Message boards : Number crunching : Dual Xeons
Author | Message |
---|---|
Dragokatzov Send message Joined: 5 Oct 05 Posts: 25 Credit: 2,446,376 RAC: 0 |
I see some Dual Xeons with hyperthreading going cheap on ebay. One of them that I am interested in only has 512 megs of ram. for ram, ot has RAMBUS 800-45 ECC, I would like to add more, but its going to be like finding a needle in a haystack. do you think it would be ok to run 4 instances of Rosetta with only 512 megs of ram? Thanks! Victory is the ONLY option! |
Ethan Volunteer moderator Send message Joined: 22 Aug 05 Posts: 286 Credit: 9,304,700 RAC: 0 |
It will run fine. . it would run better with more memory, but if it's a dual xeon it probably has a 10k scsi (or better), so scratch would be faster than on a normal machine. |
Webmaster Yoda Send message Joined: 17 Sep 05 Posts: 161 Credit: 162,253 RAC: 0 |
Do you think it would be ok to run 4 instances of Rosetta with only 512 megs of ram? Thanks! Figures vary, but it's not unusual for Rosetta to use 160MB per WU. The system would need 640MB plus whatever the O/S needs. With only 512MB, there will probably be so much swapping to disk that it's not worth running with HT on. If it had 1GB, it might be OK. *** Join BOINC@Australia today *** |
dgnuff Send message Joined: 1 Nov 05 Posts: 350 Credit: 24,773,605 RAC: 0 |
It will run fine. . it would run better with more memory, but if it's a dual xeon it probably has a 10k scsi (or better), so scratch would be faster than on a normal machine. I'd go along with this call. The one thing I would suggest is that you don't attach to too many projects. Two, or at the absolute most three. I'm tempted to suggest that you might want to avoid CPDN, because I believe (but am willing to be corrected if wrong) that it has a fairly heavy memory footprint. Anyone want to set the record straight on CPDN's memory needs? Figures vary, but it's not unusual for Rosetta to use 160MB per WU. The system would need 640MB plus whatever the O/S needs. That's 160 MB of VIRTUAL memory, and FWIW it's headed for 200 MB on my XP machine here. However the working set size is around 70M, so that's a far more accurate measure of the physical chip it'll use. Four of those ought to clock in at about 280 MB, so you still have 200Mb and change left for everything else. Bottom line. It'll be cosy, but it probably won't thrash. |
Paul D. Buck Send message Joined: 17 Sep 05 Posts: 815 Credit: 1,812,737 RAC: 0 |
Though I have 1G RAM, there is nearly 300K free physical I have 11 BOINC processes running or suspended in memory. Sulfur is showing 54M Three Rosetta are showing 70M to 80M each BOINC View is 100M ... |
River~~ Send message Joined: 15 Dec 05 Posts: 761 Credit: 285,578 RAC: 0 |
I see some Dual Xeons with hyperthreading going cheap on ebay. One of them that I am interested in only has 512 megs of ram. for ram, ot has RAMBUS 800-45 ECC, A couple of points to note. Firstly, if you limit to 2 processes it doesn't matter much if you have HT turned on or off. Sure with HT on the Task Manager will say you are wasting 50%, but that is becasue it measures by clock time not CPU resource. Second, how much RAM it needs before the virtual memory swaps become so bad that they outweigh the advantage is almost impossible to say. Sorry Paul, but the working set size does not tell you a lot. The issue is whether the disk accesses for VM slow down the occasional page fetches which would come from RAM if there was more, and to know that you need to know how many of the page faults come from the same or from different places. So my advice is to run BOINC for a few days with HT on, setting max processors = 4 in prefs, then for a few days with = 3, then for a few days with =2 and see how they compare. Running with 3 cpus means in effect that HT is in use on one core but not on the other. On whichever core is running just one task you get the full speed as if HT was not running. This setting will mean you get more jitter in the run times of results, but on Rosetta who's going to notice that ;-) |
Message boards :
Number crunching :
Dual Xeons
©2025 University of Washington
https://www.bakerlab.org