Will be S40 in short time or you wait for the new afficial app and S40 will after?
Well I have lot of ideas to optimise a code, but the main optimisations are done. I would like to try out some new other tricks (I do it for fun), but it goes slowly because I have lot of work (current company) and lot of other task (own company). Probably I will release an S40 code but it will not faster than S39L. The S40 will have higher code density. It's good for power consumption but doesn't mean higher speed. It's funny, isn't it?
The source level optimisations would be much more lucky, that means more and easier possibilities.
Well this is weird - everyone else with 3DNow! seems to be going faster under D40 than S39L, but I appear to be going slower. I know the WU size varies, but effect seems to be about 10% slower under D40 than S39L.
Well this is weird - everyone else with 3DNow! seems to be going faster under D40 than S39L, but I appear to be going slower. I know the WU size varies, but effect seems to be about 10% slower under D40 than S39L.
My processor is an Athlon XP 2800+ (Barton core: 512k L2 cache @ 2.08GHz MMX(+) 3DNow!(+) SSE)
Should I switch back to S39L, or wait a few more WUs to see what happens?
Your results are not all from the same major datafile. The three slowest D-40 results are 1447.5, the next faster from 0784.0, and the fastest one so far from 1058.5.
I'm pretty sure I've seen appreciable offsets for compact sets from different major datafiles, and certainly seen drift and the occasional blip even in the same.
Folks confidently reported differences down in the few percent range are being optimistic.
I've noticed that the server is much less likely to keep giving me results form the same major datafile than it was just a few weeks ago. Back then it would go weeks at a time, while now all four of my machines have process results from more than one datafile in the last week.
Okay thanks. I'll wait for a good few more WUs and see what happens.
If the difference had been less than 5% I wouldn't have queried it; and since my fastest D40 result is faster than my slowest S39L, maybe D40 is still the best for me to use.
I'm certainly doing better than a few months ago when I was running the standard app and had accidentally underclocked my CPU by nearly half (the FSB got dropped to the lowest possible setting when I flashed the BIOS to fix another problem). Back then WUs were taking 24000!
There's enough variance that you really need about 4 of 5 dozen to be sure if the gain's really small. I did a hardware upgrade and bumped my CPU up 200mhz before the switchover, but after adjusting for the CPU speedup I'm finding that while thier ranges do overlap slightly D40's mean is about on par with the fastest S39L's I've gotten.
Athlon XP 2600+ resalts S39L
)
Athlon XP 2600+
resalts S39L (only 2) - 4578.95, 5131.28
resalts D40 (first 2) - 4281.70 (S39L 65%, D40-35%), 4139.52
D40 faster S39L 12 % (+-7%)
I'am like D40 :)
Akosf, do you know if your
)
Akosf, do you know if your D40 optimizations will make it into the next official Einstein application?
RE: Akosf, do you know if
)
I'm waiting for the new official client, but as far as I know, it will not consist the "combined precision" part like as [S38,S39,S39L,D40].
RE: RE: Akosf, do you
)
To Akosf: Will be S40 in short time or you wait for the new afficial app and S40 will after?
RE: Will be S40 in short
)
Well I have lot of ideas to optimise a code, but the main optimisations are done. I would like to try out some new other tricks (I do it for fun), but it goes slowly because I have lot of work (current company) and lot of other task (own company). Probably I will release an S40 code but it will not faster than S39L. The S40 will have higher code density. It's good for power consumption but doesn't mean higher speed. It's funny, isn't it?
The source level optimisations would be much more lucky, that means more and easier possibilities.
Well this is weird - everyone
)
Well this is weird - everyone else with 3DNow! seems to be going faster under D40 than S39L, but I appear to be going slower. I know the WU size varies, but effect seems to be about 10% slower under D40 than S39L.
S39L results: 4047 3926 3908 3824 3916 3997 3905 3892 3917 4076
S39L / D40 mixed WU result: 3906
D40 results: 4347 4225 4318 4107 4010
My processor is an Athlon XP 2800+ (Barton core: 512k L2 cache @ 2.08GHz MMX(+) 3DNow!(+) SSE)
Should I switch back to S39L, or wait a few more WUs to see what happens?
RE: Well this is weird -
)
Your results are not all from the same major datafile. The three slowest D-40 results are 1447.5, the next faster from 0784.0, and the fastest one so far from 1058.5.
I'm pretty sure I've seen appreciable offsets for compact sets from different major datafiles, and certainly seen drift and the occasional blip even in the same.
Folks confidently reported differences down in the few percent range are being optimistic.
I've noticed that the server is much less likely to keep giving me results form the same major datafile than it was just a few weeks ago. Back then it would go weeks at a time, while now all four of my machines have process results from more than one datafile in the last week.
Okay thanks. I'll wait for a
)
Okay thanks. I'll wait for a good few more WUs and see what happens.
If the difference had been less than 5% I wouldn't have queried it; and since my fastest D40 result is faster than my slowest S39L, maybe D40 is still the best for me to use.
I'm certainly doing better than a few months ago when I was running the standard app and had accidentally underclocked my CPU by nearly half (the FSB got dropped to the lowest possible setting when I flashed the BIOS to fix another problem). Back then WUs were taking 24000!
There's enough variance that
)
There's enough variance that you really need about 4 of 5 dozen to be sure if the gain's really small. I did a hardware upgrade and bumped my CPU up 200mhz before the switchover, but after adjusting for the CPU speedup I'm finding that while thier ranges do overlap slightly D40's mean is about on par with the fastest S39L's I've gotten.
brainchild:
)
brainchild: SSE2+3DNow!
(hellish, but possible!)