Crunch3r enabled for BOINC V5.8.x 'CPU-affinity' and 'report results immediately'.

He or some other could enable this features in V5.10.7 too?

---

In the latest 5.10, if you set the connection interval to 0 and use the additional work buffer preference to maintain your cache, it effectively reports results as soon as they are finished.

As for affinity, no, I think I read somewhere on boinc_dev that the gains from setting affinity might have not been significant enough to justify trying to code it in the client.

---

In the latest 5.10, if you set the connection interval to 0 and use the additional work buffer preference to maintain your cache, it effectively reports results as soon as they are finished.

As for affinity, no, I think I read somewhere on boinc_dev that the gains from setting affinity might have not been significant enough to justify trying to code it in the client.

It was tested, and the results were a bit of a mixed bag. Some CPUs under some OSs had a slight increase in efficiency. Most were very much the same, but there were a couple of cases that were much worse when CPU affinity was turned on.

---

BOINC WIKI

In the latest 5.10, if you set the connection interval to 0 and use the additional work buffer preference to maintain your cache, it effectively reports results as soon as they are finished.

As for affinity, no, I think I read somewhere on boinc_dev that the gains from setting affinity might have not been significant enough to justify trying to code it in the client.

It was tested, and the results were a bit of a mixed bag. Some CPUs under some OSs had a slight increase in efficiency. Most were very much the same, but there were a couple of cases that were much worse when CPU affinity was turned on.

This makes me wonder how much the question of cpu affinity was tested. Long before cpu affinity was included in any 'optimised' versions of BOINC, I ran two computers, dual P3 and HT P4 (Willameta) both with small L2 cache and got virtual affinity by haaving a very small connect interval, <0.1 days, and got over 10% improvement on Seti by running Seti/Einstein combination compared to Seti/Seti combination. There was no discernable change in Einstein performance.

This effect is no longer seen in my present computers that have much larger L2 cache.

Andy

When L2 is shared (like with any HT or Core or C2D), affinity will have no effect. It is when L2 is not shared on the same chip (like with Pentium D), or on another chip altogether. Affinity should be a machine-level setting.

---

Dublin, California
Team: SETI.USA

When L2 is shared (like with any HT or Core or C2D), affinity will have no effect. It is when L2 is not shared on the same chip (like with Pentium D), or on another chip altogether. Affinity should be a machine-level setting.

I have the QX6700, this is two times Core2 Duo.. with 2 x 4096 KB Cache per Core(2 Duo)..
So normally this are two separately CPUs, or? So 'CPU-affinity' will make a little speep-up?

Or it's not like this, if you don't have 'CPU-affinity' the WUs are jumping from Core(thread) to Core(thread), or two WUs on one Core(thread)?

---

I have the QX6700, this is two times Core2 Duo.. with 2 x 4096 KB Cache per Core(2 Duo)..
So normally this are two separately CPUs, or? So 'CPU-affinity' will make a little speep-up?

Anytime the L2 is not shared across ALL L2 cache, there is the possibility that a thread needs something that is not in the L2 it expects. When that happens, speed is lost.

So yes, with a C2Q, which has 2x L2, affinity should improve speed.

---

Dublin, California
Team: SETI.USA

I have the QX6700, this is two times Core2 Duo.. with 2 x 4096 KB Cache per Core(2 Duo)..
So normally this are two separately CPUs, or? So 'CPU-affinity' will make a little speep-up?

Anytime the L2 is not shared across ALL L2 cache, there is the possibility that a thread needs something that is not in the L2 it expects. When that happens, speed is lost.

So yes, with a C2Q, which has 2x L2, affinity should improve speed.

I would like to profit from the feature in the new V5.10.7, if two results are back, my uncrunched WU will delete.
So the science will be faster..
BUT, who give me now the 'CPU-affinity' for the new Version? :-(

BTW.
This 'delete-feature' is in V5.8.17 and higher.
But where I can get this versions and most important: they are stable?
Because then I could take Crunch3rs V5.9.0!

Now I have V5.8.11 with V5.9.0

---

When L2 is shared (like with any HT or Core or C2D), affinity will have no effect. It is when L2 is not shared on the same chip (like with Pentium D), or on another chip altogether. Affinity should be a machine-level setting.

I have the QX6700, this is two times Core2 Duo.. with 2 x 4096 KB Cache per Core(2 Duo)..
So normally this are two separately CPUs, or? So 'CPU-affinity' will make a little speep-up?

Or it's not like this, if you don't have 'CPU-affinity' the WUs are jumping from Core(thread) to Core(thread), or two WUs on one Core(thread)?

Without 'CPU-affinity' the QX6700 will be ONLY slower?
Nothing other will happen?

---