It has been suggested to set up a thread for this -
Maybe Travis can move some of the posts from the other thread to over here.


I set up 2 new machines. Both q6600 -

Same video
Same Power Supply
Same Hard Drive
Same Memory
Same clocking - Stock
Same Case
Same Motherboard

In fact, these are identical in every way possible. They even look alike.

I can post photos if requested.

Plugged them in to the same power source in case of power outages
Plugged them in to the same network hub

Turned them on at the exact same time.

Attached MW and Seti on both. And set to no new work.
Since during the attach BOINC likes to download a work unit, I aborted the SETI WU, but let the MW finish.


They both start out with zero credits.

Here they are.


Box1
Box2

Box2 will be running SETI, Box1 will be running MW

Box 2 SETI


I will click the start button on both at the same time.


I am not sure what the best time frame would be for testing - 7 days? Suggestion ?


If you choose you can set up a q6600 as well to back up my findings.


And they are off and running.


SETI BOX
CPU type GenuineIntel
Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz [EM64T Family 6 Model 15 Stepping 11]
Number of CPUs 4
Operating System Microsoft Windows XP
Professional x64 Editon, Service Pack 2, (05.02.3790.00)
Memory 2045.59 MB
Cache 244.14 KB
Swap space 2998.23 MB
Total disk space 76.68 GB
Free Disk Space 68.14 GB
Measured floating point speed 2276.25 million ops/sec
Measured integer speed 7105.46 million ops/sec


MW Box

GenuineIntel
Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz [EM64T Family 6 Model 15 Stepping 11]
Number of CPUs 4
Operating System Microsoft Windows XP
Professional x64 Editon, Service Pack 2, (05.02.3790.00)
Memory 2045.55 MB
Cache 244.14 KB
Swap space 3962.42 MB
Total disk space 76.68 GB
Free Disk Space 70.39 GB
Measured floating point speed 2286.8 million ops/sec
Measured integer speed 7128.74 million ops/sec




**** EDIT *** After starting up SETI, it downloaded 5 240 hour astropulse, I changed my prefs to run only SETI Enhanced assuming these are shorter WU's. And reset.
Approx 5 minutes were lost. Should not be a problem


**** Second edit ***

Now I remember why I don't run SETI!!!

2/22/2009 15:39:33|SETI@home|Temporarily failed download of setigraphics_6.03_windows_intelx86.exe: HTTP error
2/22/2009 15:39:33|SETI@home|Backing off 1 min 0 sec on download of setigraphics_6.03_windows_intelx86.exe
2/22/2009 15:39:33|SETI@home|Temporarily failed download of setiathome-6.03_AUTHORS: HTTP error
2/22/2009 15:39:33|SETI@home|Backing off 1 min 0 sec on download of setiathome-6.03_AUTHORS
2/22/2009 15:39:33|SETI@home|Temporarily failed download of setiathome-6.03_README: HTTP error
2/22/2009 15:39:33|SETI@home|Backing off 1 min 0 sec on download of setiathome-6.03_README
2/22/2009 15:39:33|SETI@home|Temporarily failed download of seti_603.jpg: HTTP error
2/22/2009 15:40:03|SETI@home|Started download of 16ja09ab.12794.18068.11.8.77
2/22/2009 15:40:19||Project communication failed: attempting access to reference site
2/22/2009 15:40:19|SETI@home|Temporarily failed download of setiathome_6.03_windows_intelx86.exe: HTTP error

---

.

Which is why I suggested using the published CPCS for the hosts from BOINCStats. Project availability problems don't effect the outcome of the experiment, as long as you can get a representative sample of the work stream. Of course right now for SAH it's debatable how long that might take! ;-)

The problem with the Combined BOINC Statistics CPP Grid (and the others as well I'm assuming) is that it only tracks hosts which are common to both of the projects in question.

Your wish is granted.

Milkyway Host


SETI Host


I have set the cache on both boxes to 8 days. Not that MW will use it, but with the problems that SETI has, I thought it might be wise to at least have a cache of SETI there.


** I am going to take liberty of running FreeHal on both boxes - this is a non-CPU intensive program and should not impede in the challenge. If there are those that think Freehal should not be run, then let me know and I will remove it. **

---

.

LOL...

Looks good from here... So the Green Flag is officially waved! :-D

Alinator

Attached MW and Seti on both.

I was wondering why on both. I would have thought that one should have one and one the other. Theoretically files installed by SETI could taint MW, and vica vica versa of course.

---

Good luck then. Waitin' for final results.

Attached MW and Seti on both.

I was wondering why on both. I would have thought that one should have one and one the other. Theoretically files installed by SETI could taint MW, and vica vica versa of course.

I don't know, just because I guess.

I have not problem in detaching.


Done!



** I will do my best that if one project goes down and runs out of cache,,, more than likely it will be MW, since I can cache more work at SETI. That I will suspend the other until work is available. **

---

.

I saw your post in the grandstand about the cache setting. If you are going to run both projects on both hosts, that's probably a good idea given the problems with availability at SAH lately.

Regarding Ice's observation, I suppose you could run the test both ways to see if that made any difference. It has been documented that a heterogeneous set of projects can improve performance in some cases, but I don't think it should make that big a one in this case.

Regarding FreeHAL, I leave that up to you. Personally, I would suggest starting without it though, in order to get an initial baseline with the fewest variables possible.

Alinator

giving all setis the full grant its at 1716 against 2489 at MW, but certainly I don't know how many seti wus are ready but not reported.
So the first 11,5 hours seem to be going to MW.

Very fine test Kevin, its really great that you are doing it. So we will get clear numbers.

I know you don't like to compare opt. appl. but couldn't you do them too, afterwards? To get clear data for that, too.

---

Dragons can fly because they don't fit into pirate ships!

giving all setis the full grant its at 1716 against 2489 at MW, but certainly I don't know how many seti wus are ready but not reported.
So the first 11,5 hours seem to be going to MW.

Very fine test Kevin, its really great that you are doing it. So we will get clear numbers.

I know you don't like to compare opt. appl. but couldn't you do them too, afterwards? To get clear data for that, too.

I think it is still too early to determine anything. I think at least a week, possibly longer. Someone suggested 4 weeks, I don't know if we need to go that far.

Just at the end of the test, clear the caches at the same time, and let SETI finish granting the quorums.

To testing opt app to opt app. I can, but I don't see what this would accomplish, it is like comparing speeds between a boat and a plane.

---

.

The reason I suggested 2 to 4 weeks is this will improve the odds of seeing the widest spectrum of Angle Ranges (AR) for SAH. There is still a fairly non linear aspect to the credit rate vs AR across the whole apectrum of drift angles.

Likewise there is some difference in rate per slice here, albeit not as large as SAH's.

A longer run tends to average that out better.

Alinator

To testing opt app to opt app. I can, but I don't see what this would accomplish, it is like comparing speeds between a boat and a plane.

I understand what you're saying here. But at the end of the day I might want to know where it's best for me to crunch, which project, if I want to maximise my credits. I could say, in this case, that I would not be interested in the stock comparisons between SETI and MW - if I intend to use op apps and not stock apps. I would want to know which returns the most - the SETI op app or the MW op app. (Hypothetically since I can't crunch in SETI anyway.) That is relevant even if I have GPU since I don't have GPU in every box I am crunching.

However, I can understand that the stock to stock comparison is vital since it also affects the MW credit awarded in op apps and GPU, irrespective of whether I or anyone else crunches in SETI. So I'm very interested to see this test outcome.

---

But at the end of the day I might want to know where it's best for me to crunch, which project, if I want to maximise my credits.

This is not to see what give you personaly more credits per hour but to achieve credit parity between the stock applications.

To get the data for your own machine you will have to run and watch yourself, especially as the optimisations may vary considerably for different OS/CPU/GPU setups.

---

Grüße vom Sänger

But at the end of the day I might want to know where it's best for me to crunch, which project, if I want to maximise my credits.

This is not to see what give you personaly more credits per hour but to achieve credit parity between the stock applications.

To get the data for your own machine you will have to run and watch yourself, especially as the optimisations may vary considerably for different OS/CPU/GPU setups.

That's a very good point, but it usually comes down to which project(s) you want to crunch. Credit is nice to have, but credit is more of an indication of 'I did more for this project than that project'.

---

To testing opt app to opt app. I can, but I don't see what this would accomplish, it is like comparing speeds between a boat and a plane.

I understand what you're saying here. But at the end of the day I might want to know where it's best for me to crunch, which project, if I want to maximise my credits. I could say, in this case, that I would not be interested in the stock comparisons between SETI and MW - if I intend to use op apps and not stock apps. I would want to know which returns the most - the SETI op app or the MW op app. (Hypothetically since I can't crunch in SETI anyway.) That is relevant even if I have GPU since I don't have GPU in every box I am crunching.

However, I can understand that the stock to stock comparison is vital since it also affects the MW credit awarded in op apps and GPU, irrespective of whether I or anyone else crunches in SETI. So I'm very interested to see this test outcome.

I understand. How much will you pay me :) I really dislike running SETI, this little crunchoff is slowing me down in other ventures. True, it is only 2 quads out of my farm, but I would like to get back to my previously scheduled program soon.
And for me not having 2 boxes sitting at standard clocking not overclocked - well, this is just criminal!


Either way - you really want to know what pays better. Stock, is ABC-64, PrimeG-64, and even Cosmo is doing better than MW or Seti(from what I have heard) as well as E@H.

But Ice, maybe I could do that for you. But I am fairly sure the winner would be MW, optimized to optimized. But since I have not run SETI in a long time I could not be positive.

---

.

True, it is only 2 quads out of my farm

*envy* I have one quad, an overclocked Q6600 @ 3.26GHz, along with the Radeon HD4870 I got running today, and three dual-core laptops. Do you think being a student would have something to do with that? :P

---

I know you don't like to compare opt. appl. but couldn't you do them too, afterwards? To get clear data for that, too.

Another thing to consider there would be that different opt apps are preferable on different systems.

My experience shows that MW opt SSE3 outshines Seti's opt SSE3 astropulse on an AMD FX74, but on the core 2 duo it's the other way round. Although the developers of the opt apps at Seti would deny it, their opt apps are coded (whether by design or accident) using tools (and optimisations?) that prefer Intel CPUs.

I have found (in my experience and by my calculations, which I can't check while Seti is down for maintenance) that on an E6750 running at 3.2GHz, Astropulse gives more credits. On an AMD FX74 underclocked to 2.4GHz (due to inadequate cooling for the next few days) MW dramatically outshines Astropulse.

Thus the system makes a big difference to which is better. I would not be surprised if the same applies to stock apps as well.

---

Another thing to consider there would be that different opt apps are preferable on different systems.

My experience shows that MW opt SSE3 outshines Seti's opt SSE3 astropulse on an AMD FX74, but on the core 2 duo it's the other way round. Although the developers of the opt apps at Seti would deny it, their opt apps are coded (whether by design or accident) using tools (and optimisations?) that prefer Intel CPUs.

I have found (in my experience and by my calculations, which I can't check while Seti is down for maintenance) that on an E6750 running at 3.2GHz, Astropulse gives more credits. On an AMD FX74 underclocked to 2.4GHz (due to inadequate cooling for the next few days) MW dramatically outshines Astropulse.

Thus the system makes a big difference to which is better. I would not be surprised if the same applies to stock apps as well.

MW has always done better on AMD's while SETI has done better on Intel's

I am only planning on testing on the q6600 chip.

---

.

How much will you pay me :)

I've had a good think about that, but decided to keep my hands in my pockets. Maybe not knowing is best. Half the fun is trying to work out where it is best to crunch, and with what resources ;)

---

[quoteMW has always done better on AMD's while SETI has done better on Intel's

I am only planning on testing on the q6600 chip.
[/quote]

My reply was intended to show Kalessin howcomplicated it could be if tests were done to compare all the options. Any test done by anyone (currently yourself and kelsieb) that compares credits in any two projects should be welcome, as long as nobody takes one set of results as being truth for all possible machines :)

---

[quoteMW has always done better on AMD's while SETI has done better on Intel's

I am only planning on testing on the q6600 chip.

My reply was intended to show Kalessin howcomplicated it could be if tests were done to compare all the options. Any test done by anyone (currently yourself and kelsieb) that compares credits in any two projects should be welcome, as long as nobody takes one set of results as being truth for all possible machines :)[/quote]

True statement!

This is one reason that I have always said that cross project parity is just someone's dream or nightmare as the case may be.

You can only really compare within a certain project.

---

.