First post, by Ozzuneoj
- Rank
- l33t
I was doing some testing recently and noticed that some Trio64V+ cards are noticeably sluggish in Windows 98SE while others are super fast and responsive. Even cards that are nearly identical but have slightly different configurations, such as the two STB Powergraph 64V cards I have picked for this test. These results are all repeatable and the performance difference between the cards exists on either a K6-III system or a Pentium III system.
Below are pictures of the cards, a picture of dragging a window (with contents shown) quickly on the desktop with each card and a benchmark comparison from Wintune97, including one with the BIOS chip swapped from the slow card to the fast card. The tests were all done at 800x600x32bit at 85Hz (75Hz does it too) on a PIII 850 on a 440BX with the same exact drivers. Absolutely nothing else is changing between the cards aside from powering the system down to swap cards and powering it back on. Also, it's worth noting I am using PS2Rate to force the mouse to update at 200hz, which definitely exaggerates the window-dragging sluggishness on the slower card, but it has no impact on the benchmark numbers.
I have around a dozen Trio64V+ cards of varying makes and models (mostly STB and Diamond). I would say that about one third of them are "fast" and the other two thirds are "slow", but that was with some fairly haphazard testing so it could be inaccurate. Can anyone think of any reason for this?
The only significant differences between the cards are the memory chips, but with 8x8bit EDO and 4x16bit EDO memory chips, they should both have a 64bit total bus width, and even the 60ns chips should be fast enough to run at the 50Mhz or 60Mhz that these cards apparently can operate at (remember, EDO timings are weird, you can't just convert ns to Mhz).
After reading the Trio64V+ datasheet it seems that there are three possible memory configurations for these cards. It says: Supports standard fast page mode and EDO DRAMs (60 MHz) and 1-cycle EDO DRAMs (50 MHz)
I'll be honest, I never even knew about this faster single-cycle mode of EDO. What exactly enables or disables this feature? Is it possible that one of these cards is running in single cycle mode while the others is not and that is totally destroying the performance of the card without it?
The only other notable differences (aside from small component placement differences, which could also be related) are the S3 chip revision (G1E3DE on the fast card, H1C3BE on the slow one) and the existence of an additional IC near the PCI connector on the faster card. It appears to be a Motorola LS14... but I'm not sure if I've identified that correctly or if it would have any impact on any of this.
Anyway... I'm stumped. Normally, I wouldn't care about the desktop performance of a 2D S3 Trio card that much but the difference here is so drastic that the slower cards feel barely better than an ISA card where as the faster ones feel more like a Matrox or Voodoo 3. I feel like if people aren't doing back to back comparisons of these cards then the issue may not have been documented before.
If no one can come up with a quick solution I will do more testing with other cards to possibly narrow down more differences between them. For what it's worth, I believe both of my Diamond Trio64V+ cards performed like the fast card but I need to do more testing to be sure.
