Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)
On 3 Apr 2004 14:21:55 -0800, firstname.lastname@example.org (muhaha) wrote:
>By 4/8/16 pipelines per gpu, they mean that the instruction gets
>broken down to 4/8/16 µops or is it n-time superscalar ? If pipelined,
>how come they always manage to break it by four ?
In the GF2 MX400 card I have the 2 pipelines means it can process 2 x
16 bit data words (pixels) simultaneously. On some computers this is
called parallel vector processing. This means that the GPU can carry
out the same processor operation on two lots of data at the same time.
If you had a 4x pipeline, it would be able to process 4 x 16 bit words
in a 64 bit pipeline. Or it could process 2 x 32 bit pixels.
If I use 32 bit colour, it can only process one 32 bit pixel at a
time. This results in much lower benchmarks, the graphics card
benchmarks about half the speed at 32 bits compared to 16 bits.
To my knowledge the pipelining only affects the data, not the
You know, a lot of things, in this world, they make no sense. But now Im here, everthing make no sense.