i just read someone saying they turned off ht? am i correct? is he saying he just set his total cpu's used to 1? on the preferrences

---

Would it make any sense to use algorithms that use HT explicitly for computation, rather than running two processes? That is, for example, one might consider linking a FFTW subroutine explicitly written to take advantage of two threads ( a sort of mini-MPI ). Number crunchers will tend to do this more frequently as multiprocessors w/ HT proliferate.

---

May this Farce be with You

...Number crunchers will tend to do this more frequently as multiprocessors w/ HT proliferate.

I don't think so...

Look up what HT is.

Good luck,
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

...Number crunchers will tend to do this more frequently as multiprocessors w/ HT proliferate.

I don't think so...

Look up what HT is.

Perhaps I'm not being precise, but multiprocessors on the same 'chip' should permit the first level of parallelism (assuming only two exist). So algorithms need to change to take advantage of this. HT isn't quite MP, but it is close enough to offer parrallelism advantages, if the algorithms are re-structured.

So my question remains: is it time to introduce a more general algorithm for the optimized clients?

One counter argument is that there is overhead associated with fine grain parallelism that may be greater than that seti experiences with coarse grain parallelism. Furthermore, seti may be happy with turn around times of a couple of hours. But if the next seti involves longer compute times, then one way to increase the 'activity' is to reduce turn arounds with better algorithms.

---

May this Farce be with You

i just read someone saying they turned off ht? am i correct? is he saying he just set his total cpu's used to 1? on the preferrences

From what you write it sounds like he's saying he disabled hyper-threading on his PC, I think it's done in the bios.

---

Help and BOINC documentation is available here.

You can turn HT on and off in the bios, and you can also decide on how many processors to use in the preferences. These two setting accomplish different things, however.

If you turn off HT in the bios, you will process only 1 WU at a time, but it will use 100% of your processor (since there is only the one).

If you leave HT on, and set your preferences to use only one processor, you will then process one WU at a time, but only use 50% of your processor (since there are now two, but you are using only one).

With HT on, you will generally process WU's slower, but you will process twice as many, with the result being more work done.

---

You can turn HT on and off in the bios, and you can also decide on how many processors to use in the preferences. These two setting accomplish different things, however.

If you turn off HT in the bios, you will process only 1 WU at a time, but it will use 100% of your processor (since there is only the one).

If you leave HT on, and set your preferences to use only one processor, you will then process one WU at a time, but only use 50% of your processor (since there are now two, but you are using only one).

With HT on, you will generally process WU's slower, but you will process twice as many, with the result being more work done.

thanks that answered my question

---

On my 2.8G HT it takes about 4:30 per WU with HT. Without HT it's about 3:30. So, I can either do 1 in 3:30 or 2 in 4:30. You do the math.

---

My movie https://vimeo.com/manage/videos/502242

Um, actually, you will see about a 60% increase, not a doubleling. Though you have 2 in flight, you do not get the full effect of a theoretical second processor. So, you see a throughput gain such that you get 1.5 to 1.8 times the processing power.

Not sure if I am making it clear or not ...

All other things being equal, you do not have the full two logical processors.

---

On my 2.8G HT it takes about 4:30 per WU with HT. Without HT it's about 3:30. So, I can either do 1 in 3:30 or 2 in 4:30. You do the math.

Theres really no math to do IMO...If you have a HT CPU & don't run the Projects using the HT Capability of the CPU then your just short changing yourself.

I run 7 PC's with HT Capability & all of them run 2 WU's under 2 hours time, a few as low as 1:30 Hr's time & I wouldn't think of running them without HT turned on ...

---

...
One counter argument is that there is overhead associated with fine grain parallelism that may be greater than that seti experiences with coarse grain parallelism. Furthermore, seti may be happy with turn around times of a couple of hours. But if the next seti involves longer compute times, then one way to increase the 'activity' is to reduce turn arounds with better algorithms.

The coarse grain parallelism for s@h works very well. There is simply no advantage or any need to use any finer grained parallelism.

Regards,
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

On my 2.8G HT it takes about 4:30 per WU with HT. Without HT it's about 3:30. So, I can either do 1 in 3:30 or 2 in 4:30. You do the math.

Theres really no math to do IMO...If you have a HT CPU & don't run the Projects using the HT Capability of the CPU then your just short changing yourself.

I run 7 PC's with HT Capability & all of them run 2 WU's under 2 hours time, a few as low as 1:30 Hr's time & I wouldn't think of running them without HT turned on ...

thanks

---

Um, actually, you will see about a 60% increase, not a doubleling. Though you have 2 in flight, you do not get the full effect of a theoretical second processor. So, you see a throughput gain such that you get 1.5 to 1.8 times the processing power.

Not sure if I am making it clear or not ...

All other things being equal, you do not have the full two logical processors.

thanks

---