Author: Jeremiah Penery
Date: 08:04:57 05/07/02
Go up one level in this thread
I tried to write this post like 4 times, and I kept doing stupid stuff, like going to a new webpage, and losing the post. Blah. On May 07, 2002 at 07:05:16, Aaron Gordon wrote: >I suppose people just don't read my posts, if so, it's just not sinking in. You and Slater both accused me of not reading your posts. Maybe you guys are the ones not reading mine. :) >This >will be the third or fourth time I've said this. Those benchmarks and crap you >see.. the Athlon is NOT OPTIMIZED. Look at it again. Let it soak in. Yes, lets >go ahead and reread that one more time. THE ATHLON IS NOT OPTIMIZED IN THOSE >TESTS. My memory is about the same bandwidth at 150/300DDR as a P4 but with much >lower latency. Without the bios settings it is *2* times slower. >Here's a screenshot for you. Mine is the topmost result. > >ftp://speedycpu.dyndns.org/pub/overclockstuff/pics/athlonxp/sisoftmem.jpg I know all the BIOS settings and tweaks. I know the Sandra numbers, but I don't think that particular test reflects real-world performance in a measurable way. Once you saturate the CPU bus with data, an increase in bandwidth doesn't really matter. The Athlon's CPU bus has a peak theoretical transfer rate of 2.1GB/s at 133MHz. If you run the FSB up to 150MHz, it only goes up to 2.4GB/s. Running asynchronously with a memory bandwidth of 2800 (on Sandra's test) isn't going to give a noticable performance increase. >This as far as tweaking goes is considered slow & still matches & due to being >much faster latency wise, beats the P4. If you read up a bit you can buy a >motherboard that can run 166(333DDR) or 200(400DDR)fsb with stock PCI/AGP speeds >which means you really aren't overclocking anything (KT266a/333 can do this with >absolutely NO problems). Just pick up the proper ram (PC2700/3000/3200) & punch >it up to 166 or 200fsb and you're good to go. You'd score around 2,800 in that >particular test if not a little more. If you do it that way, the memory speed is asynchronous to the CPU bus speed - read the previous paragraph I wrote. If you actually push the FSB up to 200MHz (synchronous), the theoretical CPU peak bandwidth goes to 3.2GB/s. Unfortunately, this is not so easy to do - most components will choke at that kind of speed. By the way, on my stock-speed EpoX 8kHA+, I get about 2000 on that Sandra benchmark. >In Quake3 the FPS is massively increased with those options enabled in the bios >(Q3 is bandwidth hungry). Stock speeds my AthlonXP 1900+ (1.6GHz) beats a >P4-2.2GHz in Quake3 by a large margin. At 1.8GHz/150(300DDR) a 2.6 gets blown to >pieces. If I use my AMD optimized DLL's the FPS is increased another 10-20%. Of >course if you have all the bios options disabled it'll be a bit slower (if you >use the DLL's it'll still be faster though, even with a ~500mhz disadvantage to http://www6.tomshardware.com/cpu/02q2/020506/p4b-06.html All other benchmark numbers for P4 vs. Athlon on Quake3 are fairly similar to this. I find it difficult to believe that _all_ hardware websites do not optimize for Athlon (even the AMD-biased websites). I don't think any kind of optimization can overcome that huge gap that you see. Maybe I'm wrong - I don't care a whole lot about Quake3 performance anyway. :) >a P4). You can test this for yourself. Make sure you get a quality board (only >$80-100) and enable 4 way interleaving, 4/8k page blocks, set the CL latency to >2T instead of 3T (there will be another option for 5/6T, put it to 5T). There >will be other options. They vary from board to board.. such as fast cpu command >decode, cpu drive strength, USCW or WC AGP options, 4x agp enabled, 128 or more >AGP aperture size, etc. Yes, yes, I have of course done all those things. :) >I'm looking forward to the tests, Slate. Also Slate, you must have done >something wrong with your 1.73GHz AthlonXP testing because it should mame that >P4 in crafty. Take a look at some benchmarks of various Tbird/AthlonXP speeds vs >different speed P4's. The AthlonXP's use AthlonXP binaries, P4 used P4, etc. >http://speedycpu.dyndns.org/crafty/craftybench.jpg > >Even a dinky Tbird 1GHz @ 1.68 beats a P4 Northwood at almost 2.5GHz not to >mention that P4 was running 137fsb(548MHz QDR). I'm not sure how you tested that >AthlonXP.. but having 12 divx movies playing while benchmarking is generally a >BAD thing. ;) I totally agree on this. The AthlonXP should blow away the P4 on Crafty.
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.