Prime95 version 25.7

[James Heinrich]

Would it be possible to extend the TF assignments slightly such that a client (or via the webpage) can specify a preferred TF range? Right now there's "low limits" and "regular", but perhaps a user (for whatever reasons, limited internet connection for example) prefers to do long assignments, and would like to only TF in the 2^67-2^72 range, whatever assigments are available and most useful in that depth). Another user may prefer to do mid-depth work (e.g. 2^63-2^65), etc. If, truly, there is no useful work to be done in those ranges then assign something outside the specified range, but consider it as a preference.
Is there any downside to this (other than the work to implement it) ?

[Prime95]

Originally Posted by petrw1

Hey, wait a minute:  Worker 1 is faster one 1 CPU than both?!?!?!?

This is very possible on a hyperthreaded machine. Hyperthreading takes one real CPU and makes it look like two to the operating system. So whether you have Worker 1 use one logical CPU or two, the work is being by the one physical CPU.

Using the hyperthreaded CPU in LL testing usually results in a speed difference in the range of plus or minus 5%.

[petrw1]

Originally Posted by Prime95

This is very possible on a hyperthreaded machine. Hyperthreading takes one real CPU and makes it look like two to the operating system. So whether you have Worker 1 use one logical CPU or two, the work is being by the one physical CPU.

Using the hyperthreaded CPU in LL testing usually results in a speed difference in the rane of plus or minus 5%.

I tried the same thing with another Hyper.

Software Version Windows,Prime95,v25.7,build 3
Model Intel(R) Pentium(R) 4 CPU 3.40GHz
Features Single core, 2 hyperthreaded CPUs, Prefetch,SSE,SSE2
Speed 3.391 GHz (3.540 GHz P4 effective equivalent)
L1/L2 Cache 16 / 1024 KB
Computer Memory 1016 MB configured usage 200 MB day / 200 MB night

Same results:

[Dec 3 08:50] Worker starting
[Dec 3 08:50] Setting affinity to run worker on any logical CPU.
[Dec 3 08:50] Setting affinity to run helper thread 1 on any logical CPU.
[Dec 3 08:51] Resuming primality test of M28462163 using FFT length 1536K, 2 threads
[Dec 3 08:51] Iteration: 2753075 / 28462163 [9.67%].
[Dec 3 08:51] Iteration: 2753200 / 28462163 [9.67%]. Per iteration time: 0.075 sec.
[Dec 3 08:51] Iteration: 2753300 / 28462163 [9.67%]. Per iteration time: 0.049 sec.
[Dec 3 08:51] Iteration: 2753400 / 28462163 [9.67%]. Per iteration time: 0.049 sec.
[Dec 3 08:51] Iteration: 2753500 / 28462163 [9.67%]. Per iteration time: 0.049 sec.
[Dec 3 08:51] Iteration: 2753600 / 28462163 [9.67%]. Per iteration time: 0.049 sec.
[Dec 3 08:51] Stopping primality test of M28462163 at iteration 2753658 [9.67%]
[Dec 3 08:51] Worker stopped.
[Dec 3 08:51] Worker starting
[Dec 3 08:51] Setting affinity to run worker on any logical CPU.
[Dec 3 08:51] Resuming primality test of M28462163 using FFT length 1536K
[Dec 3 08:51] Iteration: 2753659 / 28462163 [9.67%].
[Dec 3 08:51] Iteration: 2753700 / 28462163 [9.67%]. Per iteration time: 0.047 sec.
[Dec 3 08:51] Iteration: 2753800 / 28462163 [9.67%]. Per iteration time: 0.047 sec.
[Dec 3 08:52] Iteration: 2753900 / 28462163 [9.67%]. Per iteration time: 0.114 sec.
[Dec 3 08:52] Iteration: 2754000 / 28462163 [9.67%]. Per iteration time: 0.047 sec.
[Dec 3 08:52] Iteration: 2754100 / 28462163 [9.67%]. Per iteration time: 0.047 sec.
[Dec 3 08:52] Iteration: 2754200 / 28462163 [9.67%]. Per iteration time: 0.047 sec.

[Phantomas]

Originally Posted by pertw1

L1/L2 Cache 8 / 512 KB

Maybe both threads are fighting for the cache. I would try to set the cachesize manualy to the half of your true cachesize in p95.

[starrynte]

after about a minute running prime95, what appears as "System" starts "stealing" cpu cycles..."System" has been using 10% cpu for about 4 hours now, causing prime95 to run 15% slower than normal...any reasons why "System" keeps using cycles?
(i have two workers on a hyperthreaded machine)

[petrw1]

Originally Posted by starrynte

after about a minute running prime95, what appears as "System" starts "stealing" cpu cycles..."System" has been using 10% cpu for about 4 hours now, causing prime95 to run 15% slower than normal...any reasons why "System" keeps using cycles?
(i have two workers on a hyperthreaded machine)

As I posted just a few before yours.

I found that on both my Hyperthreading machines with v25.7 they are actually about 5-8% SLOWER with two workers than with just one.

I suppose the other possible answer is Virus / Spyware.

[petrw1]

Model Intel(R) Core(TM)2 Quad CPU Q9550 @ 2.83GHz 
Features 4 core, Prefetch,SSE,SSE2,SSE4 
Speed 2.833 GHz (6.055 GHz P4 effective equivalent)

In the case above is Prime95 determining that each core is 6.055 X P4 Ghz
OR
is it for the entire machine (all 4 cores). If so it is very low but then I've only had it on v5 for 2 weeks so there wouldn't be much data yet.

In comparison I have this Duo:

Model Intel(R) Core(TM)2 Duo CPU E6550 @ 2.33GHz 
Features Dual core, Prefetch,SSE,SSE2 
Speed 2.327 GHz (7.818 GHz P4 effective equivalent)

which I know isn't 30% faster per core than the Quad above. It has about 1 month of v5 time in.

[Prime95]

The P4 effective speed is for each core. It is initially calculated at something like 1.6 * your CPU speed (a Core2 is architecturally faster than a P4). Hours per day and rolling average also figure into the calculation so it will change over time.

[petrw1]

Originally Posted by Prime95

The P4 effective speed is for each core. It is initially calculated at something like 1.6 * your CPU speed (a Core2 is architecturally faster than a P4). Hours per day and rolling average also figure into the calculation so it will change over time.

Fair enough but doesn't 7.818 (now 7.896) seem excessively high even for a Duo. It's only a 2.33 Ghz. It is more than twice the default of 1.6 times higher. I have NOT overclocked.

[Prime95]

Originally Posted by petrw1

Fair enough but doesn't 7.818 (now 7.896) seem excessively high even for a Duo. It's only a 2.33 Ghz. It is more than twice the default of 1.6 times higher.

You are correct. The client is underestimating the speed or Core 2s. Thus your rolling average is much higher than it should be. This will be fixed in 25.8 build 3. I've also added better initial client estimates for Core i7 and Phenom.

Note this isn't a serious problem. It has no affect on the work assigned.

[petrw1]

Originally Posted by Prime95

This will be fixed in 25.8 build 3.

Is it imminent? Any reason not to use Build 2 in the meantime?