I would like to attach an Intel with little cache to check how that does, but I can't get my hands on a such just now... certainly not one I could let run Linux.
and I think we had someone else here who complained about extreme runtimes on his Celeron. I checked and that box has 128 KB of cache...
I remember that, but the comment was made on the forecast for the ETA, which can be off by miles. That Celeron was a Coppermine, we'll know only in a few days what the real runtime was.
Today I downloaded (for fun) a document from AMD about optimizing code in C , C++ and assembler for the Opterons & Athlon 64s. This is nice reading (400 pages!!!). And it's surprising what huge penalties are suffered performance-wise for things you wouldn't expect to.
For example if you are working with 10 byte Floating points and store them in an array without padding, basically you are toast :-)
CU
BRM
LOL, I've downloaded a couple of these documents too, during the last days, because I want to refresh my assembler knowledges.
Very interesting stuff(about the K10 too).
LOL, I've downloaded a couple of these documents too, during the last days, because I want to refresh my assembler knowledges.
Very interesting stuff(about the K10 too).
cu,
Michael
Yeah, brings back memories as well, like juggling with the few registers the 486 had ... a long time ago for me.
@Annika:
Quote:
I would like to attach an Intel with little cache to check how that does, but I can't get my hands on a such just now... certainly not one I could let run Linux.
I've got a Dual CPU PIII Coppermine (256KB L2 at core spped), but it happened to receive two >500 credit WUs a short time ago, so it will be busy for some time :-(
I've now put my AMD Athlon XP 1800+ (Palomino), nicknamed "the heater" :-) on the beta test. We'll see...
I would like to attach an Intel with little cache to check how that does, but I can't get my hands on a such just now... certainly not one I could let run Linux.
Forget about the cache, this Athlon XP 2600+(Thoroughbred B) has just 256kB L2 cache and is doing fine without any errors with 4.18 and 4.21(the last result). My X2 has 2 · 1MB L2 cache(2GB RAM) and is running 3 VMWare sessions, two of them dedicated crunchers, one is home for a complete InterShop solution(2 x httpd, 1 x application server and 1 x Sybase RDBMS). Beside that all that other stuff one is using. This will for shure produce a lot of cache misses.
Did you really try out everything so far, even reboot(BIOS check about C&Q)?
cu,
Michael
p.s. My long time expirience tells me, that even if the mathematical chances are pretty low, there is a reasonable amout of situations where two errors hit you at the same time. E.g. one can change an app and at the same time get some kind of harware error. Shurely, in case of problems the first thought is about the new app and that sometimes makes it very hard to find the real origin.
LOL, I've downloaded a couple of these documents too, during the last days, because I want to refresh my assembler knowledges.
Very interesting stuff(about the K10 too).
Yeah, brings back memories as well, like juggling with the few registers the 486 had ... a long time ago for me.
Oh what a damn modern thing. ;-)
I Started with an SC/MP microprocessor(self building project of the magazine Elector) :-)
Later on I also wrote assembler routines for an Epson HX 20. I'd never want to miss these exciting times.
Btw, at that time I always dreamed about an 68000 developer board, but it was just too expensive.
p.s. My long time expirience tells me, that even if the mathematical chances are pretty low, there is a reasonable amout of situations where two errors hit you at the same time. E.g. one can change an app and at the same time get some kind of harware error. Shurely, in case of problems the first thought is about the new app and that sometimes makes it very hard to find the real origin.
Good point. If it's the machine and not the app, the BOINC benchmark should now show a somewhat lower performance, shouldn't it. I mean not that the benchmark in itself is meaningful, but it should show a degradation on Annika's PC if it's a problem that wasn't there a few days ago? Might be worth trying.
About the 486 being a modern thing :-)...well I started assembler on a 6502 with 3,5 KB of available RAM, I just left it behind on a 486 :-)
p.s. My long time expirience tells me, that even if the mathematical chances are pretty low, there is a reasonable amout of situations where two errors hit you at the same time. E.g. one can change an app and at the same time get some kind of harware error. Shurely, in case of problems the first thought is about the new app and that sometimes makes it very hard to find the real origin.
Good point. If it's the machine and not the app, the BOINC benchmark should now show a somewhat lower performance, shouldn't it. I mean not that the benchmark in itself is maningful, but it should show a degradation on Annika's PC if it's a problem that wasn't there a few days ago? Might be worth trying.
Good idea and shurely worth trying. I only hope the CPU isn't throttling, then the benchmark might wake it up, which the einstein app is not able to do.
Quote:
About the 486 being a modern thing :-)...well I started assembler on a 6502 with 3,5 KB of available RAM, I just left it behind on a 486 :-)
Nice cpu. ;)
I started with 256B of RAM and dip switches for setting addresses and data. The system was later on expanded more and more, afair to 24kB RAM and 8kB EPROM. At the beginning, the parts for a 4kB expansion card cost about 495 DM, later a 16KB RAM / 8kB EPROM parts set was about 295 DM. Maybe I swapped the prices, it's too long ago. :)
After that HX 20 came the first 80286@20MHz and then.........
So go ahead and think about what we get in 30 years(don't know if I live any more then) and what it will cost. I don't believe anybody can really imagine.
Just rebooted, Cool and Quiet is off, CPU has a perfectly normal temperature and the benchmarks are just the same as they always were (too low, of course, but that is because I'm not running the latest BOINC client- talk about that, you think the client version might be of interest here?).
Just rebooted, Cool and Quiet is off, CPU has a perfectly normal temperature and the benchmarks are just the same as they always were (too low, of course, but that is because I'm not running the latest BOINC client- talk about that, you think the client version might be of interest here?).
Boinc Client doesn't matter, benchmark is ok, considering the older core client.
In case progress doesn't accelerate now and top shows constantly > 98%, there must be another place where the cpu cycles burn around. ;)
The WU should not be the reason, because your wingman has finished it in regular time.
I would use a degugger(man gdb) to find out which parts of the code are mostly in work. You may have a look in our forum thread, where Daran and me spoke about such things(http://www.heise.de/foren/go.shtml?read=1&msg_id=12702507&forum_id=74338). Maybe there is a different place, where the CPU is losing its time.
If Bernd Machenschalk is interested, you can make a backup of your boinc folder and send it to him, so he can run it in a VM under the same OS.
But first of all I would run memtest and have a look at the L1/L2 cache and memory speed.
The first WU has completed successfully on my host, a A64 3700+ running Ubuntu 6.06.
The run time is unchanged from the previous app.
The WU hasn't been uploaded yet, but my hast has connected the server and there was no error.
Thanks Bernd :-)
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman
I would like to attach an
)
I would like to attach an Intel with little cache to check how that does, but I can't get my hands on a such just now... certainly not one I could let run Linux.
RE: RE: and I think we
)
LOL, I've downloaded a couple of these documents too, during the last days, because I want to refresh my assembler knowledges.
Very interesting stuff(about the K10 too).
cu,
Michael
RE: LOL, I've downloaded a
)
Yeah, brings back memories as well, like juggling with the few registers the 486 had ... a long time ago for me.
@Annika:
I've got a Dual CPU PIII Coppermine (256KB L2 at core spped), but it happened to receive two >500 credit WUs a short time ago, so it will be busy for some time :-(
I've now put my AMD Athlon XP 1800+ (Palomino), nicknamed "the heater" :-) on the beta test. We'll see...
CU
BRM
RE: I would like to attach
)
Forget about the cache, this Athlon XP 2600+(Thoroughbred B) has just 256kB L2 cache and is doing fine without any errors with 4.18 and 4.21(the last result). My X2 has 2 · 1MB L2 cache(2GB RAM) and is running 3 VMWare sessions, two of them dedicated crunchers, one is home for a complete InterShop solution(2 x httpd, 1 x application server and 1 x Sybase RDBMS). Beside that all that other stuff one is using. This will for shure produce a lot of cache misses.
Did you really try out everything so far, even reboot(BIOS check about C&Q)?
cu,
Michael
p.s. My long time expirience tells me, that even if the mathematical chances are pretty low, there is a reasonable amout of situations where two errors hit you at the same time. E.g. one can change an app and at the same time get some kind of harware error. Shurely, in case of problems the first thought is about the new app and that sometimes makes it very hard to find the real origin.
RE: RE: LOL, I've
)
Oh what a damn modern thing. ;-)
I Started with an SC/MP microprocessor(self building project of the magazine Elector) :-)
Later on I also wrote assembler routines for an Epson HX 20. I'd never want to miss these exciting times.
Btw, at that time I always dreamed about an 68000 developer board, but it was just too expensive.
cu,
Michael
RE: p.s. My long time
)
Good point. If it's the machine and not the app, the BOINC benchmark should now show a somewhat lower performance, shouldn't it. I mean not that the benchmark in itself is meaningful, but it should show a degradation on Annika's PC if it's a problem that wasn't there a few days ago? Might be worth trying.
About the 486 being a modern thing :-)...well I started assembler on a 6502 with 3,5 KB of available RAM, I just left it behind on a 486 :-)
CU
BRM
RE: RE: p.s. My long
)
Good idea and shurely worth trying. I only hope the CPU isn't throttling, then the benchmark might wake it up, which the einstein app is not able to do.
Nice cpu. ;)
I started with 256B of RAM and dip switches for setting addresses and data. The system was later on expanded more and more, afair to 24kB RAM and 8kB EPROM. At the beginning, the parts for a 4kB expansion card cost about 495 DM, later a 16KB RAM / 8kB EPROM parts set was about 295 DM. Maybe I swapped the prices, it's too long ago. :)
After that HX 20 came the first 80286@20MHz and then.........
So go ahead and think about what we get in 30 years(don't know if I live any more then) and what it will cost. I don't believe anybody can really imagine.
cu,
Michael
Just rebooted, Cool and Quiet
)
Just rebooted, Cool and Quiet is off, CPU has a perfectly normal temperature and the benchmarks are just the same as they always were (too low, of course, but that is because I'm not running the latest BOINC client- talk about that, you think the client version might be of interest here?).
RE: Just rebooted, Cool and
)
Boinc Client doesn't matter, benchmark is ok, considering the older core client.
In case progress doesn't accelerate now and top shows constantly > 98%, there must be another place where the cpu cycles burn around. ;)
The WU should not be the reason, because your wingman has finished it in regular time.
I would use a degugger(man gdb) to find out which parts of the code are mostly in work. You may have a look in our forum thread, where Daran and me spoke about such things(http://www.heise.de/foren/go.shtml?read=1&msg_id=12702507&forum_id=74338). Maybe there is a different place, where the CPU is losing its time.
If Bernd Machenschalk is interested, you can make a backup of your boinc folder and send it to him, so he can run it in a VM under the same OS.
But first of all I would run memtest and have a look at the L1/L2 cache and memory speed.
cu,
Michael
The first WU has completed
)
The first WU has completed successfully on my host, a A64 3700+ running Ubuntu 6.06.
The run time is unchanged from the previous app.
The WU hasn't been uploaded yet, but my hast has connected the server and there was no error.
Thanks Bernd :-)
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman