Frequently Asked Questions about the Afterburner accelerator for the Mac Pro

AppleInsider

As part of the launch of the new Mac Pro, Apple introduced a piece of hardware called Apple Afterburner that could be added to the configuration. AppleInsider explains what the card is, what it can do, and whether it is worth adding to an already powerful desktop.

Apple's Mac Pro at WWDC 2019

Editor's note: This has been updated on December 10, with everything that Apple has disclosed about the card during the Mac Pro launch.

What is the Afterburner card for the Mac Pro?

Simply put, Afterburner is a card for the Mac Pro that is designed for use in video production. Rather than relying on the processor or graphics cards for some tasks, the Afterburner takes over for some tasks, specifically those relating to video processing between formats, freeing up the rest of the system components to perform other tasks.

When fitted into a Mac Pro's PCI Express x16 slot, video editors will be able to manage high-resolution, high-bitrate videos with ease, without any of the stuttering or waiting that could occur in less-powerful systems. The smoother the workflow, the easier for the editor to work, and the faster they can complete the video project.

0

Subscribe to AppleInsider on YouTube

What does it change?

In particular, Apple's inspiration for the Afterburner is to eliminate the "proxy workflow."

Due to the high file sizes and bit rates of native camera codecs, there's a lot of data that needs to be processed during editing. The proxy workflow involves making versions of video files in a codec and format that can be more easily edited and using those smaller files to define the edits in a piece.

Later, when compiling the final result, the edits made to the proxy videos are applied to the original video files.

By using Afterburner, Apple intends for video production to be able to use the original video file codecs and formats without creating proxy versions. Eliminating this step can save time, both in removing the transcoding process for the proxies at the start of editing, and in compiling the final edits at the end.

What's that FPGA or ASIC thing?

On a more technical level, the Afterburner is a card with what Apple refers to as a Field Programmable Gate Array (FPGA), or a programmable Application-Specific Integrated Circuit (ASIC). This effectively means it is a card that has chips created for a specific task, rather than a general-usage chip.

While graphics cards offer the ability to perform computing tasks, they are still using chips that are "general" in usage, albeit within the field of graphics. They can be used for a variety of tasks, which makes them flexible, but not necessarily able to offer optimal performance compared with a made-for-purpose chip.

In the Afterburner's case, the chips are specifically designed to handle just specific tasks. They are optimized to handle the encoding and decoding of ProRes and ProRes RAW codecs, so they can handle far more than a similar-performance graphics card could when pushed.

It is claimed the Afterburner includes over a million logic cells, allowing it to process up to 6.3 billion pixels per second.

The Afterburner installed inside a Mac Pro

While the use of an ASIC is beneficial, Apple's decision to use an FPGA makes the card far more useful. While an ASIC isn't typically able to be changed after it has been created, an FPGA can be reconfigured after the fact by instructions to change the layouts of chip gates.

In effect, this means Apple could update how Afterburner is configured to improve it in the future, to support new codecs and to improve performance for existing versions.

What can it do?

The card is built to accelerate ProRes and ProRes RAW codecs, namely the encoding and decoding of the codecs, which is a processing-heavy task in most cases. Apple claims the card is capable of handling up to six streams of 8K ProRes RAW video simultaneously at 30 frames per second, making it extremely useful for video editors working at the highest possible level.

On less demanding video specifications, it is able to work on up to 23 streams of 4K ProRes RAW video at 30 frames per second, or at 4K ProRes 422, up to 16 video streams.

In terms of compatibility, Apple advises it will work with ProRes and ProRes RAW codecs in Final Cut Pro X, QuickTime Player X, and "supported third-party apps," though at this time it is unclear what these will be.

Is this a brand new concept?

Despite Apple describing Afterburner as a "Game-Changing" accelerator card, the idea of having hardware to perform this type of task isn't new.

Red Camera, the manufacturers of the high-resolution cinematography cameras, produced the Red Rocket, an accelerator card range that can process and transcode video, with a focus on codecs used by the camera range.

The Afterburner revealed on-stage at WWDC 2019's keynote

Do I need one?

As Afterburner is highly specific in what it can do, the only potential buyers of the card are those in the video editing field. The card won't provide any real performance benefits outside of that capacity, so unless you happen to be a video editor or someone in that field, the answer is no.

There is also little need to get the card if the user does video editing, but doesn't use the ProRes or ProRes RAW codecs. For those who need to edit their iPhone videos or uses an entry-level DSLR for video capture, these situations generally won't use codecs that necessitate the use of the Afterburner card at all.

Those in the creative industry dealing with high-resolution, high-bitrate video on a daily basis will certainly see the advantage of acquiring the Afterburner.

Does it have to be used for video?

Based on how Apple is presenting the card in its initial form, Afterburner will only really be useful for those with video production needs.

That being said, as it is an FPGA ASIC card and therefore able to be reconfigured in firmware updates, it is entirely plausible that Apple could enable users to change the configuration of the card so it can be used for completely different tasks.

For example, one configuration could be used for specific graphical tasks that are hard for a normal GPU to pull off, such as raytracing. It could possibly be set up to optimally perform a specific type of repetitive calculation or algorithmic action for the user's work, or at a stretch, mining cryptocurrency.

While plausible, Apple would have to identify different use cases for an FPGA ASIC card to provide additional processing assistance, and offer the option to reconfigure it at all. This all depends on the company's intentions for the card, and possibly those of Mac Pro users.

Will it work on other hardware?

So far, Afterburner is only identified as being able to work with the Mac Pro, and Apple always has the option of making it work with future Mac models, but there is no official word on compatibility elsewhere than the Mac Pro at present.

As the Afterburner appears to connect via PCIe 3.0, crucially without the secondary slot used by MPX modules, there is a possibility that it could be used in other setups. Depending on the physical size of the card, it could potentially be used as part of an external enclosure, allowing it to be used in a similar way to eGPU setups for MacBooks.

Ultimately it depends on how Apple makes the card function, and what it enables within macOS.

As a PCIe 3.0 card, there is also the feasibility of inserting it into a PC, but it is unclear whether it would be technically possible to use the card at all in that way, even if drivers were provided.

How much will it cost?

At the time of the launch of the Mac Pro, it was revealed the card costs $2,000 in the system configuration part of the ordering process. The card has also been found with its own listing in the online Apple Store, showing the same $2,000 price tag. As of December 10, the card is not available to purchase separately.

But what about...

We're still getting a lot of questions about the Mac Pro and its hardware. As we collect questions and gather answers, we'll update accordingly.

kirsch

On a more technical level, the Afterburner is a card with a Field Programmable Gate Array (FGPA), or a programmable Application-Specific Integrated Circuit (ASIC). This effectively means it is a card that has chips created for a specific task, rather than a general-usage chip.

An FPGA is exactly the opposite of chips created for a specific task. It is a chip that can be configured to do a specific task. The "Programmable Application-Specific" is an oxymoron. The way FPGAs work, an image is loaded onto them which reconigures the gates on the chip in order to do something specific. The image can be updated or changed completely by the vendor (Apple) after the card has shipped and is sitting on the customer's computer. They would do this to add new codecs, fix bugs, improve performance, etc.

It makes sense for Apple to use an FPGA for this and not an ASIC, which would be cost-prohibitive at the volumes this is likely to sell at.

scdundas

It wood be super cool if this was also released in a Thunderbolt enclosure for use with the Apple's other systems.

I realize that's unlikely to happen and it wouldn't have quite as high of performance. But it would be so useful for people who do a lot of video but don't need all the power the Mac Pro provides. 

sflocal

Queue the whiners and haters saying Apple should offer a $99 version of the Afterburner card so they can play Fortnight.

macinfish

kirsch said:

On a more technical level, the Afterburner is a card with a Field Programmable Gate Array (FGPA), or a programmable Application-Specific Integrated Circuit (ASIC). This effectively means it is a card that has chips created for a specific task, rather than a general-usage chip.

An FPGA is exactly the opposite of chips created for a specific task. It is a chip that can be configured to do a specific task. The "Programmable Application-Specific" is an oxymoron. The way FPGAs work, an image is loaded onto them which reconigures the gates on the chip in order to do something specific. The image can be updated or changed completely by the vendor (Apple) after the card has shipped and is sitting on the customer's computer. They would do this to add new codecs, fix bugs, improve performance, etc.

It makes sense for Apple to use an FPGA for this and not an ASIC, which would be cost-prohibitive at the volumes this is likely to sell at.

I'm an FPGA designer myself and I was just about to add the same comments.  You are spot on.

It would be great if Apple would allow the user to create their own FPGA coprocessor design and s/w driver, but I'm betting the answer is NO.  Custom Afterburner designs would be a very cool tool for designers working on machine learning, AI, neural networks, other codecs, etc..  Also, I really wish the EDA companies (Xilinx, Intel PSG, ModelSim, Aldec, etc.) would port their tools to macOS.

mcdave

Odd to use an FPGA only for ProRes.  I’m assuming other formats will become available soon.

I guess the mass market for this tech is Photo RAW import/HEIF export in MacBook Pros.

zimmie

kirsch said:

On a more technical level, the Afterburner is a card with a Field Programmable Gate Array (FGPA), or a programmable Application-Specific Integrated Circuit (ASIC). This effectively means it is a card that has chips created for a specific task, rather than a general-usage chip.

An FPGA is exactly the opposite of chips created for a specific task. It is a chip that can be configured to do a specific task. The "Programmable Application-Specific" is an oxymoron. The way FPGAs work, an image is loaded onto them which reconigures the gates on the chip in order to do something specific. The image can be updated or changed completely by the vendor (Apple) after the card has shipped and is sitting on the customer's computer. They would do this to add new codecs, fix bugs, improve performance, etc.

It makes sense for Apple to use an FPGA for this and not an ASIC, which would be cost-prohibitive at the volumes this is likely to sell at.

I wouldn't quite call an FPGA the opposite of a chip created for a specific task. Sure, the silicon wasn't etched for just the one task, but a blank FPGA doesn't do anything useful. Once configured, it has been created for a specific task (modulo RISC-V and the like). It can be reconfigured later, but that makes it no longer suitable for the original task.

I also don't think an ASIC run would be cost-prohibitive. They don't need to use the smallest process, so they can shop around for fabs. Last time I priced a run, it was about $100k setup (as long as you provide the Verilog), some small amount per wafer, then some larger amount per chip depending on the packaging technology you want. A low-run ASIC should be well under $100 each for 10k units.

tipoo

If someone can weasel out who makes it, I'd be interested. 

Intel is the obvious choice, Xilinx is my dark horse. 

What?!!

zimmie said:

kirsch said:

On a more technical level, the Afterburner is a card with a Field Programmable Gate Array (FGPA), or a programmable Application-Specific Integrated Circuit (ASIC). This effectively means it is a card that has chips created for a specific task, rather than a general-usage chip.

An FPGA is exactly the opposite of chips created for a specific task. It is a chip that can be configured to do a specific task. The "Programmable Application-Specific" is an oxymoron. The way FPGAs work, an image is loaded onto them which reconigures the gates on the chip in order to do something specific. The image can be updated or changed completely by the vendor (Apple) after the card has shipped and is sitting on the customer's computer. They would do this to add new codecs, fix bugs, improve performance, etc.

It makes sense for Apple to use an FPGA for this and not an ASIC, which would be cost-prohibitive at the volumes this is likely to sell at.

I wouldn't quite call an FPGA the opposite of a chip created for a specific task. Sure, the silicon wasn't etched for just the one task, but a blank FPGA doesn't do anything useful. Once configured, it has been created for a specific task (modulo RISC-V and the like). It can be reconfigured later, but that makes it no longer suitable for the original task.

I also don't think an ASIC run would be cost-prohibitive. They don't need to use the smallest process, so they can shop around for fabs. Last time I priced a run, it was about $100k setup (as long as you provide the Verilog), some small amount per wafer, then some larger amount per chip depending on the packaging technology you want. A low-run ASIC should be well under $100 each for 10k units.

Based on your comment, I have a very hard time  believing you do any work with FPGAs or ASICs. Im impressed you dropped Verilog in there so you do know something about it so ill try not to be an ass (i know, already failed there).  I am purely a hobbyist but i meed to clear up this real lack of fact going on here. 

ASIC != FPGA ever. Literally opposite as Kirsch said. Im glad him and macinfish made that point for me but ill explain and maybe the article will correct some of its GLARING ERRORS. FPGA is field programmable as they say, and zimmie pay attention, they can be updated with new versions of the code (verilog etc...) and still Do they old job using the new configuration of the gates. Sorry zimmie I know you know probably more than me but your wording is misleading. 

Heres where the article is wrong and nobody has pointed it out. ASIC is application specific hardware. Literally the hardware has been designed and baked into the silicon and there is no changing the configuration. The benefits are that these get actually faster than FPGAs and more efficient. Mostly more efficient as there is nothing extraneous on board. It does what it is supposed to and that is it. An fpga may be used to make the general processor in one design but a dedicated hardware processor will always be better in every way except the ability to reconfigure it.  

theluker

Apple confirmed this IS an FPGA ASIC... they confirmed this in an interview after saying it can be rewritten for many other tasks besides video... Also why does this article imply it's not going to be in the base model? It was impled by apple to be in all models at the event... and for 6 grand USD and 8k Canadian for a 580x and 256gb of SSD etc one of these suckers better be hiding in there!

theluker

macinfish said:

kirsch said:

On a more technical level, the Afterburner is a card with a Field Programmable Gate Array (FGPA), or a programmable Application-Specific Integrated Circuit (ASIC). This effectively means it is a card that has chips created for a specific task, rather than a general-usage chip.

An FPGA is exactly the opposite of chips created for a specific task. It is a chip that can be configured to do a specific task. The "Programmable Application-Specific" is an oxymoron. The way FPGAs work, an image is loaded onto them which reconigures the gates on the chip in order to do something specific. The image can be updated or changed completely by the vendor (Apple) after the card has shipped and is sitting on the customer's computer. They would do this to add new codecs, fix bugs, improve performance, etc.

It makes sense for Apple to use an FPGA for this and not an ASIC, which would be cost-prohibitive at the volumes this is likely to sell at.

I'm an FPGA designer myself and I was just about to add the same comments.  You are spot on.

It would be great if Apple would allow the user to create their own FPGA coprocessor design and s/w driver, but I'm betting the answer is NO.  Custom Afterburner designs would be a very cool tool for designers working on machine learning, AI, neural networks, other codecs, etc..  Also, I really wish the EDA companies (Xilinx, Intel PSG, ModelSim, Aldec, etc.) would port their tools to macOS.

100%... they did confirm it can be rewritten for other tasks but they didn't specify if end users or the community will be able to do so or we have to reply on them to decide what it will be used for... making this compute power based only around videographers would be sickening to me as someone in the scientific community. Apple sometimes seems to forget that the word professional goes a little further than photographers and videographers.

zimmie

What?!! said:

zimmie said:

kirsch said:

On a more technical level, the Afterburner is a card with a Field Programmable Gate Array (FGPA), or a programmable Application-Specific Integrated Circuit (ASIC). This effectively means it is a card that has chips created for a specific task, rather than a general-usage chip.

An FPGA is exactly the opposite of chips created for a specific task. It is a chip that can be configured to do a specific task. The "Programmable Application-Specific" is an oxymoron. The way FPGAs work, an image is loaded onto them which reconigures the gates on the chip in order to do something specific. The image can be updated or changed completely by the vendor (Apple) after the card has shipped and is sitting on the customer's computer. They would do this to add new codecs, fix bugs, improve performance, etc.

It makes sense for Apple to use an FPGA for this and not an ASIC, which would be cost-prohibitive at the volumes this is likely to sell at.

I wouldn't quite call an FPGA the opposite of a chip created for a specific task. Sure, the silicon wasn't etched for just the one task, but a blank FPGA doesn't do anything useful. Once configured, it has been created for a specific task (modulo RISC-V and the like). It can be reconfigured later, but that makes it no longer suitable for the original task.

I also don't think an ASIC run would be cost-prohibitive. They don't need to use the smallest process, so they can shop around for fabs. Last time I priced a run, it was about $100k setup (as long as you provide the Verilog), some small amount per wafer, then some larger amount per chip depending on the packaging technology you want. A low-run ASIC should be well under $100 each for 10k units.

Based on your comment, I have a very hard time  believing you do any work with FPGAs or ASICs. Im impressed you dropped Verilog in there so you do know something about it so ill try not to be an ass (i know, already failed there).  I am purely a hobbyist but i meed to clear up this real lack of fact going on here. 

ASIC != FPGA ever. Literally opposite as Kirsch said. Im glad him and macinfish made that point for me but ill explain and maybe the article will correct some of its GLARING ERRORS. FPGA is field programmable as they say, and zimmie pay attention, they can be updated with new versions of the code (verilog etc...) and still Do they old job using the new configuration of the gates. Sorry zimmie I know you know probably more than me but your wording is misleading. 

Heres where the article is wrong and nobody has pointed it out. ASIC is application specific hardware. Literally the hardware has been designed and baked into the silicon and there is no changing the configuration. The benefits are that these get actually faster than FPGAs and more efficient. Mostly more efficient as there is nothing extraneous on board. It does what it is supposed to and that is it. An fpga may be used to make the general processor in one design but a dedicated hardware processor will always be better in every way except the ability to reconfigure it.  

You seem to have missed my point that an FPGA does one thing at a time. If you want it to do something else, you need to reconfigure the entire gate array, which takes hundreds of milliseconds at the small end up to several seconds for large arrays. That's more similar to an ASIC than it is to a general-purpose CPU running software.

An FPGA is a reconfigurable ASIC. Most vendors even have toolchains which let you take exactly the Verilog you would use for an ASIC run and hook it up to the FPGA's I/O pins.

Of course, if you implement a general-purpose core like RISC-V on an FPGA, it becomes a general-purpose CPU, but then it isn't as suitable for the specific task another configuration could do more efficiently.

UrbaneLegend

I'm convinced this Afterburner card will be forgotten by this time next year. For it to be worthwhile it has to be spectacularly cheap compared to either a CPU or GPU upgrade and it needs to support different codecs. 

ProRes/RAW is easy to work with and the figures Apple released don't blow me away and many RED users are getting similar performance with R3D media with off the shelf GPUs and that's a much more processor intensive codec to work with than ProRes.

You'll be hard pushed to find a production that is confined to a single codec so the benefits will be moot for most.

If the Afterburner supports 3rd party codes it would be useful if 2 or more could be installed but I don't see that happening. CPUs and GPUs are rapidly getting faster and will dwarf this proprietary one-trick pony.

Those hoping this does more than codec work will be sadly disappointed. Look at the cooling, this thing is not very powerful.

d_2

Thinking...  maybe the AB slot has another purpose intended for the future. Some type of custom card with a pile of ARM processors to offload AI or other highly computational ‘side projects’, for lack of a better term. Almost like a server farm on a card.

cy_starkman

some serious flexing going on in the forum today.

the article trolled us all with its weird claim of FPGA and ASIC.

Apple allowing it to be field programable would be amazing and so not Apple like. let’s wait and see in the next year.

will it work in a EGPU box on a MBP, that will probably be answered in a month.

marcia funebre

(Sorry wrong thread, can't move this post) For me an Important question is: What does one have to do in order to move regularly bought SSD drives from previous external use into the slots of the new Mac Pro? (It is rather unsettling how little information there is from Apple.) One would assume that SATA is pretty standard but Apple loves to irritate with proprietary pitfalls...

mangakatten

Just a curious question, when will you need it at least? Sure it’s super helpful if you edit 8k-video. But for example say you buy the baseline Mac Pro and edit just a couple of 4K ProRes? I have no idea but I would guess the machine can handle that small load without problem? 

fastasleep

marcia funebre said:

(Sorry wrong thread, can't move this post) For me an Important question is: What does one have to do in order to move regularly bought SSD drives from previous external use into the slots of the new Mac Pro? (It is rather unsettling how little information there is from Apple.) One would assume that SATA is pretty standard but Apple loves to irritate with proprietary pitfalls...

fastasleep

marcia funebre said:

(Sorry wrong thread, can't move this post) For me an Important question is: What does one have to do in order to move regularly bought SSD drives from previous external use into the slots of the new Mac Pro? (It is rather unsettling how little information there is from Apple.) One would assume that SATA is pretty standard but Apple loves to irritate with proprietary pitfalls...

macsince1988

kirsch said:

On a more technical level, the Afterburner is a card with a Field Programmable Gate Array (FGPA), or a programmable Application-Specific Integrated Circuit (ASIC). This effectively means it is a card that has chips created for a specific task, rather than a general-usage chip.

An FPGA is exactly the opposite of chips created for a specific task. It is a chip that can be configured to do a specific task. The "Programmable Application-Specific" is an oxymoron. The way FPGAs work, an image is loaded onto them which reconigures the gates on the chip in order to do something specific. The image can be updated or changed completely by the vendor (Apple) after the card has shipped and is sitting on the customer's computer. They would do this to add new codecs, fix bugs, improve performance, etc.

It makes sense for Apple to use an FPGA for this and not an ASIC, which would be cost-prohibitive at the volumes this is likely to sell at.

Technically, a Field Programmable Gate Array (FPGA) is a subset of Application-Specific Integrated Circuits (ASICs).

The major types of ASICs are:

• Fully custom-designed chips, where all of the circuitry is created from scratch and the design is locked in during the manufacturing process.
• Cell-based chip designs, where much or all of the circuitry is created from a set of pre-designed cells (e.g., a CPU core or a GPU core that can be used in multiple chip designs) that are uniquely configured for that chip.  Again, the design is locked in during the manufacturing process.
• A gate array, where a common set of logic gates can be configured to create multiple customized designs.  Every ASIC created from a gate array is exactly the same chip for most of the manufacturing process.  The design of a specific chip is locked in during the final stages of the manufacturing process where the logic gates are connected to perform the functions of a specific chip.  Once manufacturing of a gate array is completed, it’s functionality can not be changed.
• A Field Programmable Gate Array is a variation of a gate array, using a common set of logic gates can be configured to create multiple customized designs.  The difference is that its functionality is not determined until AFTER the manufacturing process is completed.  After it is manufactured, an FPGA must be programmed (in the field) before it can do anything.  Because the functionality is not defined as part of the manufacturing process, an FPGA can also be re-programmed in the field if the functionality needs to be changed.

Although the common foundation of both a standard gate array and an FPGA enable them to be a cost-effective type of ASIC for low-volume applications, the field programmability of an FPGA also enables its functionality to be changed if the application’s needs change (e.g., a new codec or changes to a standard that the FPGA supports).

welshdog

For years I edited on a Quantel Editbox and later Henry compositing system. Both had huge FPGA cards that allowed Quantel to change features of the system without having to design a new board - which for years had been the only way they could make major improvements to their mahcines. One thing I think Apple could do here is work with software companies to offer Afterburner services to specfic apps. I would think 3D companies would be interested in using Afterburner to speed up certain processes. Who knows maybe even Adobe could use it to help and expand After Effects. Blackmagic Designs might be able to use it to speed up DaVinci.

As others have said, third parties probably won't be able to access Afterburner without working directly with Apple - that's probably a good thing. But it does strike me as having a lot of potential in the pro video world. Probably scientific work too.