-
The Cheese Grater is back
-
[Eric Santiago] “I hope this Afterburner card isn’t a one year pony.”
Not sure how it could be.
The whole point of an FPGA is that you can easily re-configure it to point it at accelerating other types of footage handling if your needs or formats shift.
As such, I’d expect this approach would have an extremely long life expectancy.
Creator of XinTwo – https://www.xintwo.com
The shortest path to FCP X mastery. -
[Bill Davis] “As such, I’d expect this approach would have an extremely long life expectancy.”
This may be different, but as products go, that hasn’t been the track record.
– Oliver
Oliver Peters – oliverpeters.com
-
[Bill Davis] “Not sure how it could be.
The whole point of an FPGA is that you can easily re-configure it to point it at accelerating other types of footage handling if your needs or formats shift.
As such, I’d expect this approach would have an extremely long life expectancy.”
Until off the shelf computers can achieve ‘good enough’ results w/o needing to buy special-purpose hardware that requires specific support in order to be utilized.
That was the gist of the desktop video revolution was it not?
-
[Andrew Kimery] “Until off the shelf computers can achieve ‘good enough’ results w/o needing to buy special-purpose hardware that requires specific support in order to be utilized.
That was the gist of the desktop video revolution was it not?”
The problem is certain higher-end video tasks have outstripped the capability of general-purpose CPU hardware, at least to the extent Intel is willing to commit. E.g, Xeon does not have hardware-acceleration for Long GOP formats, so Apple had to put that in the T2 chip due to the iMac Pro and (maybe) Mac Pro — even though the MacBook Pro has Quick Sync and doesn’t need it.
A full-custom ASIC (like the T2) is very expensive to design and has long lead time. The cost of a custom ASIC must be amortized over very large production runs, yet many customers might not need that capability. E.g, T2 is a security chip but there was likely no other place to put the transcoding logic so it went there. That probably consumed valuable transistor budget that could have been spent on other things.
By contrast an FPGA implementation is much quicker and cheaper to design, and can be targeted to just the systems that need it. But FPGAs burn more power so are not suitable for a laptop. They can be reprogrammed in the field with new algorithms. If it becomes important to have hardware support for Google’s AV1 codec, that could be added to deployed machines.
It appears the main reason for the Afterburner FPGA board was hardware acceleration for demosaicing high-resolution ProRes RAW, but it’s plausible it could also be programmed for RED RAW, and it’s not limited to that. They can even be used to accelerate Long GOP formats like H264 and HEVC.
Apple is not the first to use FPGAs for specific video processing needs:
https://www.streamingmedia.com/Articles/ReadArticle.aspx?ArticleID=129580
https://www.xilinx.com/products/intellectual-property/1-4iso32.html
However it could give Apple a proprietary hardware performance advantage. Even if they expose the Afterburner capability via an API to other Mac app vendors, this won’t work on a PC. A PC vendor would have to fund, design, support and rally software vendors to a similar device. Economies of scale could give Apple an advantage, even for the “niche” 2019 Mac Pro. They might sell more Afterburner cards in 6 months than the total historical sale of RED ROCKET cards. Plus it’s a single-vendor hardware/software solution with a single support source. A high-end customer would not have to do three-way conference calls between RED, Microsoft and HP.
-
The potential is obviously there, but how much will it cost? How quickly will Apple expand support to other codecs? Will there be APIs so any third party can program for it? If Afterburner is a super-niche product how many third parties will actively support it?
I don’t doubt the performance of Afterburner, but the perceived cost/benefit is going to determine whether or not it’s a ‘game changer’ or just a really expensive piece of niche hardware for what’s widely seen as an already expensive workstation.
-
[Bill Davis]”Not sure how it could be.”
I’ve had many FPGAs in my equipment over the years. Fairlight upgraded my original card after about nine years. My latest FPGA from them is currently compatible with Fairlight in Resolve but for how long, I don’t know. However that is audio and they are monster processors for that purpose so have lasted much longer. In video, two of my Blackmagic cameras have FPGA cards on board. My three year old Ursa Mini has an older card that can’t be upgraded with the latest firmware. So three years was all I could get before the software over ran the capabilities of the card. Given the fast moving world of codecs and pixel aspects and depths, it is conceivable that an FPGA card might not last any longer than two to three years. Is the Afterburner upgradable? Unfortunately my camera isn’t but the Fairlight card is PCIe so yes it can be if required.
FPGAs are just processors and that world moves just as fast. Their capacity to be reprogrammed make them flexible for sure but not any more able to withstand development of processors and demands.
-
Preach it, Brother Shane!
-
I read elsewhere that the SSD memory in this cheesegrater is not an m2 connector but a proprietary variation on the standard connector, so if you want to add or upgrade the SSD’s, Apple will be your only source on that. Unless and until someone else copies that connector.
-
Wouldn’t it be possible to add fast ssd storage via the pcie slots? There’s 8 of them on this Mac.
-
[Andre van Berlo] “Wouldn’t it be possible to add fast ssd storage via the pcie slots? “
It can’t be standard SSD internal drives. If possible, it would have to be card-mounted, like the regular NVMe drives.
– Oliver
Oliver Peters – oliverpeters.com
Reply to this Discussion! Login or Sign Up