Some bored computer geek must be out there wondering – how does the crappiest video card introduced in 2000 compare with a mid-range graphics card of 1995?
S3 Trio 3D/2X otherwise known as the S3 ViRGE
86C368 S3 Trio 3D/2X, AGP 2x, 4MB EDO-DRAM
ATi Radeon 7000 otherwise known as the Radeon VE
Transcend Radeon 7000, 32MB 64bit DDR-SDRAM
One must consider the fact that back then, in 1995, S3 was the standard when it came to 2D graphics. Of course, it was a no brainer that the ViRGE really didn’t do 3D graphics. However, in 2D it was the fastest at… oh well, everything, I heard, and image quality was the best you could get back then. Benchmarks are hard to find from that era, you know.
Here is a very rare review of a S3 ViRGE based card.
In the other corner, ATi has always had really bad Linux drivers, but fortunately the OSS community stepped in with some drivers of their own that enable DRI. ATi is also famous for the good image quality back then, when compared to the Geforce, Geforce 2 and TNT cards.
So I did my own benchmarks. Since I ran Linux on that computer, and the ViRGE only did 2D, I chose the only 2D performance benchmark available – x11perf. X11Perf has a lot of benchmarks, but since only a few of them are really intensive and separate the wheat from the chaff, only a few are included here.
Celeron 433MHz, 66MHz FSB, 192MB 66MHz SDRAM, Asus P2B-F.
These are the number of times the card drew something on the screen, for example, a rectangle 500×500 pixels, or stippled, or xored or something like that.
Drawing 500×500 rectangles
10000 reps @ 0.5285 msec ( 1890.0/sec): 500×500 rectangle
S3 Trio 3D/2X
5000 reps @ 1.0166 msec ( 984.0/sec): 500×500 rectangle
Oops. The S3 card is half as fast. I just took it out of the box to look at it one more time, and it just pricked me on the finger too.
Copying 500×500 pixels from pixmap to window
8000 reps @ 1.1145 msec ( 897.0/sec): Copy 500×500 from pixmap to pixmap
S3 Trio 3D/2X
280 reps @ 18.8496 msec ( 53.1/sec): Copy 500×500 from pixmap to window
In one of the rather intensive tests, it seems, there is no contest here.
SHMPutImage 500×500 squares (when you use the x11 output in gmplayer, the driver uses this technique if you don’t use xv instead)
S3 Trio 3D/2X
280 reps @ 19.4611 msec ( 51.4/sec): ShmPutImage 500×500 square
200 reps @ 28.7597 msec ( 34.8/sec): ShmPutImage XY 500×500 square
Something happened here. The S3 Trio rocks… just a little bit, taking into account how hard the Radeon just whipped the Trio’s ass in the first two tests.
Oh well, that’s all for today. This is actually the only result I took notice of that had the S3 faster than the Radeon. So now you know. Even crappy cards from the Y2K era still handle 2D faster than the S3.
You can even feel the difference, believe me. I tried just all those xorg.conf options designed to push my S3 Trio card to the limit, just try man s3virge. But performance never improved. I was convinced that it was because of low RAM (I only had 64MB RAM back then). Then I bought 128MB of SDRAM and stuck it in, but the GUI was still sluggish as hell. Even fluxbox, the most plain of window managers, was slow when I right-clicked on the desktop.
So I got the Radeon 7000. Finally, when I right clicked on the desktop in XFCE4, the menu appeared faster. Finally, GNOME didn’t feel that much like molasses any more (although it still feels that way to a certain extent), and most importantly, fluxbox feels just like it should – fast, clean and minimal. Even though the 32MB of 64bit DDR SDRAM seems paltry by today’s standards, it’s more than enough for my purposes. I certainly love my Radeon 7000.
Then later I got the CuMine Pentium III Celeron CPU at 600MHz…. but that’s another story.