Re: Raven. A homemade Atari-like computer
Posted: 20 Jun 2025 09:21
I got a W32i with working interleaved mode and that certainly boosts the internal vram bandwidth.
How much practical difference this makes on an Atari is questionable but I can see this being a big deal on PC back in the day, where pretty much everything automatically took advantage of 2D hardware acceleration and on-card caching of sprites thanks to Windows and DirectDraw.
Bandwidth:
Mach32 vs W32i (interleaved) vs W32i (non-interleaved)
Video move shows a significant difference in vram-to-vram copy speed over the W32i with interleaving disabled.
VDI operations:
Mach32 vs W32i (interleaved) vs W32i (non-interleaved)
We can see that the Mach32 is accelerating more operations compared to w32i.
Gembench W32i (interleaved) vs W32i (non-interleaved)
Gembench Mach32 vs W32i (non-interleaved)
We see the Mach32 driver taking advantage of more accelerants here.
And as an interesting experiment, if we run the test on a lower resolution then the difference becomes minimal and the non-interleaved W32i test is almost the same speed as the interleaved one.
My best guess is that the card is nowhere near starved for vram bandwidth at the slow pixel clock needed for this resolution, together with the tiny area of graphics that Gembench is using for blit related tests.
How much practical difference this makes on an Atari is questionable but I can see this being a big deal on PC back in the day, where pretty much everything automatically took advantage of 2D hardware acceleration and on-card caching of sprites thanks to Windows and DirectDraw.
Bandwidth:
Mach32 vs W32i (interleaved) vs W32i (non-interleaved)
Video move shows a significant difference in vram-to-vram copy speed over the W32i with interleaving disabled.
VDI operations:
Mach32 vs W32i (interleaved) vs W32i (non-interleaved)
We can see that the Mach32 is accelerating more operations compared to w32i.
Gembench W32i (interleaved) vs W32i (non-interleaved)
Gembench Mach32 vs W32i (non-interleaved)
We see the Mach32 driver taking advantage of more accelerants here.
And as an interesting experiment, if we run the test on a lower resolution then the difference becomes minimal and the non-interleaved W32i test is almost the same speed as the interleaved one.
My best guess is that the card is nowhere near starved for vram bandwidth at the slow pixel clock needed for this resolution, together with the tiny area of graphics that Gembench is using for blit related tests.