Production guessestimate needed

[dragongoddess]

What can I expect production wise for the following. I know exact numbers are not available but a ballpark figure would help.

XP2700+ PC3200 512mb Kingston hyperX
XP2400+ PC3200 512mb Kingston hyperX
XP2000+ 256mb PC2100
AMD 1.4ghz
P4 2.4B dual channel 256mb pc3200 kingston hyperX

[PCZ]

dragongoddess

The daily output is extremely variable so it is difficult to make an estimate.
My own rule of thumb with this client is 18k for each 1GHZ a day.
As you have nearly 10GHZ thats about 180k a day.

You certainly have enough horsepower to go green.


Brian

[dragongoddess]

thankyou

[CodeMonkey]

Hmmm. Your estimate presents interesting possibilities to do some benchmarking. Might try running just HT processor with dual clients for a few days, then just a single client for a few days more. I'm still not convinced that "HYPERTHREADING" is actually producing more output running two clients....

[PCZ]

CodeMonkey

It is almost impossible to benchmark this client.
Sometimes you can get 250 structs done in 24 hrs on a fast PC.
Then the same PC will take 7 days to do the next 250.
I have seen P3's beating Athlon XP 2600's.

HyperThreading certainly gave a good boost when I was doing F@H [about 30%]

Brian

[Brian the Fist]

Originally posted by PCZ
CodeMonkey

It is almost impossible to benchmark this client.

Brian

Thats why we added a -bench option (see the README)...

[PCZ]

You cannot predict what the output of a PC is going to be from the built in benchmark.

What people want to know is how many points per day they are likely to score.
This is impossible to predict, that's what I meant when I said it was impossible to benchmark this client.

[Mikus]

Thats why we added a -bench option (see the README)...

I'm not finding it easy to interpret the benchmark output. Here are the (arithmetic averages of several runs) benchmark numbers for two different systems. How much "better" is the one system over the other ?

mikus


SYSTEM1: Usr time Sys time
-------- --------
Maketrj 03.71 00.62
Foldtraj 39.78 10.71

SYSTEM2: Usr time Sys time
-------- --------
Maketrj 03.21 00.66
Foldtraj 42.56 10.22

[AMD_is_logical]

Originally posted by Mikus
I'm not finding it easy to interpret the benchmark output. ...

Same here. I want one number, not four. One number that tells how fast the client will actually run on a particular machine.

I suppose I can ignore the "Maketrj" line and look only at the "Fold" line, since the client normally spends almost all its time folding. But what exactly is "Usr" vs "Sys" time? I would think that a crunching program would spend almost all its time crunching, and almost no time going to the OS for I/O or whatever.

[Mikus]

Originally posted by AMD_is_logical
I suppose I can ignore the "Maketrj" line and look only at the "Fold" line, since the client normally spends almost all its time folding. But what exactly is "Usr" vs "Sys" time?

One thing I find difficult to understand is the different RATIO of "Sys" vs "Usr" time in the runs on the two systems. System1 and System2 were IDENTICAL, except for CPU (and motherboard revision level). I thought that running the SAME benchmark they would both be invoking exactly the same amount of system services. [And the individual runs I made on each system ALSO show variations in the ratio of "Sys" vs "Usr" time.]

mikus

[Mikus]

Here are the (arithmetic averages of several runs) benchmark numbers for before and after installing the 7/22 client. Are the old client and the new client equivalent ?

mikus


Old client: Usr time Sys time
-------- --------
Maketrj 03.21 00.66
Foldtraj 42.56 10.22


New client: Usr time Sys time
-------- --------
Maketrj 03.24 00.62
Foldtraj 42.85 10.12

[Brian the Fist]

The benchmark test will never change (unless we find a bug) or it would be a useless measure. If you want just one number, add the usr+sys time for foldtraj together and ignore the maketrj one.

Sys time is time spent is OS calls - this includes disk access, network access as well as memory allocation, etc etc.