Crippled Athlon XP-2000 beats Northwood P4
Linux likes it and Windows adores it
I'd love to tell you how the Athlon-XP performs at its proper 1675MHz clock speed, but the new Gigabyte 7VTXE mobo we were sent was so severely hobbled that 1265 was all I could test Linux and Windows on reliably. (AMD declined our request to send a reference mobo of their choosing.)
I started with a clean install of SuSE 7.3. I knew I was scheduled for some pain when I noticed the words "Unknown VIA SouthBridge" scrolling quickly by as I booted the install DVD.
The installation died horribly, as I expected. So I dropped the bus frequency to 100MHz, popped in the most conservative CMOS params available, and tried again. This time I got SuSE 7.3 Pro (2.4.10 kernel) installed, but I couldn't enable DMA with hdparm. Well, I thought, the kernel update ought to take care of that. I needed to do a recompile anyway, as YaST somehow got the impression I was running an Intel P2. I took it to 2.4.17 and re-booted.
And of course DMA didn't work.
So I ran make config again, and using the shotgun approach enabled every IDE-related feature and driver available. I recompiled, re-booted, and got DMA working. (Sorry I can't tell you which IDE feature did the trick. You can pretty well bet I wasn't going to spend the entire weekend changing each one individually and rebuilding the kernel twenty times.)
So then I got the bright idea that with all the tricky stuff out of the way and the system running quite nicely, I might just try 133 for the bus speed.
That was a laugh. I tried six or seven different CMOS configs and kept rebooting -- to a preposterously unstable system. Apps would crash; I'd kill them, and then something else would crash, whack-a-mole-wise. After a cold boot, apps that had crashed before ran stably, and ones that had run stably before now crashed. There was no pattern; it was entirely too random to deal with.
So back to a crippled Athlon clocking 1265MHz. At least I'd got 32-bit I/O and DMA working, and 512MB of PC266 DDR SDRAM on board.
So, how does the AXP-2K/DDR system compare with a Northwood/RAMBUS setup running at nearly twice the clock speed? A lot better than half as well, that's for sure.
We'll look at the Linux performance first, then Windows. I think you're going to as be impressed as I was with this crippled machine.
As you may know, I have an aversion to standard benchmarks. They tell you that system A has better I/O through-put than system B. What they neglect to tell you is whether or not your machine is going to do the things you normally ask it to do any faster. I prefer to compare ordinary tasks, like copying files and directories, manipulating files, and launching and running applications in the same way that any user would do it.
So, using Linux, it took the Athlon DDR system 5.0 seconds to launch Mozilla on the first go, and 2.2 seconds to launch it on the second go, compared with 5.0 seconds and 2.5 seconds on the 'faster' 2.2GHz Intel RDRAM system. (DSL connection already established and target page already stored in browser cache.)
It took the AMD system 3.0 seconds to launch the Gimp on the first go and 1.0 seconds on the second go, compared with 2.5 and 1.2 on the Intel system. Perhaps, for some mysterious reason, RAM paging on the Athlon system is a bit better than on the Intel system.
Displaying a directory of 778MB containing 10,174 files in 'detailed list' mode with Krusader took 3.7 seconds on the AMD kit and 3.2 seconds on the Intel kit. Copying that directory to another location on the drive took 3 minutes, 44 seconds on the AMD kit and 3 minutes, 51 seconds on the Intel kit.
Displaying the directory with the Bash shell took 4.0 seconds on the first go and 1.6 seconds on the second go with the AMD kit, compared with 2.5 and 0.7 on the Intel kit. This suggests that the Athlon system does better than expected in GUI, but the video card and installed drivers were identical in both systems.
This is made less confusing when we compare the Quake3 benchmarks. Using identical settings, we got an average of 214.4 FPS with the Athlon and 304.7 with Northwood in this relatively CPU-intensive test.
Having the 100/133 Gigabyte debacle under my belt, I wisely installed Win-XP with the 100MHz setting. Now this is interesting. The most stable Windows ever built wouldn't even boot with the bus subsequently set at 133. Linux ran crappily, all right; but it did actually run.
Launching Mozilla in Win-XP Pro took 4.0 seconds the first time and 2.6 seconds the second time with the crippled Athlon, compared with 4.5 and 2.5 seconds on the Intel system.
The Gimp needed 3.2 seconds and 1.7 seconds respectively to launch on AMD, and 3.5 and 2.0 seconds on the Intel kit.
Opening a directory tree comprising 2.16GB with 30,562 files in 7 sub-directories took 18 seconds on the AMD kit (no Intel reference). Once this was accomplished, displaying the sub-directories in 'detailed list' mode took only a few tenths of a second.
Copying a directory of 778MB containing 10,174 files to a different location on the same disk took 3 minutes, 16 seconds on the crippled Athlon system and 3 minutes, 13 seconds on the Northwood system.
Listing the files with the DOS shell took 2.7 seconds on the AMD kit and 4.5 seconds on the Intel kit.
The Quake3 benchmark returned an average of 196.1 FPS on the AXP and 272.2 on the Northwood. Again, the processor's brute power is a major factor here.
So, what do we know? We know that the Northwood/RDRAM system runs almost twice as fast as our crippled Athlon test-system and costs almost twice as much. And it does deliver in situations where the CPU is stressed. But we're getting very good through-put with the AMD system (or piss poor through-put on the Intel system); and if we had a mobo that worked properly, we know the CPU gap would be narrowed quite a bit on both OS's. We also know that Windows likes the Athlon better than Linux does.
As it is, the Athlon system runs quite tolerably on Linux and Windows in spite of significant obstacles from Gigabyte's engineers. If one had a more 'mature' AXP/DDR mobo, and could run the Athlon at its full potential, one would probably never notice the difference in CPU clock speed with everyday tasks, and the money saved could go into something very sweet like a really powerful graphics accelerator. For gaming, the resulting AMD system would blow the doors off the Northwood system as we tested it. ®
Q3 benchmark settings:
Color depth: 16-bit
Geometric detail: low
Texture Detail: minimum
Texture quality: 16-bit
64M DDR GeForce AGP (2 yrs old) with Nvidia's latest drivers. AGP aperture 64MB (all systems).
Two Maxtor D740X 20G ATA-133 drives. SuSE Linux Pro 7.3 (patched) on master with ReiserFS; Win-XP Pro (patched) on slave with FAT.
Intel D850MVSE mobo with 2.2 GHz Northwood P4; 512M PC800 RDRAM.
Gigabyte GA-7VTXE with 1265MHz Athlon-XP 2000; 512M PC266 DDR SDRAM.
Sponsored: Hyper-scale data management