> I know that benchmarking is very subjective, was wondering though that topline
> Intel pentium proc. do not appear to be faster, or is WU selection so much
> fine grained that such processors are loaded with the bigger WU's compared to
> say a Celeron D330 ??
> If thats true apples and oranges are mixed indeed even on processing times.
>
> Thanks for your nice doc's about the subject Paul,
>
> Regards,
>
Allthough Paul's answer is much better than mine I wanted to put my 2 cents in.
The main reason high end intel CPUs do not appear faster, in the BOINC benchmarks, is hyperthreading. The benchmarks run on all processors and then are divided by the number of processors. This makes the hyperthreaded CPUs appear about half as fast as they are just by looking at the numbers. They make up for the lower numbers by running 2 workunits at the same time. Even though it takes longer to do each workunit more workunits will get done in the same amount of time.
> Posts in this thread are starting to read like they were issued by The
> Department of Redundancy Department!
>
>
Posts in this thread are starting to read like they were issued by The Department of Redundancy Department!
i wonder why they invented this HT stuff... Would be easier just to build better processors that need less power and have a good performance instead of pretending their processor has the power of two what only makes everything more complicated and in fact doesnt bring any advantage...
some may say: hey, when you use HT on these processors they are faster compared to not using it... Maybe the are only slower than they normaly would be if you dont use HT because they were build to work with this strange mode ? and it wouldnt be too hard to create ONE processor that works like ONE and also has the same performance like these wannabe-Multiprocessor-systems... but ok, that leads to another discussion and everybody himself can decide whether he likes such marceting strategies....
> i wonder why they invented this HT stuff... Would be easier just to build
> better processors that need less power and have a good performance instead of
> pretending their processor has the power of two what only makes everything
> more complicated and in fact doesnt bring any advantage...
What most people don't realize is that if the CPU is not doing anything, well, basically it does not draw much power. Our computers glow like lightbulbs because we run the heck out of them.
So, the engineers already answered your plea ...
And they don't pretend that it can do two, just one and a half or so. And that is the point ... instead of building 2 complete CPUs increasing potential power draw, they just add a few more pieces and give you almost the effectiveness of one plus CPUs for the power draw just over one CPU's.
> some may say: hey, when you use HT on these processors they are faster
> compared to not using it...
Now we are back to the issue of speed vs. throughput ...
> i wonder why they invented this HT stuff... Would be easier just to build
> better processors that need less power and have a good performance instead of
> pretending their processor has the power of two what only makes everything
> more complicated and in fact doesnt bring any advantage...
The idea here is that a single CPU can do one thread at a time. If that thread stops for any reason, perhaps waiting for an event, the whole computer is waiting.
An example: a thread might be accessing an IDE hard disk -- once the commands are sent, the thread needs to wait to get the response.
A hyperthreaded processor runs two threads, so if one stops, the other can keep working. That is better for throughput.
> One of the reasons that the Xeons may have better performance than some of the
> P4s is that (at least the ones I know of, this may not be true for the lastest
> generations) they are using a P3 core that seems to be more effecient in many
> respects than the P4's core.
My home P4/1800 runs off the Xeon 2Ghz work server on my job.
I cannot run einstein on the Xeon server (4x Xeon) because it always gives an error with every einstein WU.
But I can say this with SETI:
P4/3600= 2h30 for 2 WU ("1h15" each)
amd2000xp = 4h for 1WU
P4/1800 = 5h30 for 1 WU
Xeon/2000 = 6h30 for 4 WU ("1h40" each)
Cel1100 = 11h for 1 WU.
I only have 3 computers running Einstein:
amd2000xp = 7h for 1 WU
P4/3600 = 10h for 2 WU ("5h" each)
P4/1800 = 13h for 1 WU.
I already have received credits for 2000xp and P4/3600 but not for P4/1800.
> I know that benchmarking is
)
> I know that benchmarking is very subjective, was wondering though that topline
> Intel pentium proc. do not appear to be faster, or is WU selection so much
> fine grained that such processors are loaded with the bigger WU's compared to
> say a Celeron D330 ??
> If thats true apples and oranges are mixed indeed even on processing times.
>
> Thanks for your nice doc's about the subject Paul,
>
> Regards,
>
Allthough Paul's answer is much better than mine I wanted to put my 2 cents in.
The main reason high end intel CPUs do not appear faster, in the BOINC benchmarks, is hyperthreading. The benchmarks run on all processors and then are divided by the number of processors. This makes the hyperthreaded CPUs appear about half as fast as they are just by looking at the numbers. They make up for the lower numbers by running 2 workunits at the same time. Even though it takes longer to do each workunit more workunits will get done in the same amount of time.
BOINC WIKI
BOINCing since 2002/12/8
Posts in this thread are
)
Posts in this thread are starting to read like they were issued by The Department of Redundancy Department!
There's a fine line between fishing and standing on the shore looking like an idiot -- Steven Wright
Posts in this thread are
)
Posts in this thread are starting to read like they were issued by The Department of Redundancy Department!
> Posts in this thread are
)
> Posts in this thread are starting to read like they were issued by The
> Department of Redundancy Department!
>
>
Posts in this thread are starting to read like they were issued by The Department of Redundancy Department!
Link to Unofficial Wiki for BOINC, by Paul and Friends
i wonder why they invented
)
i wonder why they invented this HT stuff... Would be easier just to build better processors that need less power and have a good performance instead of pretending their processor has the power of two what only makes everything more complicated and in fact doesnt bring any advantage...
some may say: hey, when you use HT on these processors they are faster compared to not using it... Maybe the are only slower than they normaly would be if you dont use HT because they were build to work with this strange mode ? and it wouldnt be too hard to create ONE processor that works like ONE and also has the same performance like these wannabe-Multiprocessor-systems... but ok, that leads to another discussion and everybody himself can decide whether he likes such marceting strategies....
> i wonder why they invented
)
> i wonder why they invented this HT stuff... Would be easier just to build
> better processors that need less power and have a good performance instead of
> pretending their processor has the power of two what only makes everything
> more complicated and in fact doesnt bring any advantage...
What most people don't realize is that if the CPU is not doing anything, well, basically it does not draw much power. Our computers glow like lightbulbs because we run the heck out of them.
So, the engineers already answered your plea ...
And they don't pretend that it can do two, just one and a half or so. And that is the point ... instead of building 2 complete CPUs increasing potential power draw, they just add a few more pieces and give you almost the effectiveness of one plus CPUs for the power draw just over one CPU's.
> some may say: hey, when you use HT on these processors they are faster
> compared to not using it...
Now we are back to the issue of speed vs. throughput ...
> i wonder why they invented
)
> i wonder why they invented this HT stuff... Would be easier just to build
> better processors that need less power and have a good performance instead of
> pretending their processor has the power of two what only makes everything
> more complicated and in fact doesnt bring any advantage...
The idea here is that a single CPU can do one thread at a time. If that thread stops for any reason, perhaps waiting for an event, the whole computer is waiting.
An example: a thread might be accessing an IDE hard disk -- once the commands are sent, the thread needs to wait to get the response.
A hyperthreaded processor runs two threads, so if one stops, the other can keep working. That is better for throughput.
In theory.
> One of the reasons that the
)
> One of the reasons that the Xeons may have better performance than some of the
> P4s is that (at least the ones I know of, this may not be true for the lastest
> generations) they are using a P3 core that seems to be more effecient in many
> respects than the P4's core.
My home P4/1800 runs off the Xeon 2Ghz work server on my job.
I cannot run einstein on the Xeon server (4x Xeon) because it always gives an error with every einstein WU.
But I can say this with SETI:
P4/3600= 2h30 for 2 WU ("1h15" each)
amd2000xp = 4h for 1WU
P4/1800 = 5h30 for 1 WU
Xeon/2000 = 6h30 for 4 WU ("1h40" each)
Cel1100 = 11h for 1 WU.
I only have 3 computers running Einstein:
amd2000xp = 7h for 1 WU
P4/3600 = 10h for 2 WU ("5h" each)
P4/1800 = 13h for 1 WU.
I already have received credits for 2000xp and P4/3600 but not for P4/1800.