A Comparison of Different Amounts of L2 Cache on a 200MHz Pentium

user image



The song used for the MP3 playback test is "Feel the Melody" by Ian Anderson (Druaga), visit his Bandcamp (https://druaga.bandcamp.com/) and watch his YouTube videos. (https://youtube.com/user/Druaga1)
How does having L2 cache on an old Pentium i200 computer impact performance versus not having it? Furthermore, how much of a difference does it make to add more L2 cache? This video attempts to show the performance differences between having no L2 cache, 256KB of onboard L2 cache, and 512KB of L2 cache including a COAST expansion.

Obviously, the greatest performance gap is in between having an external L2 cache and not having one. Adding more L2 cache on top appears to provide little performance improvement when everything else is unchanged.

This video does not address other possibly important factors, including the difference more L2 cache would make with larger amounts of RAM, as well as potential performance differences between traditional cache chips and the COAST module I used to get 512KB of L2 cache.

Since the Socket 7 era, the technology of L2 cache has greatly improved, thus the performance has, too. Starting with the Pentium Pro and Pentium II, L2 cache was moved from the motherboard to the CPU packaging. Later on when CPU manufacturing improved, the L2 cache was integrated directly into the CPU die, and that's how modern CPUs are designed today.

Tests performed:
0:15 - System Speed Test 4.70 (focused on memory test)
Not many differences are immediately noticeable in the graphs, other than the time it takes to complete the tests. Towards the ends of the graphs, there is slight staggering in the reading and moving tests between 256KB and 512KB of L2 cache.

0:59 - 256Kbps MP3 Playback
Even when playing back in full quality, MP3 playback does not lag on any instance. However, as this computer is very old and MP3 was only starting to gain popularity back then, MP3 playback uses up a lot of the CPU compared to WAV playback. Having more L2 cache seems to improve playback efficiency.

1:21 - 320x240 MPEG-1 Playback
Whoa, check out the cannibalism here! It's my Pentium i200 computer playing back a video of that same Pentium i200 computer playing the DOS port of Quake II! https://www.youtube.com/watch?v=AnaohppbHF8
As with MP3 playback, L2 cache makes a difference in video playback. I have not been able to synchronize the three clips reliably as I could with the MP3 playback test, but if you check the CPU usage graphs, you'll notice it's lower with more L2 cache. Oh yeah, that severe lagging isn't caused by this old computer playing back the video, it's in the video itself if you go watch it. I really should've used a better example, but I figured it would be funny to make a computer play a video of itself playing a game. Bleh.

1:53 - Quake II Timedemo, Software Rendering
With Quake II's software-based rendering engine, the burden is almost entirely on the CPU to produce sequential images in a 3D environment rapidly. Quake II's software rendering produces darker, chunkier textures than hardware-based OpenGL rendering, which may be better or worse depending on your preference. Inevitably, software rendering is way slower. Having L2 cache improves the frame rate by about 2 FPS, but it's not like you'd want to play a game at a pathetic 6-8 FPS. The difference between 256KB and 512KB of L2 cache is tiny, only improving the timedemo by a little over two seconds.

4:00 - Quake II Timedemo, Voodoo2 Rendering
Oh yeah, everyone loves Voodoo cards! Using a single Voodoo2 8MB card, Quake II is playable on this computer. With OpenGL hardware rendering, textures and models are a lot smoother and brighter. L2 cache differences are also farther apart; having 256KB of L2 cache over none improves the frame rate by almost 7 FPS. Another 2 FPS are gained when increasing the L2 cache to 512KB. Another thing... both the software and Voodoo2 tests ran at a resolution of 640x480.

4:49 - 3DMark 99 Benchmark
I've run the full 3DMark 99 benchmark three separate times with different cache amounts, but I only show the "game" benchmarks in this video. All tests used the same Voodoo2 card from the Quake II OpenGL tests. The scores for each test are as follows:
- 734 3DMarks for no L2 cache
- 1010 3DMarks for 256KB L2 cache
- 1048 3DMarks for 512KB L2 cache

So again, not much of a difference when you add more to the existing L2 cache. I should also mention that one of the tests was skipped due to insufficient RAM. If you want to run Futuremark's older benchmark programs, the full versions are available for free here: http://www.futuremark.com/benchmarks/legacy

Intel Pentium i200 at 200MHz, Socket 7 (note: not MMX)
64MB RAM, Two FPM SIMMs at 60ns
3dfx Voodoo2 8MB 3D Accelerator (single)
Cirrus Logic 5464 4MB Video Card
Creative Sound Blaster AWE64 Sound Card
MSI MS-5128 Motherboard

Originally posted October 19, 2016
Uploaded on Feb 4, 2019 at 02:02 AM
Category: Science and Technology

Comments (1)