The thing that bugs me is that I have a pretty good personal relationship with the "business sales guy" from the Apple flagship store here in Berlin, Germany, and he tells me that the AB would absolutely be capable of being reprogrammed to help decode other codecs as well, not just ProRes.
"capable of being reprogrammed" isn't technically the same as "will be reprogrammed". If reprogrammed the FPGA to do HEVC 4:2:2 (or all of HEVC main 10 profiles) then pretty good chance it would not do ProRes anymore. So would have an Apple card that could not do Apple codec in that state. So why would Apple want to put the card into a state that doesn't handle Apple codec? They could, but what is in it for Apple to block themselves?
Of course we have no issues with prores whatsoever on the mac pro, it's the other codecs like hevc that are causing the issues where now the m1 blazes ahead. It would just be great if the mac pro could live up to it's claim of being upgradable if they could ship a new version of the afterburner that has more capabilities.
In a better matched up world AMD's HEVC de/encoder for HEVC would be better. Largely for Macs Apple has leaned on Intel's leading edge Qucksync de/encoder to do work. [ When T2 came around they used their own in some contexts for H.264. ]. T2 is relatively old at this point and the Mac Pro doesn't use Intel CPUs with Quicksync present. For the most part it should have been the GPU's job to cover HEVC; not Afterburner's.
Afterburner doing more concurrent different codecs is a problem if want to keep the same price point. To ground this a bit say that the FPGA has 500,000 programable gates and a 64MB buffer memory. And ProRes with six 8K decoders takes 320,000 'gates' to do and 40MB buffer .
And four HEVC 4:2:0/4:2:2 main 10 profiles takes 210,000 'gates' to do and a 20MB buffer.
Well , you can't reprogram the FPGA to do both at the same time. The combination of gates and/or buffer are over budget. Can try going to a bigger and substantively more expensive FPGA. However, if customers were complaining about $2K being large and wanting it to do more , then a $3-4K price probably isn't going to make most of them happier. ( many folks are price anchored on "free" HEVC de/encode in their Intel iGPU, dGPU , iPhone/iPad chip , etc. )
The likelihood that a non-Apple GPU was going to cover ProRes de/encoding was about zero. The likelihood that a non-Apple GPU in 2022-2024 would cover at least part of the HEVC main 10 profiels was extremely high. There is a odd ball disconnect where some camera vendors jumped on 4:2:2 and the GPU folks largely (except Intel) ignored it. You are looking to Afterburner to solve that problem. The problem is "cleaning up after the GPU vendors"... I'm pretty skeptical that Apple wants to take on that task. Once the GPU vendors correct for that mismatch that kind of "value add" for Afterburner is relatively low. Even lower, on a high density of mac users who are on a fixed function logic HEVC/ProRes that Apple has made .
For HEVC 4:2:0 Adobe Premier is finally way better now (if compare Premiere at Afterburner intro to now on Windows it is way different for many GPU deployments. ). The hardware was present in many modern GPUs. It was case primarily of waiting for software to catch up. The 4:2:2 corner case is relatively small. ( If have Canon camera dumping a tons of local 4:2:2 it doesn't look small , but big picture of HEVC space, it is. )
You can't tell me that it wouldnt be possible for them to put those video engines on a pci card and sell it as afterburner 2.0 for 2500$. I would buy that, if it would let me keep and evolve my expensive mac pro.
I'm skeptical that HEVC is super small and that there is a huge block of unused "gates" on the Afterbuner FPGA. I don't think there is a "large" FPGA there. The $2K includes some healthy margin to cover development and overhead.
It is possible but the price point (and Apple's margin) seems likely to move more than $500. And once it gets too high , who is going to buy it? Especially if Apple stops selling new MP 2019 to put it into.
The other issue is could just plain buy a $2,799 Mac Studio 'full' M1 Pro , 32GB , 2TB storage just as a HEVC 4:2:2 -> ProRes ingest-transcode box. Even more so if have network video storage. If already have an Afterburner card the create more ProRes stuff. ( I know; didn't make proxies disappear). However, for $2.5K, you are in the range of getting a whole Mac Studio for that. [ Once the ProRes de/encode is in fix function, high volume logic ... FPGA just isn't price competitive. Apple is counting on that. ]
Apple made the T2 chip out of a 'chopped down' A10. A long shot, but if Apple created a "V1" or "T15D" out of a chopped down M1/M2 Pro die ( kill most of the CPU and GPU cores , flip the TB into x16 PCI_e v4 lanes , run some "BridgeOS" variant. ) leave the full Video fixed function logic (probably NPU unit I suspect is coupled to some of the workload) that could work in that price range. Wouldn't be a FPGA anymore. However, unless they were going to run that out to more than just MP 2019 targets, there would not be volume to motivate that. The leading iPhones have some ProRes so I would be shocked if the plain M2 SoC doesn't get something (and already has the HEVC coverage). So what Macs would really need these and have enough x16 PCI-e v3 bandwidth to make them highly useful?
