Thoughts on the New Mac Pro

Posted on by Larry

[ This article generated a lot of technical comments. Be sure to view the comments in this blog to learn more. ]

Updated: June 15, to reflect a variety of technical comments from readers.

On Monday, Apple gladdened the hearts of power users everywhere by providing a “sneak peek” at the new Mac Pro. Stylish, diminutive, and blindingly fast – at least according to the specs provided by Apple. Since that time, I’ve been thinking a lot about a system that is directly targeted to meet the performance needs of video editors, and other power users.

First, keep in mind that this was a “Sneak Peek” — a tantalizing glimpse of what is coming in the future, not a formal product launch. (This is similar to what Apple did a couple years ago when they provided an “advanced look” at Final Cut Pro X at the 2011 NAB SuperMeet.) Consequently, while this “peek” provided an overview, it was intentionally sparse in providing details. Partly, I suspect, because Apple wants to gather feedback from potential users before nailing down the final specs.

HIGHLY CUSTOMIZABLE

One of the key things I realized was that this system is envisioned to be highly configurable. Just as the current Mac Pro has a wide variety of options for RAM, GPU, storage, and connectivity, this unit is envisioned to be highly customizable as well.

If you think about it, the current Mac Pro is the most customizable system that Apple makes. Configuration is at the heart of the new Mac Pro as well. While I expect that there will be one physical unit, we will have a lot of choices about what goes into that unit.

This also means that we will see a variety of price points as well, depending upon how each system is configured. In this regard, the new Mac Pro is identical to the current Mac Pro.

THUNDERBOLT IS KEY

Also keep in mind that Apple views Thunderbolt as more than a fast way to move data to and from a hard disk. Apple considers Thunderbolt as a direct connection to the PCI bus of the computer, able to deliver up to 20 Gb/second of data. Think of Thunderbolt as a direct line connecting the PCI bus to the expansion chassis of your choice.

NOTE: According to a reader, Intel is claiming a throughput of Thunderbolt 2 of about 1.6 GB/second, which is still very fast.

For most people, a fast computer coupled with lots of RAM and a really fast storage system will be all they need. In fact, Philip Hodgetts has written that more than 80% of Mac Pro users don’t have any PCI cards in their system; aside from the graphics card. For those users, the new Mac Pro fits their needs for raw power, without adding tons of unneeded expansion slots.

NOTE: We used to think of PCIe card performance in terms of the number of “lanes” they used to connect to the motherboard. There were four, eight, and sixteen lane cards. The more lanes, the faster the potential communication speed between card and bus. With Thunderbolt, Apple is moving away from the concept of lanes, to straight data transfer speeds.

Thunderbolt 2 is fully-backward compatible with the original Thunderbolt. Thunderbolt devices can be connected by either copper or optical cables. Copper cables can be up to 3 meters in length (about 10 feet). Optical cables can extend up to 100 meters, for users that want to store their computers or RAIDs in a machine room for security, noise, or air conditioning reasons. Currently, optical cable lengths of 10, 20, and 30 meters are available on the market.

For users that need to expand the capabilities of their computer, for example DSP audio cards, video ingest and capture cards, mini-SAS or eSATA cards, more graphics cards, a very real question becomes “how many card slots should the computer hold?” Apple felt that picking any number of internal card slots would be limiting to some number of users. By moving all expansion cards outside the box, then connecting with the very high-speed Thunderbolt 2 data bus, Apple essentially provided a virtually unlimited number of card slots for users that need the maximum in expandability.

NOTE: As a sidelight, one Thunderbolt 2 connection provides sufficient data bandwidth to ingest uncompressed 4K images, or output video to a 4K video monitor, or support VGA, DVI, and DisplayPort computer monitors. Plus Apple put an HDMI port on the back of the Mac Pro just for good measure.

Already, ATTO and Sonnet, along with others are offering Thunderbolt to “X” converter boxes: mini-SAS, FibreChannel, eSATA, Ultra-SCSI. And vendors such as AJA, Blackmagic Design, and Matrox offer ingest and monitoring options connected via Thunderbolt.

The one missing piece is the lack of high-speed Thunderbolt-native RAID 5 storage systems, with the notable exception of Promise. There are plenty of two-drive RAID 0 and RAID 1 systems, but very, very few 5 to 10 drive RAID 5 systems, which we editors need the most. I’ve heard lots of rumors of what’s causing the problem. Without pointing fingers, I hope this bottleneck gets resolved quickly.

MULTIPLE GPUs

We also need to consider that this is a system and not focus on one single element. The new CPU is twice as fast as the current Mac Pro in floating point operations. Memory bandwidth has doubled and now supports four channels of communication between RAM and the CPU.

The big news, though, was the addition of multiple GPUs. Although the ATI FirePros were featured, I suspect other options will also be available as part of the customization options Apple offers at launch.

Now, things get interesting.

On Monday, Apple made a point to say that Final Cut Pro X would release a new version that supports the Mac Pro. That instantly made me think that all applications would need to be rewritten in order to run on the Mac Pro, which would make this new system a non-starter.

This is not the case.

Instead, think of the dual-GPUs in the Mac Pro as similar to when Apple released multi-processor CPUs. All applications would run on a multi-processor system, but until they were re-written to support multi-threading (which is the technical ability software uses to take advantage of more than one processor) the application would be limited to using only one processor. This was one of the big limitations of Final Cut Pro 7.

NOTE: In terms of Final Cut Pro X, multiple GPUs offer significant performance benefits for real-time effects playback, rendering, optical flow retiming, and exporting.

So, the Mac Pro will run all current Mac software. However, if the software wants to take advantage of the dual GPUs, it may need to be reconfigured to do so. This is not a small task for developers, but it isn’t impossible. This is what Apple was referring to when they said a new version of Final Cut Pro X would be released to support the Mac Pro.

NOTE: Once developers know they can count of dual GPUs, they can design new software from scratch to take advantage of it, the way that everyone writes software today to take advantage of multiple processors and multiple cores.

UPDATE: A reader points out: “When using OpenCL, no code modification is required (problem only for Dev’s which don’t use OpenCL). Some use CUDA-API (Nvidia) – and this requires re-coding.

UPDATE: Another reader points out that the next version of Adobe Premiere and After Effects already support Open CL.

And the performance results of optimizing for dual GPUs can be astounding. Grant Petty, CEO of Blackmagic Design, tweeted earlier this week that they have been testing Resolve 10 on the new Mac Pro and it “screams.”

SUMMARY

Apple designed the Mac Pro as its most powerful and flexible desktop computer. They architected it to reflect where they see computers going for the next ten years. They provided a wealth of Thunderbolt ports – and converters – so that all legacy monitors, storage, and cards can be supported.

This has the potential to be an amazing piece of gear and I can’t wait to learn more at the launch.

As always, I’m interested in your thoughts.

Larry


75 Responses to Thoughts on the New Mac Pro

← Older Comments Newer Comments →
  1. Bruce Payan says:

    Concerning the Areca… for the price, consider a pair of FirmTek U3/Miniswap enclosures with RAID-0 configured SSD’s. For the money, it’s certainly a contender at about 600-700 MB/s.

    • Larry says:

      Bruce:

      For speed, you are correct and there are a LOT of RAID 0 devices. However, I like the redundancy of RAID 5 when editing a major project. And those are a LOT harder to find.

      Larry

  2. dr.no says:

    Larry,

    Think of GPUs as specialized Float Point Unit or DSP of previous generation.
    They are no longer for displays.

    In order to use GPU, Nvidia created a C and C++ API called CUDA.
    Apple in the mean time created OpenCL which is a generic version of
    distributed computing, it can use CPU and GPU at the same time.
    In order to do that Apple used LLVM’s JIT technology and compiler to create OpenCL.
    In the previous generation only OpenGL API could be used to create scene for Games and 3D visualization.

    In fact Nvidia was forced to use LLVM in order to compete with OpenCL JIT advantage.
    All the Supercomputers are using GPGPU programming to build there newer machines.

    The 128 GB quote is from a white paper that Apple released
    on Mavericks called OSX_Mavericks_Core_Technology_Overview.pdf

    “With its 64-bit kernel, OS X is able to address large amounts of physical RAM. OS X
    Mavericks has been tested to support up to 128GB of physical RAM on qualified Mac computers. “

    • Larry says:

      Dr. No:

      Thanks for the clarification. In your first post, you accidentally wrote 128 MB, which just confused the heck out of me.

      Larry

  3. dr.no says:

    Larry,

    Ok I went and checked the numbers.
    AMD W9000 FirePro comes with 6 mini display port
    so one GPU can handle a 6 – 30 inch displays at 16.4 million pixels.
    even W7000 can handle 4 display ports.

    3 4K is 24.88 million pixels.

    All this is job of display driver anyway. not individual programmer
    or software. That is what I am saying about FCPX having to change
    for newer GPUs.

  4. David Maier says:

    This statement shows up again and again on the net and it is so untrue. FCP7 was never limited to only one processor. It’s a multithreaded 32bit app and uses all cores and cpus just fine.

    • Larry says:

      Actually, David, it is true. Portions of FCP 7 – for example, Log & Transfer – which are the newer elements of the application, are multithreaded. However, the bulk of the program is not, which can be demonstrated by watching CPU usage in the Activity Monitor while performing routine FCP 7 functions.

      Larry

  5. Marcus Moore says:

    The number crunching over how much bandwidth the TB2 ports gives us is interesting- but I think people who’ll dismiss this machine because the external ports can’t support 16x cards are sort of missing the point: that the primary use for that bandwidth is Graphics cards which are already taken care of in the housing. The only question will be if the OpenCL oriented FirePros can be switched out for more CUDA oriented cards. If you can, then I fail to see the downside.

    We can’t say for sure until we get more detailed specs- but it’s entirely possible that the GPUs could be user replaceable.

    The market for external chassis has been limited- but as they become more popular obviously the configurations will increase and the price will go down.

    For FCPX users specifically this is going to be a wonderful machine. And I’m thinking that its release this fall might be timed to coincide with some great updates not just for X, but Motion, Compressor, and Logic too.

  6. Headscratcher says:

    It seems to me that posts pointing out, “you can easily hook up this or that,” are aimed more at justifying the configuration than at explaining why you would force the use of externals in the first place.

    The point of the big chassis with slots and drive bays was always a nod to the fact that the engineers can’t possibly foresee all the needs of the user, and that those needs evolve over time.

    Even if TB2 were adequate to current needs (and looking at the arithmetic it looks like there are some issues) it can’t accommodate everyone. And then there’s the problem of the rat’s nest of cables and power supplies that will need to be attached to the back. I don’t think too many people are going to be rotating those cases for long.

    I don’t doubt that many users will find a way to manage – but it’s puzzling at this point why it has to be like that. Perhaps this will only become clear when they ship. The only thing I can think of that would make the design choices palatable would be a surprisingly low price for the base machine. That might allow for higher volume sales and would encourage clustering for high end users.

    People have suggested a variety of thinks it resembles – looks like a sub-woofer to me.

    • Larry says:

      Headscratcher:

      These are also good points. The key, it seems to me, is how many people ACTUALLY put cards in their systems, aside from graphics cards. If the number is as low as others have indicated, then this becomes a small issue. If not, well, this would be a good time to invest in a company that makes cables.

      Larry

  7. David Maier says:

    Hi Larry,

    actually, watching activity monitor will show you it’s multi threaded. General RT playback isn’t all that great compared to more modern NLEs but that’s because of the architecture. Rendering, effects in the timeline, all show virtually equal usage of CPU cores. One can argue how well it’s multithreaded, same goes for QT apps in general because QT just isn’t doing so well, CPU usage hardly ever goes above 50-60 per cent, but you’ll never see just one core peak in FCP7 while others are idle, never.

    • Larry says:

      David:

      Hmm… interesting. I am a, um, devoted watcher of Activity Monitor and my experience is different from yours. For me, in most operations, FCP 7 peaks one processor 50-70%, raises the second 20% or so, and the rest show no movement at all.

      I’m willing to be wrong, but what you describe has not been my experience.

      Larry

  8. David Maier says:

    Hi Larry,

    that’s interesting indeed. I have never seen such behaviour with FCP7 on any machine.
    Here is a screenshot of activity monitor and a very simple timeline with some pip and color tinting applied.
    http://www.imagebam.com/image/9e047d260714481

  9. Ted T. says:

    As someone who use a 2009 MacPro *at home* for many purposes from professional software development and photography to hobby home theater I’m with Headscratcher in regards to disk drives rather than cards: I don’t want external enclosures. I have 14TB of internal disk space right now (out of a 16TB current max) plus two optical drives. Moving this to external cases will be a messy, ugly, loud business.

    Unless Apple were to offer a matching disk enclosure for users who care about noise, cable warrens and esthetics people like me will be in a quandary, and certainly will miss the old style Mac Pros.

  10. Rick Barrett says:

    RE: Marcus Moore I suspect that the GPU will be not be easily replaced. The cooling is the issue. The GPU’s have to connect to the heat sink. It might be possible but you’d be better of getting the GPU you want rather than thinking you could upgrade.

← Older Comments Newer Comments →

Leave a Reply

Your email address will not be published. Required fields are marked *

Larry Recommends:

FCPX 10.5 Complete

NEW & Updated!

Edit smarter with Larry’s latest training, all available in our store.

Access over 1,900 on-demand video editing courses. Become a member of our Video Training Library today!

JOIN NOW

Subscribe to Larry's FREE weekly newsletter and save 10%
on your first purchase.