Right, but still, 384 MB should be fine unless you run lots of other stuff for Win XP.
Those are the magic words. 384MB should be fine as long as you only run Windows XP. Not much else. And if you run it with all the standard services on, you're in deep trouble already.
Not run Einstein (taking 55MB of RAM). Maybe that APS is something (taking 1.5MB per result).
My girlfriend tried running XP Corp on 756MB of memory.
That worked out fine as soon as she shut down all unnecessary services. Else she looked at a BSOD each time she even thought of playing Oblivion. ;-)
Mine RAM is 320 MB. I am running Linux, SETI, Einstein and QMC. SETI is using even more RAM than Einstein. I have a swap partition of 700 MB, following the old Unix thumb rule of doubling your RAM space, but Linux hardly uses more than 40 MB of it. The L2 cache of my PII Deschutes is 512k, but I was told that that is unimportant for Einstein. Cheers.
Tullio
Not if it is a Celeron... I had problems with my 2.3GHz Celeron back in the day. It just lacked the cache to crunch Einstein. And that was the S4 run. I shudder to think about an S5R2 run on it.
I know I'm going to regret this but I couldn't resist .... :).
I want to talk about Coppermines and their direct descendents, the Tualatins. I want to compare the Celerons of these two families with the next gereration - the Williamettes and Northwoods, both P4 and Celeron varieties. I have a large number of machines that span these architectures and have been intrigued to compare their various performances. When I talk about performance, I'm only considering running the EAH app. I'm not talking about more general computing tasks.
To save time, let me give you just a couple of examples of what I've seen many, many times. This machine is a Tualatin Celeron 1300 overclocked to 1560MHz. It has 256Mb RAM and runs WinXP Pro with all default services. It generates very close to 10 credits/hr. This machine is a Compaq Evo P4 1.8GHz with 512MB RAM. It runs WIN XP Pro under the same conditions. It generates less than 8 credits/hr. Even if I go to this machine which is another Evo P4 - this time a 2.0GHz Northwood, it still generates less than 9 credits/hr.
So which machine is the better cruncher, the Tualatin Celeron 1300 or the 2.0GHz Northwood P4? The Northwood has double the RAM and double the L2 cache so obviously extra RAM and L2 cache don't necessarily make a better performer.
So how do the Coppermine Celerons go? This one is a HP e-PC which runs on a 60 watt (19V, 3A) laptop style PSU. It's a Celeron 900 overclocked to 1060MHz. It generates 6.62 credits/hr which isn't that far short of what the P4 1.8GHz is producing. If you put two of these together, they would take less room than the Evo, use less power than the Evo and way out-produce it by a very large margin.
As a final example, take a look at this machine. It is a Northwood Celeron 2.4GHz running overclocked at 3.2GHz. It was given to me recently so I decided to see what it could do, running at 3.2GHz. It hasn't got much history yet but it is obviously pretty lousy as a cruncher. The first result (the finished one showing) was done mainly at 3.2GHz with a little bit at 2.4GHz. The second result (all at 3.2GHz) will complete shortly and will be 2.5 hours faster but that is still only 9.1 credits/hr. So my Tualatin Celeron 1300 running at 1560MHz is comprehensively beating a Northwood P4 Celeron running at more than double the speed.
The lousy performance of the P4 Celerons is why all Celerons get such a bad rap. The Coppermine and particularly Tualatin P3 varieties don't deserve it.
Well I don't know if it fair to single out the Celery's for additional dissing. Basically what you're proving is Netburst was an dismal failure compared to what came before and after it. ;-)
By and large you get a pretty good deal from Celeron's for the price all things considered, once Intel realized you couldn't cheap out totally and have no onboard L2. Those were the only ones which were truly terrible IIRC.
Thanks for all the replies! Just FYI, I'm now at 21 hours of crunching Einstein and I'm only 6% done, so the estimate to completion appears to be on the ball at almost 300 to go.
My processor is an Intel Celeron, at 797 mHz, with 384 meg of RAM.
In addition to Einstein, I'm running Rosetta and Seti. Rosetta, of course has very short data files, and Seti is at about 100 hours. I have each of the three allotted 1/3 of the time, though with so long a time to completion of the Einstein work, Einstein is taking all the open time. I usually do not have BOINC running continually, as I find some of my other programs really slowed if BOINC is also computing. I have it set to start after 5 minutes of inactivity. I frequently keep the computer on overnight as I am also participating in another interesting project studying shortwave radio propagation. (It's called PropNET and you can check it out at http://propnet.org.)
I will try changing to a blank screen and let you know the results of that, though it wouldn't seem like that simple rotating sky screensaver would take that much computer power. I am running no other processor-intensive programs that I am aware of.
I'll let you know in a day or two what blanking the screen does.
Thanks again and I hope that does the trick!
Quote:
OK -- Now my downloads are over 300 hours long! I thought that new versions of BOINC/Einstein would take care of this problem. I just downloaded the newest version, 5.10.13, and my latest Einstein is 312 hours! By the way, after reading an earlier response to this problem, I've set my connect to network for every 0.1 days (not that that helped any!)
I have aborted 3 previous work files because I just couldn't finish them, even running 24/7! Why do I keep getting these huge files??? I like the project, and UWM is my Alma Mater, but if there's no way to get reasonable (like 100 hours) data files, I'm going to have to give up on Einstein!
By the way, yes, the completion time appears to be accurate, as my computer's been crunching the data for 4 hours, and I'm just 1% done!!! No way I can finish in the two week timeframe. GET REAL GUYS!!!
I will try changing to a blank screen and let you know the results of that, though it wouldn't seem like that simple rotating sky screensaver would take that much computer power. I am running no other processor-intensive programs that I am aware of.
I'll let you know in a day or two what blanking the screen does.
Thanks again and I hope that does the trick!
Hi!
AFAIK the screen saver is written in "OpenGL". Depending on your graphics card & driver, the screen saver might use a lot of software emulation to do the graphics instead of using the graphics card accelerator.
Anyway,
When trying to speed up things, you can check for success rather quickly, no need to wait several days:
This is how to do it:
0) Open a command line window
1) CD to your BOINC directory (like C:\\Program Files\\BOINC or whatever)
2) there are several slot/n/ directories where n is a number. One of them has Einstein files in it for the current task. it will contain files like sun* , earth*, h1_*, Hough.out etc. You can't miss it
3) There's a file named stderr.txt that grows while E@H is computing
4) on the command line:
> type stderr.txt
to display the file's contents. Don't use a fancy editor as this might lock the file (???)
So basically every line is a comma-separated list of integers, terminated at line end with a "c" (for checkpoint)
The average number of integers on each line happens to give a measurement of the speed of computation, because about every other minute a new line is begun.
So, if you do something on your PC (for example, the E@H screen saver kicks in), and afterwards the number of integers per line in the file mentioned above has decreased significantly (some variation is normal), you'll know E@H was slowed down. If the number of integers on each line increases, E@H was accelerated.
This allows you to find the performance culprit (if any) more easily, I hope.
Don't use a fancy editor as this might lock the file (???)
No problem opening it up in Notepad. As long as you do NOT save the file when you leave Notepad, there's no problem opening any of the text files in BOINC, while BOINC is running.
So in my opinion, there's no need for difficult workarounds through the command line.
Don't use a fancy editor as this might lock the file (???)
No problem opening it up in Notepad. As long as you do NOT save the file when you leave Notepad, there's no problem opening any of the text files in BOINC, while BOINC is running.
So in my opinion, there's no need for difficult workarounds through the command line.
Cool, makes things easier=> just click on it in Explorer and open it with Notepad.
Under Linux or Mac OS X, you'll want to use
> tail -f stderr.txt
to watch it live, of course, but there's nothing like that in Windows, I guess....
...I usually do not have BOINC running continually, as I find some of my other programs really slowed if BOINC is also computing. I have it set to start after 5 minutes of inactivity.
normally you won't notice when a BOINC project is running... at least the CPU side!
But rosetta is using lot of memory (around 512MB, but they have different WUs) so I think you will only notice rosetta running due to the memory consumption (you mentioned only having 384 MB of main memory).
Seti WUs I don't know, but running Einstein in parallell to your normal work should do fine!
@Tullio , will try that one
)
@Tullio , will try that one when I come across one, currently I'm on a mix of "Weizen-Bier" and Grapefruit-juice ... yeah, really...
Right, but still, 384 MB should be fine unless you run lots of other stuff for Win XP.
Thanks, I think I got the moderator status last Friday, AFAIK, and I'm glad to be given the opportunity to contribute to the E@H forum this way.
CU
BRM
RE: Right, but still, 384
)
Those are the magic words. 384MB should be fine as long as you only run Windows XP. Not much else. And if you run it with all the standard services on, you're in deep trouble already.
Not run Einstein (taking 55MB of RAM). Maybe that APS is something (taking 1.5MB per result).
My girlfriend tried running XP Corp on 756MB of memory.
That worked out fine as soon as she shut down all unnecessary services. Else she looked at a BSOD each time she even thought of playing Oblivion. ;-)
Mine RAM is 320 MB. I am
)
Mine RAM is 320 MB. I am running Linux, SETI, Einstein and QMC. SETI is using even more RAM than Einstein. I have a swap partition of 700 MB, following the old Unix thumb rule of doubling your RAM space, but Linux hardly uses more than 40 MB of it. The L2 cache of my PII Deschutes is 512k, but I was told that that is unimportant for Einstein. Cheers.
Tullio
RE: RE: The Coppermines
)
I know I'm going to regret this but I couldn't resist .... :).
I want to talk about Coppermines and their direct descendents, the Tualatins. I want to compare the Celerons of these two families with the next gereration - the Williamettes and Northwoods, both P4 and Celeron varieties. I have a large number of machines that span these architectures and have been intrigued to compare their various performances. When I talk about performance, I'm only considering running the EAH app. I'm not talking about more general computing tasks.
To save time, let me give you just a couple of examples of what I've seen many, many times. This machine is a Tualatin Celeron 1300 overclocked to 1560MHz. It has 256Mb RAM and runs WinXP Pro with all default services. It generates very close to 10 credits/hr. This machine is a Compaq Evo P4 1.8GHz with 512MB RAM. It runs WIN XP Pro under the same conditions. It generates less than 8 credits/hr. Even if I go to this machine which is another Evo P4 - this time a 2.0GHz Northwood, it still generates less than 9 credits/hr.
So which machine is the better cruncher, the Tualatin Celeron 1300 or the 2.0GHz Northwood P4? The Northwood has double the RAM and double the L2 cache so obviously extra RAM and L2 cache don't necessarily make a better performer.
So how do the Coppermine Celerons go? This one is a HP e-PC which runs on a 60 watt (19V, 3A) laptop style PSU. It's a Celeron 900 overclocked to 1060MHz. It generates 6.62 credits/hr which isn't that far short of what the P4 1.8GHz is producing. If you put two of these together, they would take less room than the Evo, use less power than the Evo and way out-produce it by a very large margin.
As a final example, take a look at this machine. It is a Northwood Celeron 2.4GHz running overclocked at 3.2GHz. It was given to me recently so I decided to see what it could do, running at 3.2GHz. It hasn't got much history yet but it is obviously pretty lousy as a cruncher. The first result (the finished one showing) was done mainly at 3.2GHz with a little bit at 2.4GHz. The second result (all at 3.2GHz) will complete shortly and will be 2.5 hours faster but that is still only 9.1 credits/hr. So my Tualatin Celeron 1300 running at 1560MHz is comprehensively beating a Northwood P4 Celeron running at more than double the speed.
The lousy performance of the P4 Celerons is why all Celerons get such a bad rap. The Coppermine and particularly Tualatin P3 varieties don't deserve it.
Cheers,
Gary.
Well I don't know if it fair
)
Well I don't know if it fair to single out the Celery's for additional dissing. Basically what you're proving is Netburst was an dismal failure compared to what came before and after it. ;-)
By and large you get a pretty good deal from Celeron's for the price all things considered, once Intel realized you couldn't cheap out totally and have no onboard L2. Those were the only ones which were truly terrible IIRC.
Alinator
Thanks for all the replies!
)
Thanks for all the replies! Just FYI, I'm now at 21 hours of crunching Einstein and I'm only 6% done, so the estimate to completion appears to be on the ball at almost 300 to go.
My processor is an Intel Celeron, at 797 mHz, with 384 meg of RAM.
In addition to Einstein, I'm running Rosetta and Seti. Rosetta, of course has very short data files, and Seti is at about 100 hours. I have each of the three allotted 1/3 of the time, though with so long a time to completion of the Einstein work, Einstein is taking all the open time. I usually do not have BOINC running continually, as I find some of my other programs really slowed if BOINC is also computing. I have it set to start after 5 minutes of inactivity. I frequently keep the computer on overnight as I am also participating in another interesting project studying shortwave radio propagation. (It's called PropNET and you can check it out at http://propnet.org.)
I will try changing to a blank screen and let you know the results of that, though it wouldn't seem like that simple rotating sky screensaver would take that much computer power. I am running no other processor-intensive programs that I am aware of.
I'll let you know in a day or two what blanking the screen does.
Thanks again and I hope that does the trick!
John R
RE: I will try changing to
)
Hi!
AFAIK the screen saver is written in "OpenGL". Depending on your graphics card & driver, the screen saver might use a lot of software emulation to do the graphics instead of using the graphics card accelerator.
Anyway,
When trying to speed up things, you can check for success rather quickly, no need to wait several days:
This is how to do it:
0) Open a command line window
1) CD to your BOINC directory (like C:\\Program Files\\BOINC or whatever)
2) there are several slot/n/ directories where n is a number. One of them has Einstein files in it for the current task. it will contain files like sun* , earth*, h1_*, Hough.out etc. You can't miss it
3) There's a file named stderr.txt that grows while E@H is computing
4) on the command line:
> type stderr.txt
to display the file's contents. Don't use a fancy editor as this might lock the file (???)
anyway, the file will look like this:
...
10042,10043,10044,10045,10046,c
10047,10048,10049,10050,10051,c
...
So basically every line is a comma-separated list of integers, terminated at line end with a "c" (for checkpoint)
The average number of integers on each line happens to give a measurement of the speed of computation, because about every other minute a new line is begun.
So, if you do something on your PC (for example, the E@H screen saver kicks in), and afterwards the number of integers per line in the file mentioned above has decreased significantly (some variation is normal), you'll know E@H was slowed down. If the number of integers on each line increases, E@H was accelerated.
This allows you to find the performance culprit (if any) more easily, I hope.
Good luck,
BRM
RE: Don't use a fancy
)
No problem opening it up in Notepad. As long as you do NOT save the file when you leave Notepad, there's no problem opening any of the text files in BOINC, while BOINC is running.
So in my opinion, there's no need for difficult workarounds through the command line.
RE: RE: Don't use a fancy
)
Cool, makes things easier=> just click on it in Explorer and open it with Notepad.
Under Linux or Mac OS X, you'll want to use
> tail -f stderr.txt
to watch it live, of course, but there's nothing like that in Windows, I guess....
CU
BRM
RE: ...I usually do not
)
normally you won't notice when a BOINC project is running... at least the CPU side!
But rosetta is using lot of memory (around 512MB, but they have different WUs) so I think you will only notice rosetta running due to the memory consumption (you mentioned only having 384 MB of main memory).
Seti WUs I don't know, but running Einstein in parallell to your normal work should do fine!
Udo
