BTW, do you have plans to try Linux on your Intel Atom? It would be cool to try the x87 vs SSE vs SSE2 variants there. I guess the in-order execution of the Atom is really, really bad for the E@H app because it won't make use of all the tricks and quirks (especially from your latest "interleaved loop" stuff) to keep the multiple execution units busy in parallel.
Bugger, just lost a whole lot of work units after updating to 4.49 SSE2.
After downloading 4.49 then downloading the SSE2 file and replacing the original 4.49 executable file, I changed it's name and restarted Boinc.
Can you see what is wrong, yes that's right, I forgot to set the executable attribute and lost all the work.
Have now set that attribute and awaiting Boinc to download new work, this may take awhile as Boinc Manager has decided that Einstein has to wait and downloaded a heap of Rosetta WU's instead, probably due to the number of Einstein failures.
I will wait to see if it all works before trying again on the other computers.
BTW, do you have plans to try Linux on your Intel Atom? It would be cool to try the x87 vs SSE vs SSE2 variants there. I guess the in-order execution of the Atom is really, really bad for the E@H app because it won't make use of all the tricks and quirks (especially from your latest "interleaved loop" stuff) to keep the multiple execution units busy in parallel.
CU
Bikeman
Sorry, but what is an Intel Atom? Sounds interesting.
Bugger, just lost a whole lot of work units after updating to 4.49 SSE2.
After downloading 4.49 then downloading the SSE2 file and replacing the original 4.49 executable file, I changed it's name and restarted Boinc.
Can you see what is wrong, yes that's right, I forgot to set the executable attribute and lost all the work.
Have now set that attribute and awaiting Boinc to download new work, this may take awhile as Boinc Manager has decided that Einstein has to wait and downloaded a heap of Rosetta WU's instead, probably due to the number of Einstein failures.
I will wait to see if it all works before trying again on the other computers.
Well I am impressed with the 4.49 SSE2 application. I have now returned 3 work units and all have been done in under 5 hours, which I had not done before.
Previous best for a trough result was 18,900 seconds, my best now is 16,788 seconds and I have not reached the trough yet, only halfway there. This is about 11% improvement on 4.35, with more to come.
Have an AMD Opteron 285 @2.6GHz, Linux Fedora Core 3.
Sorry, but what is an Intel Atom? Sounds interesting.
It's Intel's low-power consumption CPU series intended for (among other things) smart-phones, sub-notebooks, palmtops, PDAs ... anything that should have decent performance but must not consume a lot of power. Akos has a desktop system with an Atom CPU.
The interesting thing about this is that it is 100% software compatible with the current line of x86 CPUs, so the existing software should work. But the CPU has a completely new microarchitecture. In essence, some of the sophisticated performance optimizations that were implemented since the 486-era (speculative execution, out of order execution, register renaming...) were stripped away, so the Atom is a much more simplified design compared to modern cores (or even the P III). Number one priority is low wattage.
BTW, do you have plans to try Linux on your Intel Atom? It would be cool to try the x87 vs SSE vs SSE2 variants there. I guess the in-order execution of the Atom is really, really bad for the E@H app because it won't make use of all the tricks and quirks (especially from your latest "interleaved loop" stuff) to keep the multiple execution units busy in parallel.
CU
Bikeman
Sorry, but what is an Intel Atom? Sounds interesting.
Should be a low power consuming processor, useful in laptops, but it is described as a poor cruncher.
Tullio
It's Intel's low-power consumption CPU series intended for (among other things) smart-phones, sub-notebooks, palmtops, PDAs ... anything that should have decent performance but must not consume a lot of power. Akos has a desktop system with an Atom CPU.
I may add that the northbridge Intel made for Atom is very poor, fabbed on an ancient 130nm process, making it too big (and power hungry?) for handheld devices. The Atom itself is made on the excellent 45nm High-K/metalgate process. When they shrink the northbridge to, or at least closer to, 45nm it will be useful in smaller devices like handheld GPS and cellphones. Will be great, cellphones and GPS running Linux, thats sounds quite sexy. For now the smallest Intel dare marketing it for is handheld internet devices.
AFAIK, there will be two lines of Atom, the faster ones and dualcores for notebooks and laptop. The results from Akosf so far doesnt give me much confidence that it will be that great in laptops, the in-order execution is a serious handicap, but who knows, lets see how high they can clock them with a still respectable power consumption. I know the Intel 45nm process has lots of juice in it.
Level 1: $Id: LocalComputeFstatHoughMap.c,v 1.21 2008/02/28 09:40:15 bema Exp $
Status code 5: Null pointer
function LocalComputeFstatHoughMap, file /home/bema/einsteinathome/HierarchicalSearch/EaH_build_release_einstein_S5R3_4.49_1/extra_sources/lalapps-CVS/src/pulsar/hough/src2/LocalComputeFstatHoughMap.c, line 369
2008-05-30 16:06:09.3973 [CRITICAL]: BOINC_LAL_ErrHand(): now calling boinc_finish()
called boinc_finish
##########################################################################
App is the SSE2 version.
cu,
Michael
Hm. Nothing special about these tasks. The machine still occasionally delivers succesfull and valid results, and the other tasks of the same workunits where your machine throws the errors have been completed successfully, too. Looks like a transient problem on the machine, e.g. temperature related. Weren't this the hottest days of the year so far? Well, that sounds like what a "burn in" test is done for, isn't it?
RE: You are right,
)
:-), nooo, apprentice I'd say.
BTW, do you have plans to try Linux on your Intel Atom? It would be cool to try the x87 vs SSE vs SSE2 variants there. I guess the in-order execution of the Atom is really, really bad for the E@H app because it won't make use of all the tricks and quirks (especially from your latest "interleaved loop" stuff) to keep the multiple execution units busy in parallel.
CU
Bikeman
Bugger, just lost a whole lot
)
Bugger, just lost a whole lot of work units after updating to 4.49 SSE2.
After downloading 4.49 then downloading the SSE2 file and replacing the original 4.49 executable file, I changed it's name and restarted Boinc.
Can you see what is wrong, yes that's right, I forgot to set the executable attribute and lost all the work.
Have now set that attribute and awaiting Boinc to download new work, this may take awhile as Boinc Manager has decided that Einstein has to wait and downloaded a heap of Rosetta WU's instead, probably due to the number of Einstein failures.
I will wait to see if it all works before trying again on the other computers.
RE: RE: You are right,
)
Sorry, but what is an Intel Atom? Sounds interesting.
RE: Bugger, just lost a
)
Well I am impressed with the 4.49 SSE2 application. I have now returned 3 work units and all have been done in under 5 hours, which I had not done before.
Previous best for a trough result was 18,900 seconds, my best now is 16,788 seconds and I have not reached the trough yet, only halfway there. This is about 11% improvement on 4.35, with more to come.
Have an AMD Opteron 285 @2.6GHz, Linux Fedora Core 3.
RE: Sorry, but what is an
)
It's Intel's low-power consumption CPU series intended for (among other things) smart-phones, sub-notebooks, palmtops, PDAs ... anything that should have decent performance but must not consume a lot of power. Akos has a desktop system with an Atom CPU.
The interesting thing about this is that it is 100% software compatible with the current line of x86 CPUs, so the existing software should work. But the CPU has a completely new microarchitecture. In essence, some of the sophisticated performance optimizations that were implemented since the 486-era (speculative execution, out of order execution, register renaming...) were stripped away, so the Atom is a much more simplified design compared to modern cores (or even the P III). Number one priority is low wattage.
CU
Bikeman
RE: RE: RE: You are
)
Should be a low power consuming processor, useful in laptops, but it is described as a poor cruncher.
Tullio
RE: It's Intel's low-power
)
I may add that the northbridge Intel made for Atom is very poor, fabbed on an ancient 130nm process, making it too big (and power hungry?) for handheld devices. The Atom itself is made on the excellent 45nm High-K/metalgate process. When they shrink the northbridge to, or at least closer to, 45nm it will be useful in smaller devices like handheld GPS and cellphones. Will be great, cellphones and GPS running Linux, thats sounds quite sexy. For now the smallest Intel dare marketing it for is handheld internet devices.
AFAIK, there will be two lines of Atom, the faster ones and dualcores for notebooks and laptop. The results from Akosf so far doesnt give me much confidence that it will be that great in laptops, the in-order execution is a serious handicap, but who knows, lets see how high they can clock them with a still respectable power consumption. I know the Intel 45nm process has lots of juice in it.
Team Philippines
Anyone running this Beta with
)
Anyone running this Beta with an old Client such as 4.43 (or willing to test)? See this thread.
BM
BM
I suddenly get a lot of
)
I suddenly get a lot of errors.
Host is a root server hosted at a provider running E@H as a kind of burn in test.
http://einsteinathome.org/host/1259302/tasks&offset=100
Error example:
##########################################################################
757, 758, 759, 760, 761, 762, 763, Error[1] 5: function LocalComputeFstatHoughMap, file /home/bema/einsteinathome/HierarchicalSearch/EaH_build_release_einstein_S5R3_4.49_1/extra_sources/lalapps-CVS/src/pulsar/hough/src2/LocalComputeFstatHoughMap.c, line 369, $Id: LocalComputeFstatHoughMap.c,v 1.21 2008/02/28 09:40:15 bema Exp $
ABORT: Null pointer
Level 0: $Id: HierarchicalSearch.c,v 1.192 2008/02/01 01:45:43 bema Exp $
Function call `COMPUTEFSTATHOUGHMAP ( &status, &semiCohCandList, &pgV, &semiCohPar)' failed.
file /home/bema/einsteinathome/HierarchicalSearch/EaH_build_release_einstein_S5R3_4.49_1/extra_sources/lalapps-CVS/src/pulsar/hough/src2/HierarchicalSearch.c, line 1113
Level 1: $Id: LocalComputeFstatHoughMap.c,v 1.21 2008/02/28 09:40:15 bema Exp $
Status code 5: Null pointer
function LocalComputeFstatHoughMap, file /home/bema/einsteinathome/HierarchicalSearch/EaH_build_release_einstein_S5R3_4.49_1/extra_sources/lalapps-CVS/src/pulsar/hough/src2/LocalComputeFstatHoughMap.c, line 369
2008-05-30 16:06:09.3973 [CRITICAL]: BOINC_LAL_ErrHand(): now calling boinc_finish()
called boinc_finish
##########################################################################
App is the SSE2 version.
cu,
Michael
RE: I suddenly get a lot of
)
Hm. Nothing special about these tasks. The machine still occasionally delivers succesfull and valid results, and the other tasks of the same workunits where your machine throws the errors have been completed successfully, too. Looks like a transient problem on the machine, e.g. temperature related. Weren't this the hottest days of the year so far? Well, that sounds like what a "burn in" test is done for, isn't it?
BM
BM