Message boards :
Number crunching :
Very Long Crunching Time on Celeron (Coppermine)
Message board moderation
Author | Message |
---|---|
Send message Joined: 22 Feb 10 Posts: 4 Credit: 3,576,829 RAC: 0 |
Hello! I'm set up my ancient Celeron 800 (Coppermine, Family 6, Model 8, Stepping A) box with Windows 98 on it to play some of my old games that won't work on Windows 7, and found that the background crunching on MilkWay@Home is taking forever! At the moment it's been working on a single separation unit for just over 72 hours and it's currently 44.5% complete. That doesn't seem to make sense when you compare it to my i3 330M (Arrandale, Family 6, Model 5, Stepping 2) laptop, which manages to get through a similar workunit in around 4 1/2 to 5 hours. If they were proportional based on the MIPS, the older processor would be getting through them in well under 50 hours, not 3 times that. Is there SSE2/3/4 optimisation in the application that my older box just can't take advantage of, or is it something else? |
Send message Joined: 8 May 10 Posts: 576 Credit: 15,979,383 RAC: 0 |
Hello! I'm set up my ancient Celeron 800 (Coppermine, Family 6, Model 8, Stepping A) box with Windows 98 on it to play some of my old games that won't work on Windows 7, and found that the background crunching on MilkWay@Home is taking forever! At the moment it's been working on a single separation unit for just over 72 hours and it's currently 44.5% complete. That doesn't seem to make sense when you compare it to my i3 330M (Arrandale, Family 6, Model 5, Stepping 2) laptop, which manages to get through a similar workunit in around 4 1/2 to 5 hours. If they were proportional based on the MIPS, the older processor would be getting through them in well under 50 hours, not 3 times that.It makes sense. It's ancient. The architectural improvements since then have been enormous. Is there SSE2/3/4 optimisation in the application that my older box just can't take advantage of, or is it something else?Coppermine was the first to have just SSE I think; it certainly doesn't have SSE2 required for doubles. The current application will use SSE2/SSE3/SSE4.1 if it's available already; the Arrandale you have is using the SSE3 stuff already. |
Send message Joined: 22 Feb 10 Posts: 4 Credit: 3,576,829 RAC: 0 |
Hello! I'm set up my ancient Celeron 800 (Coppermine, Family 6, Model 8, Stepping A) box with Windows 98 on it to play some of my old games that won't work on Windows 7, and found that the background crunching on MilkWay@Home is taking forever! At the moment it's been working on a single separation unit for just over 72 hours and it's currently 44.5% complete. That doesn't seem to make sense when you compare it to my i3 330M (Arrandale, Family 6, Model 5, Stepping 2) laptop, which manages to get through a similar workunit in around 4 1/2 to 5 hours. If they were proportional based on the MIPS, the older processor would be getting through them in well under 50 hours, not 3 times that.It makes sense. It's ancient. The architectural improvements since then have been enormous. Oh, I know about their capabilities (the Arrandale has SSE4.1/4.2 as well, I think). I was just wondering what might account for the discrepancies between the times and if that might be why! Thanks for the info, it was getting me worried that something was set up wrong in the older box. What seemed especially odd was that if you take into account the Whetstone and Dhrystone MIPS (i.e. the raw performance) was the sheer increase in speed that the newer instruction sets appeared to be giving the modern processor. |
Send message Joined: 19 Jul 10 Posts: 589 Credit: 18,926,785 RAC: 4,539 |
I had the same supprising discrepancy in performance when I tried an old Athlon C @ 1050 MHz on Seti. While my AthlonXP 2000+ @ 1666 MHz (which is not that much younger) makes a "normal" Seti WU in about 11 hours, the Athlon C needed 34-35 hours for that. Both were using optimized apps, however the Athlon C only the MMX one, so part of the speed diference might go on that. |
Send message Joined: 8 May 09 Posts: 3319 Credit: 520,332,607 RAC: 21,340 |
Hello! I'm set up my ancient Celeron 800 (Coppermine, Family 6, Model 8, Stepping A) box with Windows 98 on it to play some of my old games that won't work on Windows 7, and found that the background crunching on MilkWay@Home is taking forever! At the moment it's been working on a single separation unit for just over 72 hours and it's currently 44.5% complete. That doesn't seem to make sense when you compare it to my i3 330M (Arrandale, Family 6, Model 5, Stepping 2) laptop, which manages to get through a similar workunit in around 4 1/2 to 5 hours. If they were proportional based on the MIPS, the older processor would be getting through them in well under 50 hours, not 3 times that. It is a Celeron, meaning small L2 cache, meaning the whole workunit won't fit into the L2 cache meaning TONS AND TONS of swapping bits back and forth so the unit can be crunched! That is part of why some projects can't use gpu's yet, the workunits are just too big. |
Send message Joined: 20 Sep 08 Posts: 1391 Credit: 203,563,566 RAC: 0 |
Mikey is correct. Celerons are cut down cheaper versions of the full blown Pentium chips. It stems back from the old days where we had 80486SX and DX chips where the SX had the co-processor disabled. They are not really up the the task of crunching. |
Send message Joined: 22 Feb 10 Posts: 4 Credit: 3,576,829 RAC: 0 |
Hello! I'm set up my ancient Celeron 800 (Coppermine, Family 6, Model 8, Stepping A) box with Windows 98 on it to play some of my old games that won't work on Windows 7, and found that the background crunching on MilkWay@Home is taking forever! At the moment it's been working on a single separation unit for just over 72 hours and it's currently 44.5% complete. That doesn't seem to make sense when you compare it to my i3 330M (Arrandale, Family 6, Model 5, Stepping 2) laptop, which manages to get through a similar workunit in around 4 1/2 to 5 hours. If they were proportional based on the MIPS, the older processor would be getting through them in well under 50 hours, not 3 times that. Ah, I think this was the problem - I'd forgetten about the differences in cache size. I was especially confused when I benchmarked a Pentium II 450 this morning and found it was actually making better progress than the Celeron 800 even without SSE! The Celeron's 128K L2 is just too small for this sort of thing; the 512K of the Pentium II more than made up for the lower MIPS performance. |
Send message Joined: 1 Sep 08 Posts: 204 Credit: 219,354,537 RAC: 0 |
Raw MIPS are very unrelated to real world performance :p MrS Scanning for our furry friends since Jan 2002 |
Send message Joined: 27 Apr 10 Posts: 35 Credit: 90,828,595 RAC: 0 |
Celeron I have discovered your problem. You may as well just short out one of your wall sockets, that would be more efficient use of power than crunching on this thing. |
Send message Joined: 22 Feb 10 Posts: 4 Credit: 3,576,829 RAC: 0 |
Celeron I only use it to play really old games, so I don't think it matters how crap it is at crunching to be honest. Thanks for the input everyone, it was just a mental brainfart on my part for forgetting about the cache. I also set the set it to write to disk only every half-hour, since it's been rock solid so far and isn't used often, which seems to have helped quite a bit also. |
Send message Joined: 8 May 09 Posts: 3319 Credit: 520,332,607 RAC: 21,340 |
Celeron One thing you might try is to lower your cache down to very low and try different projects, I see you have signed up for a bunch, and see which ones have the lowest workunit size that will work for you. Most projects used to be between the 128 and 256 meg size, making them bad for Celerons and Durons, etc. Also bump UP the ram in the machine to 2gb, if possible, and then set it to 'leave applications in memory while suspended'. MilkyWay for instance has an 8 minute cpu task, that can't be that big of a workunit! |
Send message Joined: 1 Sep 08 Posts: 204 Credit: 219,354,537 RAC: 0 |
That old Celeron wasn't that crippled. It's got 128 kB L2 instead of 256 kB for the full Coppermine, and an FSB of 100 MHz compared to 100 or 133 MHz for the Coppermine. Everything else is similar. Since it's only got SSE1 you might want to use it in a project which doesn't benefit from SSE2+. Otherwise it will always seem absymal compared to modern hardware. MrS Scanning for our furry friends since Jan 2002 |
©2024 Astroinformatics Group