> or create a byte-array of some sort that gets translated, which while faster, still isn't fast enough. Except, of course, about 2-3 orders of magnitude faste

> or create a byte-array of some sort that gets translated, which while faster, still isn't fast enough. Except, of course, about 2-3 orders of magnitude faster than the best SVGA era cards were. I mean, you got many GB/s bandwidth now. Yes, that Tseng Labs ET4000 was amazing, with 7 MB/s over ISA, beating most (all?) other cards I bothered to try back then. Or some basic (but fast) later PCI card like S3 Trio 64+ 50 MB/s (IIRC). PCIe 3.0 can do up to 16 GB/s over 16 lanes. Even if it were 7 GB/s in practice, we're still talking about 1000x improvement compared to the best ISA SVGA card. You have also bus mastering and ability to map buffers that way that GPU itself has access to the data. Setting one pixel might be slow. But who cares when you can just have your "framebuffer" in system RAM and "MOV" away!