Hacker News new | past | comments | ask | show | jobs | submit login

Yes, in terms of Dolphin's performance today, I don't doubt it. There are probably many more optimisations that can be made: Dolphin's AArch64 JIT is still pretty incomplete and inefficient compared to the x86-64 JIT, for example.

My point is in theory, given enough time and effort, the Pi 4 should have more than enough power to emulate a 20-year old device. I'd argue that it's certainly not "way way too weak" as the reply above suggested.




The RPi4 GPU is very weak for the tricks that Dolphin does. Good luck to get Ubershaders running on that for example.

The CPU isn't much of an issue.

Emulating a platform at such a low level in real-time is very heavy resource-wise, by design.


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.


Ok, 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.


My 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.


It 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...




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: