Head 2 Head – Windows NT4 vs. 95 vs. 98 vs. 2000


By Rick Talbot
Notorious Networks
Feb 29, 2000



Introduction

    This article tests all 4 Windows operating systems on the same computer, with the same benchmark programs. We wanted to find out which OS would perform the best when no OS is given an unfair advantage over the others. You won’t find Microsoft style “let’s compare Windows 2000 to NT4 with broken drivers” style testing here. We have endeavored to use only the purest and most rigorous methodology in the known universe:
    All across the board I ran the same version of Quake II, Wintune 98, and Winbench 99. All OSes used the same swapfile setting: a single swapfile set to 320 megs (2.5 x the system memory). All OSes used the same drivers, or as close as possible to the same drivers. All OSes had their performance settings set as close (or identical) as possible.
    The only exception to this methodology was with the Windows 2000 drivers for the Voodoo2. Because 3Dfx did not create drivers for  Windows 2000, I used two approaches:

  1. I used the 3dfx NT4 drivers. They work fine, but don’t get rid of the ‘? Unknown device’ in the Windows 2000 device manager.

  2. I then tested again with Anthony Toste’s hacked Windows 2000 drivers. They install fully and work fine.



Index


1. Hardware setup
2. OS memory usage on setup
3. OS base install size
4. Part I – Application Performance test (Wintune 98 and Winbench 99)
5. Part II – Game performance test (Quake II)
6. Overall conclusion

Hardware Setup

Motherboard: Abit TX5 motherboard. 512KB pipeline cache
CPU: K6-2 400MHz, running at 75x5.5 (412.5 MHz)
RAM: 128 MB 10ns SDRAM
Video:  ATI Rage II+dvd, 2MB EDO (“3d Charger”)
CD: HP 8110i CD Rewriter
Cd: Pioneer 32x CD-ROM
Test drive: Maxtor 90680D4 6.8 gigs, 5400 rpm
Other drive: Fujitsu 4.3 gig, MPB3043ATUE, 5400 rpm
Removable: Iomega ATAPI Zip 100
Floppy: 3.5” floppy
Modem: Aopen FM56-ITU2 ISA
Network: D-link 528DE PCI Ethernet adapter
Sound: SoundBlaster AWE-64 Value ISA
Drive controler: Promise Ultra-66 Drive Controller
3D video: Creative Voodoo2 8MB


Notes:

  • The test drive (Maxtor) was run off of the Ultra 66 controller. This provided for full UDMA support in all operating systems. Therefore Windows NT 4 was on an even footing with the other OSes.

  • Windows 2000 used the built in Ultra 66 drivers. Windows NT4, 95, and 98 used the drivers from the Ultra66 driver floppy disk v1.42 .


Operating system Memory usage on startup

    Below, you will see that Windows NT 4 leaves the most physical RAM free after startup. Windows 2000 really isn’t that bad – it’s only 3 MB behind Windows 98. Note however that the Win 9x family has zero swapfile use at start, while NT 4 has 26 megs and win2K has 40 megs swapped out initially.
 

K= Kilobytes, B = Bytes, MB = Megabytes
  Handles Threads Processes Total RAM Free RAM Cache Swap Size Commit Limit Peak Kernel Paged Non-paged
Win 2K 2435 188 14 130612 K 85480 K 54516 K 40188 K 442680 B 44412 K 14740 K 12312 K 2428 K
NT4 880 88 17 130484 K 95580 K 10540 K 26712 K 446048 B 29160 K 8208 K 6968 K 1240 K
95b       128 MB 92.1 MB 21.9 MB 0          
98       128 MB 87.8 MB 23.5 MB 0          

 

OS Base Install Size

Windows 98: 170 MB
Win 95B: 95.2 MB initially, 156 MB after IE4 and DirectX 7 added
Win NT: 90.2 MB initially, 175 MB after Service Pack 6 and IE4
Windows 2000: 655 MB

 
655 MB? It appears that Windows 2000 places everything onto your hard drive. This makes sense in an office setting, where the Windows CD is not always available for new driver installation. Windows 2000 keeps all of its drivers on your computer, in the drivers.cab file, which is a portly 51 MB in size. Also, the Arial Unicode Truetype font is 23 MB.
In Windows 95B, there is the Online Services folder. Delete it from your Program Files folder and you’ll save 15 MB.

 


Part I – Application Performance Tests


Figure 1 - Winbench 99 scores



Figure 1 – WinBench 99 Scores

  • The Winbench 99 scores show that Windows 2000 is overall only slightly slower than NT4.

  • A good surprise is that NTFS filesystem performance has improved with Win2K. The High End test results show there is only around 10% performance loss (against FAT16) in Win2K instead of over 30% loss in NT4.

Windows 95 and 98 can’t come close to the NT family scores, but it does seem that Microsoft did some tweaking to bring up the High End test scores under Win 98 (Win 98 gets an 8% higher score than Win 95).

 


Figure 2- Wintune 98 test results



Figure 2 – Wintune 98 test results

  • Windows 2000 doesn’t preserve the high 2D performance of NT4. In fact, it gets the lowest MegaPixels-per-second score of the four Windows, and is only 70% the speed of NT4’s 2D.

  • All four Windows show well on the MIPS and MFLOPS calculation results, all scoring within spitting distance of eachother. The exception is the Win95B MIPS score. It’s strange, but I don’t have an explanation.

  • Windows 2000 beats the others handily in the Integer and MMX application simulation scores, and ties in the floating point application simulation.

  • Windows 2000 improves slightly on NT4’s already better RAM handling speed. In reality the difference between the highest and lowest scores is only 5%. Not enough to make much difference to the average person.

  • Windows 2000 has the highest cached hard disk and uncached hard disk performance scores. The lame duck here is Win95B, which gives us 11 MB per second less than win 98.


Summary of part I

    It looks like Windows 2000 performs comparably to Windows NT4, and even boasts a more efficient NTFS implementation. While the increased memory and processor requirement of Win2k pulls it back somewhat from the NT4 level of performance, it still runs circles around both 95 and 98.
    It appears that Microsoft did make improvements with Windows 98. The Wintune 98 MIPS and HD cached scores show this, as well as the WinBench 99 High End scores.

 


Part II – Game Performance Comparison


Figure 3 - 166 MHz performance



Figure 3 – 166 MHz performance results

    If you’ve got a slow machine, then Windows 95 or 98 are definitely the superior platforms for gaming. Especially if you’ve got a 3dfx card. The two NT setups fall far behind.



Figure 4 - 300 MHz performance



Figure 4 – 300 MHz performance

    At 300 MHz, Windows NT 4.0 edges ahead and takes the top performance. Again Win2K is at the bottom of the heap, but only trails by about 3 frames/sec.





Figure 5 - 412.5 MHz performance test



Figure 5 - 412.5 MHz (75 bus speed x 5.5)

  • Here all of the Windows perform very close to one another, especially at 800x600, where the differences average about .5 frames/sec.

  • Note the performance of the Win2k-alt column. This is the hacked NT5 driver by Anthony Toste. It brings Windows 2000 in line with the others. [update March 8/2001: Now that 3Dfx is out of business, we can see that their failure to make proper drivers for the Voodoo 2 was a sign of their financial situation.]



Figure 6 - Software Performance test 412.5 MHz



Figure 6 – Software performance (412.5 MHz)

    Here is a software rendering test of all four systems. We see that NT4 is again the king, probably because of its faster 2D video capability. Windows 2000 falls behind Windows 98 by 0.8 fps.


Figure 7 - software performance 3dnow example




Figure 7 – 3dNow and CPU optimizations test

  • Since I was testing with the Quake II modified by AMD, I could always run it in 3dnow! mode (which I did NOT do for any of the other tests). Figure 7 compares regular software mode with 3dnow! enabled software mode. It also shows the performance gains that can be had when your cpu is optimized. K6-2(or 3) users can use PowerTweak 2.0 or RawPower 1.0 to optimize their CPU.

  • This is especially important for K6-2 (and 3) owners whose motherboards do not enable certain cache optimizations that can add 100-200 MIPS in integer performance. This optimization gives us 5.3 more fps in the chart above (‘soft’ and ‘soft + cpu’).

  • The 3dnow! software in Quake II can provide 2.9 more fps than the standard software drivers (see ‘soft’ and ‘soft + 3dnow’).

  • The ultimate kick comes when your cpu had all optimizations turned on, plus your game supports 3d now. As seen in the ‘soft + 3dnow + cpu’ column, we get a full 10.3 fps increase (50%) when this happens.




Summary of Part II

    Windows 95 or 98 are better gaming machines for those with slower systems. At higher speeds NT4 takes over, but is only really useful for games that support glide or OpenGL, since Microsoft, hoping to force people to Windows 2000, limited DirectX support to only version 3.0. Windows 2000 makes an alright gaming platform, but you may find some performance loss over Win 95/98. However, 2K is much more stable than any version of Windows so far, so you might enjoy less crashes while 'fragging'. Both NT4 and 2K have better modem performance, so you might find online gaming to be smoother. Expect performance loss if you plan on gaming with an NTFS partition.


Overall Conclusion

    Windows 2000 is a fully capable successor to both NT and 9x families. It offers the highest stability of any windows yet. It has higher processor and memory requirements, but you won’t need 200 MHz of additional speed to feel the same level of performance - as Intel had implied early in the year. A 233 will feel slightly sluggish in Win2K. My experience has shown me that anyone with a 400 MHz or faster machine will find that their computer does not seem any slower after moving to Windows 2000.
 

Windows 2000 is the overall best OS for the person who does more than just game, and who has 128 MB (or more) of memory, and who is confident in their computer skills.

END