3 ms·
To elaborate a bit more on the GPU peak FP32 perf: > VideoCore VI @ 500MHz: 500 [MHz] x 2 [slice] x 4 [qpu/slice] x 4 [processor] x 2 [op/clock] = 32 Gflop/s
by my123 5y ago
To elaborate a bit more on the GPU peak FP32 perf:
> VideoCore VI @ 500MHz: 500 [MHz] x 2 [slice] x 4 [qpu/slice] x 4 [processor] x 2 [op/clock] = 32 Gflop/s
which is the RPi4 config.
compared to a Snapdragon 800: 147.9Gflops FP32, and that's still a very old part.
And about the Wii itself:
> With the Wii, Nintendo opted to go with an innovative control scheme, rather than to focus on processing power. It offered 12 gigaflops, which represents a 1.27x improvement over its GameCube predecessor
The margins are way too slim when counting emulation overhead.
- Reason077 5y agoOk, sounds like the Pi’s VideoCore VI is a lot less performant than I thought. I’d imagined it’d be closer to other mobile GPUs.
- my123 5y agoIt was probably a sane technical choice for them. RPi4 is made on the same node as the Snapdragon 800 (28nm), but with a much more recent (and bigger) CPU complex, using the Cortex-A72. As a bunch of RPis are headless anyway...
- 0xcde4c3db 5y agoMy understanding is that the VideoCore architecture was initially designed as more of a "media codec coprocessor", and was incidentally able to grow enough matrix math throughput to be pressed into service as a GPU.