000005) at address 0x00509400 read attempt to address 0x021AFF80
Crunch3r's original SSE-Version is crunching away since it came out, though :|
after installing Optimised by akosf (S-39L)
my calculation speeds more than doubled
both on Celeron 1.70Ghz (almost tripled) and Pentium3 700Mhz
S39L runs really super well on Celerons with small L1 L2 cash !!!
small celerons' cash is a real bottle neck of these processors :/
I have a question though - wandering in results database i've found that many users still use slow original exe file . I just wonder would not it be possible to make BOINC update exe file depending on the user's CPU so that all users run fastest exe available matching their CPUs best ?
I just wonder would not it be possible to make BOINC update exe file depending on the user's CPU so that all users run fastest exe available matching their CPUs best ?
It's really impressing what can be achieved with an optimized client. I'm using a Pentium M running at 1.6 GHz. The execution time for one WU using the original client was round about 8 hours. This dropped to 2.5 hours using version A36 of the optimized application. Now with S39L the time is reduced down to 1.5 hours. I'm speechless !
Hi all Some words about
)
Hi all
Some words about seti_K78 for SETI.
I upgraded 4 comps to this verion - 2 Athlon 64 with SSE2/SSE3 support - it seems seti_K78 faster than Crunch3r's dedicated verions for SSE2/SSE3 - very cool :)
But another 2 comps are Athlon XP with only SSE support. I thought it's OK cause seti_K78 claimed to be optimisation of Crunch3r's SSE version and it has the same size indeed, but... both PC returned computation error immediately on all WUs in cache :(.
Installation procedure and BOINC version are the same on all 4 PCs. So maybe seti_K78 contains some instructions Athlon XP cant understand?
Or what? Please give me some ideas/support in this
P.S.
from stderr:
***UNHANDLED EXCEPTION****
Reason: Access Violation (0xc0000005) at address 0x00509404 read attempt to address 0x01F0FFC0
very interesting! I tested it
)
very interesting! I tested it on a XP2100+, which is SSE capable, but not SSE2 and had the same error (see Message 28811)
AndyK
Want to know your pending credit?
The biggest bug is sitting 10 inch in front of the screen.
RE: from
)
There are small cycle with two load instruction to prefetch datas into L1 cache. I did it long time ago, so I can't remember exactly.
I see that there is one free byte, so you can exchange one of them (the critical one in your case) with a prefetchnta (mmx) instruction.
Try to modify:
00109404: 8B 4C DE 80 83 EB 10 75 F3 89 EA 90 ->
00109404: 0F 18 44 DE 80 83 EB 10 75 F2 89 EA
RE: 00109404: 8B 4C DE 80
)
SETI_K78b
Well, I modified, but I'm not interested in SETI.
RE: RE: 00109404: 8B 4C
)
Hmm... will this run on a SSE-capable Pentium III-M?
[edit] Just tried, it definitely runs not on SSE-capable Pentium III-M:
***UNHANDLED EXCEPTION****
Reason: Access Violation (0xc0000005) at address 0x00509400 read attempt to address 0x021AFF80
Crunch3r's original SSE-Version is crunching away since it came out, though :|
Aloha, Uli
RE: ***UNHANDLED
)
Ok. Perhaps they are wrong. I deleted them.
after installing Optimised
)
after installing Optimised by akosf (S-39L)
my calculation speeds more than doubled
both on Celeron 1.70Ghz (almost tripled) and Pentium3 700Mhz
S39L runs really super well on Celerons with small L1 L2 cash !!!
small celerons' cash is a real bottle neck of these processors :/
I have a question though - wandering in results database i've found that many users still use slow original exe file . I just wonder would not it be possible to make BOINC update exe file depending on the user's CPU so that all users run fastest exe available matching their CPUs best ?
RE: I just wonder would not
)
Check out this link.
RE: ... S39L runs really
)
Funny, I've never noticed any cash packed with any of my CPUs :-{ Is this kind of like an instant rebate?
Perhaps I should have shook the package. :-)
microcraft
"The arc of history is long, but it bends toward justice" - MLK
It's really impressing what
)
It's really impressing what can be achieved with an optimized client. I'm using a Pentium M running at 1.6 GHz. The execution time for one WU using the original client was round about 8 hours. This dropped to 2.5 hours using version A36 of the optimized application. Now with S39L the time is reduced down to 1.5 hours. I'm speechless !