Author: Vincent Diepeveen
Date: 10:27:51 07/03/05
Go up one level in this thread
On July 03, 2005 at 12:13:15, K. Burcham wrote: > >We know the actual switching for chess kns is in the transistor count. >Of course I am talking all things equal when making a comparison, optimizing, >bottlenecks, latency, cache size, etc... > >I have been curious about the transistor count of the new dual core processors. >I was not sure how many transistors were in each core. >In the X2 4800 dual core we can see that each core has 115 million transistors. >This would be about the same as the FX-57 transistor count. >This would explain why they are about the same in the bench tests with most >software. >Then with multi-processing software we can see a big jump in the bench tests >with the dual core. > >You can also see here how Intel has trouble getting the full benefits of the >transistor switching of the 169 million. The 3.73 Intel is not beating the FX-57 >at with 114 million transistors. You cannot compare AMD transistors with Intel. Intel is doing a lot of effort in order to produce the chips cheaper. Please realize very well that the real speed the cpu gives is based upon the L1 cache and the clock of it and all the penalties one and another can give (so the effective retire speed of the execution units and how deep it is pipelined). By pipelining deeper and having a weak L1 and tracecache, intel takes care they can cheaply produce the actual chip. However this is not the majority of the transistors. The vaste majority of the transistors is not the real cpu we care about, but the level 2 cache. A second optimization intel is doing there to produce it cheaper is by making the L2 cache more compact. Intels transistors have a far higher density than AMD. Intel can put 100 million transistors so to speak at the same area where AMD can only fit 50 million. The result of that is that the intel cpu's produce more heat at a smaller die area. The result of that is also you just can't compare a cheapo intel chip with an AMD chip. Additional the AMD chip has more layers than the P4. The opteron is a 9 layer processor. So the production cost of 1 A64/opteron is far higher than that of a P4, in general spoken. Further intel is producing their cpu's in bigger factories using 300 mm factories, versus amd uses 200 mm machines. So intel could, even if they would produce the same cpu, far cheaper. The advantage intel has from this is they make more profit a cpu and can produce more cpu's. At this moment AMD has problems producing enough cpu's for the total world market. In theory their production capacity start 2005 could never supply more than 30% of world demand of cpu's. So any comparision between amd and intel, other than how fast it runs your software, or how much profit they make, is very hard to do. AMD has designed a processor fast for most codes, intel has designed something as cheap as possible, and some compiler team (even when serving microsoft) then must make up for the problem of having bad cpu's. The compilers code must take care that specific test software (especially specint and specf type software) is fast at the intel cpu's and slow at amd. Another thing intel is far superior is in providing test machines to hardware sites. To get an AMD one you must fight hard. Logically most write positive reviews about intel cpu's, even if they are slow everywhere. So the transistorcount comparision doesn't make sense at all, as the space in the same technology that intel occupies is far smaller. >Transistor Count: > >Athlon 64 X2 4800+ >2 x 2400mhz core frequency >233 million > > >AMD Athlon 64 FX-57 >2800MHz >San Diego >114 million >90nm SGOI @ AMD, Fab 30, Dresden > > >AMD Athlon 64 FX-55 >2600MHz >Clawhammer >105.9 million >130nm SOI @ AMD, Fab 30, Dresden > > >Pentium 4 XE >3.73 GHz >169 million >90nm
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.