pouët.net

The psp only have a fixed pipeline gpu a bit like the one of the ps1 but faster (166mhz) also the video memory is directly accessible via the processor, the ps2 have an architecture which pre-dates the vertex shaders (150mhz for the gpu me think).

The processors of the 2 consoles are customized version of the mips, the ps2 (~295mhz) have a 128 bits r5900 with extra simd instructions + 2 vector units coprocessors (VUs) programmable with microcode + a mmu and a secondary r3000 (~33mhz) processor to handle io/sound and ps1 emulation, everything is dma orientated (10 channels).

The psp (333 mhz) have 2 r4400 (or approximatively) + 1 vector coprocessor (vfpu) with fixed instructions, the second processor have it's own memory (2megs) and is a pain in the ass to operate, there's 1 dma channel.

Both have fpu coprocessors.

That more or less sums it up.

Actually, the video memory is only indirectly accessible by the CPU. It's mapped into the CPU's memory address space, but all memory access operations go through the graphics engine. However, the main memory is accessible by the GPU aswell, 'cause it's connected to the main memory bus.

Yeah the video memory accessible by the cpu belongs to the gpu alright but since it's mapped it doesn't really matter from a programmer standpoint.

How can the gpu access anything else than it's own 2 megs of memory
(0x4000000 for the cpu, 0x0 for the gpu), some sort of wrapping or mirrored main memory somewhere after the vidmem ?

Bump.

Another way of running one's own code on the ps2 surfaced not so long ago:

http://www.psx-scene.com/forums/freevast/58292-official-free-mc-boot-releases.html

That's one less excuse to not to consider developing demos on that console.

nice