Is the smaller working set size showing itself in reduced 'register misses', or less data needing pulled from the L2 cache/ram after a different app is ran. With my machine being dualcore, if the cores don't share cache that could be some of the gain as well, but couldn't explain why single core procs also saw a gain.
~13K seconds C37
~11K seconds S38 = ~85% of the C37 Time
~7.8K seconds S39L = ~60% of the C37 time, ~71% of the S38 time
I never tried the standard client nor the S39. These times are on a P4 520 Prescott 2.8 that has only a 16K L1 cache.
My Laptop however, a Pentium M Banias 1.4GHz with 32K L1 cache is flying through the results ad did the first unit in 1797.5 seconds, the second unit is going much slower so we will see what the average is.
How can we tell if we are getting the small or large work units?
How can we tell if we are getting the small or large work units?
I suggest that this occurs because of the two different ligo sites. So there are "r" units and "z" units. I experienced the r wus to be smaller so they are getting done earlier but tell me if I'm wrong.
How can we tell if we are getting the small or large work units?
I suggest that this occurs because of the two different ligo sites. So there are "r" units and "z" units. I experienced the r wus to be smaller so they are getting done earlier but tell me if I'm wrong.
i doubt that this is accurate.....
Currently I do have "r" units which are larger and "z" units which are smaller....
Here are the results for my
)
Here are the results for my P3s comparing S39L to S39:
P3mobile 1GHz: 2.9% faster
P3T 1.26GHz: 5.5% faster
On Dothan 1,73: -
)
On Dothan 1,73: - ~6,4%
great work akosf!
:
your thoughts - the ways :: the knowledge - your space
:
About 5% better on my a64x2
)
About 5% better on my a64x2 2.4gig.
Is the smaller working set size showing itself in reduced 'register misses', or less data needing pulled from the L2 cache/ram after a different app is ran. With my machine being dualcore, if the cores don't share cache that could be some of the gain as well, but couldn't explain why single core procs also saw a gain.
RE: Here are the results
)
The second surprise.
I was waiting these results, but I expected bigger difference before them.
Thanks Stef!
Thank all crunchers!
AMD Athlon(tm) 64 Processor
)
AMD Athlon(tm) 64 Processor 3400+ comparing S39L to S39: -5%
Intel(R) Pentium(R) 4 CPU 3.00GHz : -5,5%
Great work akosf :)
Anders n
A64 3200+ @2.5GHz from
)
A64 3200+ @2.5GHz
from 3388s(s39) down to 3270s(s39L)
:)
Wish you can understand my English:)
~13K seconds C37 ~11K seconds
)
~13K seconds C37
~11K seconds S38 = ~85% of the C37 Time
~7.8K seconds S39L = ~60% of the C37 time, ~71% of the S38 time
I never tried the standard client nor the S39. These times are on a P4 520 Prescott 2.8 that has only a 16K L1 cache.
My Laptop however, a Pentium M Banias 1.4GHz with 32K L1 cache is flying through the results ad did the first unit in 1797.5 seconds, the second unit is going much slower so we will see what the average is.
How can we tell if we are getting the small or large work units?
RE: How can we tell if we
)
I suggest that this occurs because of the two different ligo sites. So there are "r" units and "z" units. I experienced the r wus to be smaller so they are getting done earlier but tell me if I'm wrong.
Greetings, Santas little helper
RE: How can we tell if we
)
The name of the small one starts with "z"
microcraft
"The arc of history is long, but it bends toward justice" - MLK
RE: RE: How can we tell
)
i doubt that this is accurate.....
Currently I do have "r" units which are larger and "z" units which are smaller....