Author: Aaron Gordon
Date: 06:36:22 05/08/02
Go up one level in this thread
On May 07, 2002 at 21:37:37, Jeremiah Penery wrote: >On May 07, 2002 at 12:16:30, Aaron Gordon wrote: > >>Sisoft sandra is fairly accurate when it comes to memory bandwidth. It's >>basically a souped up STREAM benchmark. Also, I notice a very large difference >>between 2.0gb/s and 2.4gb/s. This is 20%. In many applications that are memory >>intensive this gives a MUCH needed boost. > >My argument was that the Sandra memory bandwidth test doesn't reflect any kind >of real-world performance. I'm sure it's accurate as far as raw bandwidth, >under specific conditions. It comes very close to the theoretical peak >bandwidth, in fact - which is why I don't think it's completely realistic; very >few applications will saturate the bus like that. There are many applications and games that are heavily memory bandwidth dependant. >>Also, about running 166/200fsb synchronous. Like I said.. the 8k3a+ and other >>similar boards have 1/5, 1/6 PCI multipliers and similar AGP multipliers to keep >>everything within spec while keeping the bus and ram at 166 or 200MHz(333 & > >I was talking about running the CPU bus at the normal 133MHz and running the >memory at 166MHz (333 DDR) - _a_synchronous operation - it usually doesn't >produce much speedup, because the processor bus (still at 133MHz) is already >saturated, no matter if the memory bus is putting out 20GB/s. It's not always saturated. You'd need a RAID array blasting away with a highend Geforce4 in 4x agp mode to get it nice and bogged down. >>400DDR). Thats why I said you can just grab the board & slap in the appropriate >>memory and you're good to go. If you are at 133fsb(DDR) then ~2.0gb/s is about >>all you're going to get, the theoretical max would be 2.1gb/s. > >Yep. > >>About the Quake3 benchmark, yes, optimization can overcome that gap quite >>easily. Get Quake3 and run the tests (even at 133fsb w/ the tweaks you have >>now). I guarantee with a similar video card you will get (easily) %20 more fps >>than the AthlonXP on the page of a similar clock speed. No if's and's or but's >>about it. Use my DLL's and get ANOTHER 20% :) > >As I said, I'm not completely interested in Quake3 benchmarks, but maybe I'll >give it a try sometime. It's not like it matters once you're over about 100FPS >anyway, because the monitor doesn't refresh so fast. I use the fps in various timedemos in Quake3 to gauge how much fsb helps, how much memory speed helps & etc. It's definitely plenty of FPS to play with but I also use it for testing. >>Most of the speed increases come from increasing memory bandwidth. With the >>right board & memory you can get some ridiculous scores without having to modify >>anything on the board (just a few settings in the bios, nothing more). Dual >>channel DDR (2x faster than current DDR) is right around the corner.. hopefully > >Of course there has been an nForce chipset with dual-channel DDR support for >months. :) I'm talking about a QUALITY chipset. :) >>VIA puts out a decent chipset (like the KT266a & KT333) so we can take advantage >>of such insanity. :) > >Via is trying to push DDR400 on us, but I haven't heard much about them doing >any kind of dual-channel thing, which sucks. IMO, AMD should make their >processors use a 200MHz bus, instead of the 133 they currently use. Even >without increasing the speed of the processor, the performance would increase >quite a bit, especially with respect to the P4 on bandwidth-intensive >applications. It's a mystery why they keep everything on the old 133MHz bus - >it has to artificially hold the Athlon's performance down. Absolutely. Bus/memory speed has been a problem area for a very long time. I get a pretty large speed increase just going from 133 to 150... going from 133(266DDR) to 200(DC-DDR 800) will be crazy stuff. :) I can't wait.. especially for the Sledgehammer.
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.