Author: Robert Hyatt
Date: 10:31:59 01/31/03
Go up one level in this thread
On January 31, 2003 at 02:49:03, Matt Taylor wrote: >On January 30, 2003 at 14:24:27, Robert Hyatt wrote: > >>On January 30, 2003 at 13:28:42, Christopher A. Morgan wrote: >> >>>Question re dual processor differences. >>> >>>Tiger Direct offers a dual AMD MP 2200+ for $1,650, and a dual Intel Xeon 2.4 >>>for $3,000. Both without an operating system. >>> >>>What difference in performance would I expect between these two machines? The >>>AMD dual at roughly ½ the price seems to be the much better buy, although the >>>Xeon should have HT, I believe? >>> >>>For a Windows operating system which is better (I am a single user, no network >>>use, no server use), Win XP Pro or Win 2000 Pro? >>> >>>Thanks! >> >> >>there are plenty of people here that can give you a good performance comparison >>between the two. From what I have seen, AMD generally has the performance edge >>until you step into the dual market. Then you have to be very careful as several >>AMD >>tests posted here by others (not by me as I have no AMD boxes here at all) >>suggest that >>the AMD duals have a memory bottleneck that limits performance. However, it is >>also likely that there are both good and bad chipsets for supporting duals. >>Intel has >>a "workstation" class dual xeon chipset and a "server" class chipset. The >>server class >>chipset has better memory performance. >> >>It it were _my_ money, I would benchmark the program(s) I want to run on the box >>before making the decision. As I said, there are certainly bad AMD chipsets for >>duals. There are _also_ bad Intel chipsets for duals. >> >>Common sense says "benchmark" or get data directly from someone that has the >>_specific_ chipset you are looking at. >> >>All duals are not created equal. Hyper-threading is yet another issue. >>Remember >>that you have to run two threads to take full advantage of one physical CPU. On >>AMD >>this is not true. Hyper-threading speeds things up significantly. But that >>second thread >>also has a cost, particularly if you don't have a second thread to run. :) > >There are only 2 dual-CPU chipsets for Socket A. They are the AMD 760 and AMD >760MPX. Both of my duals are based on the 760MPX. There may be performance >difference; the only listed difference I could find was the memory footprint. > >I posted my numbers a while back. My AthlonMP 2000 system had something like 1.5 >MN/sec. My AthlonMP 1600 had something like 1.1 MN/sec. IIRC, your dual Xeon 2.8 >GHz was ~2.1 MN/sec, yes? Extrapolating to AthlonMP 2400, it would not quite >match 2.1 MN/sec, though it would be close and -much- cheaper. > >Just like you, I believe the issue is a memory bottleneck. The SMP chipset >squanders valuable bus cycles. Intel's chipsets use interleaving to give each >processor "dedicated" bandwidth. > >-Matt the 2.1M depends on the test set, of course. And other things as well. I really hate to compare even Crafty NPS numbers across machines, unless it is done one very specific way: default settings on _everything_, except for the mt=N to enable extra cpus, and using the "bench" command. That makes sure everyone runs the same positions, with the same size hash, and it needs to be the _same_ version of Crafty as there is variance between versions for lots of reasons. I'll try to run the bench test. OK just did. 2.134M nodes per second (2,134K nodes per second if you prefer K). That is run using the intel compiler in feedback optimizer mode, dual 2.8 xeon with mt=4 and hyperthreading on... Bob
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.