I'm still a bit puzzled. The changes I made from 4.16 to 4.17 were some that I would have expected to slow things down. However on my Xeons they sped things up quite noticably. My current guess is that CPUs with large caches will rather benefit from these changes, while they slowed down things on all others (though not much).
I'm still a bit puzzled. The changes I made from 4.16 to 4.17 were some that I would have expected to slow things down. However on my Xeons they sped things up quite noticably. My current guess is that CPUs with large caches will rather benefit from these changes, while they slowed down things on all others (though not much).
Does this sound reasonable?
BM
Hi Bernd
Well I only have 512 L2 cache, which isn't exactly huge these days and noticed a speedup here...
I didn't run 4.16 but now on 4.17 my Opteron with 1mb L2 cache seem to be a little bit faster than the X2 with 512kb L2 cache. On 4.01 they were alike.
Thanks for the link Michael, I missed out on 4.16 so I didn't see the post.
So I gather 'small x' is for debugging, to show how often some bit of code is executed, or some value is returned.
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman
I'm still a bit puzzled. The changes I made from 4.16 to 4.17 were some that I would have expected to slow things down. However on my Xeons they sped things up quite noticably. My current guess is that CPUs with large caches will rather benefit from these changes, while they slowed down things on all others (though not much).
Does this sound reasonable?
BM
I have a 512K L2 cache on my Pentium II and 4.17 was slower than 4.16 and also 4.01. What I saw was that 4.17 used very little of my 320 MB RAM, which means it did not cache disk.SuSE Linux 9.3.
I have a 512K L2 cache on my Pentium II and 4.17 was slower than 4.16 and also 4.01. What I saw was that 4.17 used very little of my 320 MB RAM, which means it did not cache disk.SuSE Linux 9.3.
I payed most attention to the SSE code, the "generic" on hasn't changed much, so my statement about cache referrs to the SSE code only.
The Einstein@Home Application doesn't use much memory at all, just a few MB, plus the stuff that the code for the OS it's running on needs (code, libraries etc.). It also doesn't write much. It'ts totally CPU bound.
so far the switch for me from v4.01 to v4.17 has speeded up one of my systems 22.5% and another by 16.7%. I have just updated my other remaining system and will have to monitor it for a while to know on it.
Thor-nordic23
Member of: Linux Users Everywhere
RE: What does 'small x'
)
small x
Michael
Team Linux Users Everywhere
I'm still a bit puzzled. The
)
I'm still a bit puzzled. The changes I made from 4.16 to 4.17 were some that I would have expected to slow things down. However on my Xeons they sped things up quite noticably. My current guess is that CPUs with large caches will rather benefit from these changes, while they slowed down things on all others (though not much).
Does this sound reasonable?
BM
BM
I installed 4.17 on an P4
)
I installed 4.17 on an P4 1.7. Compared to 4.01 there is a speedup of roughly 25%!
See 37890519 (4.17) and 37874836 (4.01).
Michael
Team Linux Users Everywhere
RE: I'm still a bit
)
Hi Bernd
Well I only have 512 L2 cache, which isn't exactly huge these days and noticed a speedup here...
Gray
I didn't run 4.16 but now on
)
I didn't run 4.16 but now on 4.17 my Opteron with 1mb L2 cache seem to be a little bit faster than the X2 with 512kb L2 cache. On 4.01 they were alike.
Thanks for the link Michael,
)
Thanks for the link Michael, I missed out on 4.16 so I didn't see the post.
So I gather 'small x' is for debugging, to show how often some bit of code is executed, or some value is returned.
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman
RE: I'm still a bit
)
I have a 512K L2 cache on my Pentium II and 4.17 was slower than 4.16 and also 4.01. What I saw was that 4.17 used very little of my 320 MB RAM, which means it did not cache disk.SuSE Linux 9.3.
RE: I have a 512K L2 cache
)
I payed most attention to the SSE code, the "generic" on hasn't changed much, so my statement about cache referrs to the SSE code only.
The Einstein@Home Application doesn't use much memory at all, just a few MB, plus the stuff that the code for the OS it's running on needs (code, libraries etc.). It also doesn't write much. It'ts totally CPU bound.
BM
BM
679091authentic mobile AMD
)
679091authentic mobile AMD Athlon XP-M 2400+ Linux 2.6.8-24.24-default (Suse 9.2)
Computation times ( in seconds )
v4.01 average: 31881.45 v4.01 std dev: 775.56 [sample of 15 results]
v4.16 average: 27903.87 v4.16 std dev: 301.39 [sample of 5 results]
v4.17 average: 27677.37 v4.17 std dev: 297.40 [sample of 13 results]
Compare v4.16/v4.01 = 12.48% faster
Compare v4.17/v4.01 = 13.19% faster
Compare v4.17/v4.16 = +0.81% faster
"Entia non sunt multiplicandam praeter necessitatem"
(OKHAM)
so far the switch for me from
)
so far the switch for me from v4.01 to v4.17 has speeded up one of my systems 22.5% and another by 16.7%. I have just updated my other remaining system and will have to monitor it for a while to know on it.
Thor-nordic23
Member of: Linux Users Everywhere
Don't be a windows crash dummy, USE LINUX!!