[FFmpeg-devel] Hardware purchase request Apple M2

Neal Gompa ngompa13 at gmail.com
Sat Feb 18 13:43:22 EET 2023

On Tue, Feb 7, 2023 at 4:41 PM Martin Storsjö <martin at martin.st> wrote:
> Hi Devin, Thilo,
> On Tue, 7 Feb 2023, Thilo Borgmann wrote:
> > Am 07.02.23 um 21:19 schrieb Devin Heitmueller:
> >> On Mon, Feb 6, 2023 at 6:24 AM Thilo Borgmann <thilo.borgmann at mail.de>
> >> wrote:
> >>> Martin pointed out we don't need an M2 for FATE as the instruction sets
> >>> are equal on both machines.
> >>
> >> It may be worth noting that while the M1 and M2 have the same
> >> instruction set, there is no guarantee the video processing blocks are
> >> identical.  At a minimum we know that the ProRes acceleration supports
> >> 8K resolutions in the M2.
> Ah, good point. Yeah they're probably not identical, but I would kinda
> expect them to be mostly similar.
> >> I got burned by a subtle change in behavior for videotoolbox between
> >> Intel and M1,
> Yeah those two implementations are entirely totally different - they
> differ on essentially every single observable detail.
> >> so it's possible there are comparable differences between M1 and M2.
> Yes it's somewhat possible that something could differ, but TBH I don't
> expect the difference to be very big.
> E.g., I've tracked bugs in Apple's HW HEVC encoder, where the same bug has
> been observable across many generations of iPhones up until the M1 (until
> the bug was fixed in the latest iOS and macOS updates last year).
> >> If the tests include validating hardware acceleration then there may
> >> still be benefits in running the tests on both platforms.
> >
> > Thanks for throwing that in. AFAICT we still don't cover hw accells in FATE,
> > though.
> Yeah we don't do that currently - but we probably should; at least some
> level of smoke testing would be extremely useful.
> In any case - I'm not arguing against using the available funds for
> getting an M2, that sounds like a totally reasonable thing to do - I'm
> just saying that I don't think it gives much extra value in the context of
> a separate FATE instance.

I don't know this for sure, but I suspect that we're going to see
future enablement of AV1 decoding on M2, judging by the Safari release
notes: https://developer.apple.com/documentation/safari-release-notes/safari-16_4-release-notes

Quote: "Added WebRTC support for hardware AV1 decoding on supported
device configurations."

Of course, it's just a guess, but if we start testing hwaccels, that
might be relevant eventually.

真実はいつも一つ!/ Always, there's only one truth!

More information about the ffmpeg-devel mailing list