Apple’s Phil Schiller previewed the new Mac Pro at the keynote of the World Wide Developer Conference in San Francisco on June 10th. Sure to annoy as many users as delighting them, the new hardware marks a significant departure from the currently shipping Mac Pro tower. The key discussion concerns seem to be around expandability and graphics card support. Actually, discussion isn’t the right word. I’ll call it what it is: the latest internet shitstorm to hit our industry since the previous darkest day in recent history: May 6, 2013 – the Adobe Creative Cloud announcement.
What Apple didn’t do with the new Mac Pro — and what seems to be the source of the greatest outcry among editors and artists — is create a traditional, slot-filled PC with plenty of expansion and user exchangeable parts. If you’re a user who needs to load up a machine with multiple cards or likes to add or remove your own internal hardware, you’re not going to be happy with the new machine.
Instead of making a computer that could be easily upgraded to work well for everyone, it is clear they are attempting to re-invent the PC with a tightly integrated package that will work *really* well for *most* everyone. Some argue that is Apple telling us what the user needs and forcing it upon them. They’re not forcing anything on anyone, but there are certainly situations where the new hardware will not be an effective solution. And there are alternatives for those users.
Some users who have reacted to the announcement have said Apple doesn’t know anything about the PC market. I think this hardware, as well as Apple’s business success says exactly the opposite. The fact that Apple generates more profit from PC sales than the top five PC makers combined says Apple knows a bit about what they’re doing outside of iPhones and iPads.
A followup argument to this is that Apple should license the OS to external hardware manufacturers in order to provide users what they need. They tried this in the dark days of Apple and it failed miserably. Having been the owner of a PowerPC clone, I know this first-hand. It didn’t work well. There are some who will disagree, but there are significant benefits to controlling both the hardware and software OS as Apple does.
There are quite a few more questions than answers at this point, but the fact that there aren’t any internal expansion slots (or user serviceable/changeable parts) shouldn’t be the end of the discussion. The fact is, there’s a lot to like about the new Mac Pro if you take a moment to examine some real world applications. For many artists out there, this could be the ideal machine. And as one software manufacturer said, in their short time with the new box, it is the “best out of the box performance” they have ever seen.
In a nutshell, let’s start with the facts (as we know them) about what is coming later this year.
- CPU: Appears to be one CPU (Xeon E5 chipset), based upon Phil Schiller’s comments during the keynote. With configurations offering up to 12 cores of processing power
- Memory: Looks like 4 slots are available. Includes a four-channel DDR3 memory controller running at 1866MHz. It delivers up to 60GB/s of memory bandwidth (vs 30GB/s for current MacPro)
- Graphics: Two FirePro GPUs in each system, supporting up to three 4K displays. State-of-the-art AMD FirePro workstation-class GPU with up to 6GB of dedicated VRAM.
- Internal Storage: PCI/E Flash storage, up to 1250MB/s transfer
- Peripherals: 4 x USB3, 6 x Thunderbolt 2, 1 HDMI 1.4, 2 x GigE
- Network: 2 x GigE, 802.11ac Wi-Fi
- Expansion (Thunderbolt 2): Six Thunderbolt 2 connections, each with 20Gbps bandwidth
The new MacPro is less than half the height of the curent tower and, according to Schiller, occupies “1/8th the space”. It seems to bear a remarkable heritage with the Apple Cube, which was groundbreaking at the time from an industrial design standpoint. It was a beautiful computer, but a bit ahead of its time. A tangible difference to the Apple Cube (of which I was an owner) is that the 2013 model has much more expandability due to Thunderbolt, so its built-in limitations are less of an issue. There’s a comparison to be made to another old-school Apple product, the Newton. It was also a bit ahead of its time, and could be considered the predecessor to the iPhone and iPad.
The new version is built around a triangular empty core in the center, with the CPU and two GPUs forming the three sides. Each side has heat sinks, which dissipates the heat from the sides through a fan in the top. Apple maintains the fan is quiet — and if it is at least takes some design cues from the retina MacBook Pro then there is hope that it will, in fact, be quiet. My guess it will be a very quiet workstation.
The computer rotates around on a base stand, revealing peripheral and expansion ports on the back, with a nice touch of lighted outlines around the ports, allowing users to easily see the connection points. The functionality has been a bit mocked in online reactions, pointing out that the idea of having all the wires connected and then rotating will easily create a spider web of cable tangle-ness. Those folks are free to pick up and rotate their MacPro tower around by hand to connect things. The rotate option seems better to me.
As far as industrial design, I’ll make a judgement when I see it in person. Trashcan, R2D2, whatever. After the iPad announcement, many made fun of it, calling it a big iPhone with name that sounded like a women’s hygiene product. In the end, how’d that work out for Apple?
We mention the form factor, because it does lead to an lack of internal expandability. Apple calls it “our most expandable Mac yet,” but that’s a bit disingenuous as it is only expandable in an external sense via USB3 and Thunderbolt 2. It appears as though no internal parts will be customizable by the end user. So expansion in traditional terms? Not so much.
What Apple has done here is effectively say…look, we’ll give means for external expansion to those users who need it. But what they didn’t do is build a box with lots of internal slots and a beefy power supply to support those cards — only to have many of their customers not use the expansion capability. As I mentioned earlier, many feel this is forcing their customers to do something, but one could also argue it’s simply a different take on what a workstation means.
Through its lack of internal expansion, the hardware doesn’t support PCIe3 expansion, which could limit its use moving forward. But speaking in practical terms, other than an external GPU, what limits does Thunderbolt 2 place on real world hardware? A four drive RAID 0 6G SSD SAS array can provide approximately 1.5GB/sec in bandwidth. This fits well within the 20Gbps (~2.5GB/sec) theoretical limit of Thunderbolt 2, so we can expect similar performance from a Thunderbolt array. Apple bills 4K prominently in the graphics section of their Mac Pro page, so that deserves some attention. For 4K, you’d need close to 4Gbps for 12bit 24fps uncompressed playback (double that for 48fps), so both would also be well within the spec for Thunderbolt 2 connections. So our takeaway is that Thunderbolt 2 will be adequate for all but a few edge cases other than housing a GPU in the chassis.
It’s also important to note that the lack of internal expansion means that users will be spending money to add external enclosures. These expansion boxes range from $400 for half-length boxes to $1,000 and more for larger boxes.
It appears as though the new Mac Pro has four slots for RAM, which look to be user installable due to their position inside the case. As with all things “New Mac Pro”, this is an educated guess as well. While the new DDR3 RAM running at 1866Mhz is a speed king, the maximum size for currently shipping ECC modules near that speed is 16GB. This means that a maximum of 64GB looks to be able to be installed in the system at this point in time if you want to use matched RAM. There’s slower RAM available, including load-reduced RAM, but that might cause issues.
64GB could be quite limiting for many (but not all) high end applications. Fast flash storage could change the dynamic of RAM v. “hard drive” space and what is important, but this is an initial red flag. This situation may change by the time the product ships later this year, but it is certainly a point to pay attention to moving forward.
Built-in storage on the Mac Pro is limited to PCI/E flash storage which provides about 1GB/s of transfer speed. So while storage space will be limited compared to adding hard drives, there is a *significant* improvement in performance when moving to flash. This is a night and day game changer for performance. One key will be how much storage is included (and how expensive it is). The bump from 512GB to 768GB on a Retina MacBook Pro is about $400, so having over 1TB of storage means close to a $1,000 price tag. But with the fast transfer speed, this could be significant speed benefit for everyday users for things such as After Effects hash cache. The jury is still out as to whether the storage is upgradable, as with previous products such as the MacBook Air or Retina MacBook Pro.
What about large scale storage, such as what we use currently with the built-in hard drives in the tower Mac Pro? If you want more storage, you’ll have to use drives externally via USB3 or Thunderbolt, which means an extra connection as well as external power.
We’ve seen a lot of posts from people saying that more connections means more points of failure and that’s not a risk willing to take. While true that it’s a risk, I feel the risk is overblown. Externally connected RAID systems are incredibly common on desktop systems via fibre, thunderbolt, or esata. On top of this, we have breakout boxes for video i/o and other things. I’ve been using external peripherals such as the Promise R4 array and the Blackmagic Intensity Extreme without issue with my Mac Book Pro. In fact, the benefits of being portable and only having to buy one Thunderbolt peripheral and share it among multiple systems could a big cost savings.
Frankly, saying the tangle of multiple cables is ugly might be more of a valid concern. But not a big concern. As far as external drive chassis, I can see manufacturers making units that would fit nicely under the Mac Pro , similar to expansion peripherals that work with the Mac Mini.
The new Mac Pro will ship with one Intel Xeon E5 chipset CPU, available with up to 12 cores. According to Apple, this single chip provides “Up to 2x faster speed” than the currently shipping Mac Pro. We’re a bit perplexed by the inclusion of only one CPU. Yes, it’s much faster than the current generation. But for RAW image processing power and hard core rendering we can see real benefit from having a multiple-CPU system. More cores = faster processing.
It doesn’t seem as though cost would be deciding issue for Apple when deciding whether to include dual CPUs. From our research, it appears as though the cost of the dual GPUs would be greater than dual CPUs. And this pales in comparison to the cost of the flash storage. A sensible reason for not having more than one CPU might be heat generation and dissipation. Perhaps the form factor and layout make having two CPUs out of the question. But our hope is that one day we’ll see the addition of another CPU, at least as an option.
That being said, there are a couple of things to keep in mind. We’ve been following GPU tech for years here at fxguide, and the fact is that the performance to watts/heat ratio on GPU is far superior to what we’re seeing in CPU development. In other words, the general purpose processing power is increasing much more dramatically on the GPU, while at the same time reducing power consumption and heat generation per processing cycle. Time and time again, presentations at the NVIDIA GPU Tech Conference made this clear.
At the user end, we’re seeing more and more reliance on the GPU for processing power in all of our applications. This is just starting to be tapped. From OpenGL shaders being used to do grading and effects in Autodesk Smoke and Adobe After Effects, to The Foundry using their blink technology to support identical processing on both the GPU and the CPU, to Adobe Premiere Pro and Blackmagic Design’s Resolve utilizing CUDA and OpenCL processing….it is certainly the trend in the industry. The fact is, much of the heavy lifting of image processing is being done by the GPU today and is likely to be more prominent with the shift to 4K. This doesn’t mean the CPU is irrelevant by any stretch, but the GPU may prove to be more important for overall performance in many applications.
One other problem of relying on pure CPU benchmarks and horsepower as a litmus test for a “pro system” is that it doesn’t fully take into account how we work with software on the job. By this I mean that the amount of time that many of us are doing full up processing and maxing out the CPU and GPU in an app can be quite minimal. We spend a lot of time in the app simply working with the software in various ways: setting up a vfx composite, doing keying, masking, kerning of type, color correcting, animating, etc. During that process, we’re generally not maxing out the CPU or GPU by any stretch. It’s one reason I feel that my Retina MacBook Pro *feels* much faster than my Mac Pro tower. Even though the Quadro K5000 and more CPU cores render faster…due to the tech (flash storage, decent Quadro mobile GPU), it feels as though it keeps up.
Before you jump on me and say that’s bullshit…yes, I understand that this lack of full CPU/GPU utilization is not always true. And I would very much like to see a dual CPU system in the future. Editing multiple streams of 2K and higher video, debayering RED footage, applying realtime LUTs to footage, complex grading pipelines in Resolve, background renders in After Effects — all those things can max out your system. And that’s why having two CPUs would be a welcome addition to the new Mac Pro. But the point is not every “pro” user has those concerns, and for many of those users the new hardware will be big leap forward…because the overall hardware *is* a leap forward.
There is no built-in optical drive.
The new MacPro has not one, but two ATI FirePro GPU units built in. This is a critical feature and one key spec that is incredibly important from a “pro” standpoint. And these don’t seem to be entry level GPUs. In trying to guess the level of GPU that Apple is including in the new MacPro, it looks as though the specs seem close in performance to something like the ATI Fire Pro S9000 card (which has 6GB VRAM and 3.23 TFLOPS of performance). This isn’t your mother’s Radeon. These seem to be at the top end of the ATI product line, performance that OS X users haven’t had access to in the past due to limited support. It’s also a good thing they’re top of the line, as from the design of the computer, it looks as though it is a custom GPU hardware build in order to fit in the case. In other words, not user-replaceable.
The inclusion of two cards by default is a big step forward and a critical aspect of the base system. While PCI expansion is available through a Thunderbolt external enclosure, the spec doesn’t allow using an externally housed GPU card as a display device. You can use a GPU card in an external Thunderbolt expansion chassis for CUDA or OpenCL processing. But even with the 20Gbps improved bandwidth of Thunderbolt 2, it still isn’t high enough to matter. In most situations it simply doesn’t make sense to offload CUDA/Open CL processing to an external GPU card due to the limited bandwidth. Having the GPU (or GPUs in this case) connected to the RAM and CPU by the fastest means possible is key to maximizing processing power.
The NVIDIA issue
One big outcry voiced among pro users after the WWDC keynote is the apparent lack of an NVIDIA graphics chipset in the new MacPro. While this could be problematic for many due to the lack of CUDA support — and let me just say at fxguide, we’re huge fans of NVIDIA’s GPU tech — recent improvements in the ATI product line has led many software manufacturers to beef up their OpenCL support. It’s important to note that OpenCL performance is not up to par with CUDA performance in many implementations, and OpenCL programming between NVIDIA and ATI isn’t as simple as using the same code twice. In other words, manufacturers still need to verify and bless the various graphics cards.
The lack of CUDA in the Mac Pro could actually lead to an improvement in the OpenCL situation and in the end lead to less dependence on one manufacturer for GPU computing. But in the end, who really knows? In this specific case, emerging OpenCL support allows Apple to only offer a “custom build” AMD GPU and still give pro users much of what they need. All this being said, we would prefer to have options on the GPU front…and would very much welcome the addition of NVIDIA hardware in the new Mac Pro.
A real shortcoming will be the lack NVIDIA support for certain After Effects users. Not for the majority use cases, since all but one AE GPU accelerated feature is supported with both CUDA and OpenCL. But it’s a big one: If you want an effective workflow, Adobe requires an NVIDIA GPU to do fast raytracing. By this I mean that while you can do raytracing utilizing the CPUs on your system, it is simply too slow to rely on as a creative technique in real production.
With the new Cinema 4D support in CC as well as options such as Element 3D, it may not be as much as a showstopper. But in a collaborative environment, do you really want to take a chance that you’re not going to be able to render something that another artist can? Probably not. Artists have been asking for Adobe AE to support raytracing on more than NVIDIA cards since the day it was released. This will hopefully give the team at Adobe reasons to do so in a timely manner, though I’m sure it won’t be a simple or straightforward process.
One possible workaround is that even though you can’t run a display device off a GPU card in an external Thunderbolt PCI expansion box such as the Sonnet line, you can (in theory) run a CUDA card for general purpose processing. The OS should recognize the GPU in the external box, and therefore After Effects *should* be able to use it for raytracing. With Thunderbolt 2, bandwidth shouldn’t be an issue for something like the AE raytracer, since you’re not really streaming a lot of imagery or layers in real time. You’re basically loading images to the card in buffers, doing the render (which takes the majority of the time), and then offloading the images.
These negatives being said, we do have concrete examples of “pro” apps which have made significant changes to support ATI and reduce the reliance on a single GPU manufacturer. And give support to the idea that having ATI-only cards — and serious high level cards — is OK in the long run. Maybe even preferred.
First, the upcoming version of Adobe Premiere Pro CC dramatically increases support for AMD cards. The shipping software has support for OpenCL in a wide variety of ATI Radeon and FirePro offerings. In addition, the two built-in cards could dramatically increase performance for export, since according to Adobe “configurations containing multiple GPUs, Premiere Pro CC can use all of them during export.”
Next up, Resolve. There were lots of initial complaints on Twitter from Resolve users because in the past, there have been performance and tool benefits to the app’s NVIDIA CUDA processing. But Blackmagic Design’s Grant Petty was quick to postthe company’s thoughts regarding the new Mac Pro, and it’s incredibly positive:
We have been testing with DaVinci Resolve 10 builds and this screams. Its amazing and those GPUs are incredible powerful. I am not sure what I can say as I am only going off what Apple has talked about publicly here in the keynote for what I can say right now, however there is a whole new OpenCL and DaVinci Resolve 10 has had a lot of performance work done to integrate it and its really really fast. Those GPUs are very powerful and have lots of GPU memory so this is the Mac we have been waiting for! We have lots of Thunderbolt products too so video in and out is taken care of.
Petty’s full post can be seen in the Blackmagic forums.
Finally, MARI. Phil Schiller mentioned that The Foundry would be showing off one of their products in a special session on Tuesday. We learned from The Foundry that they would be showing a tech preview MARI running on OS X, which has been in development for a while. According to product manager Jack Greasley, they’re getting really good performance on current hardware. As part of the development process, they had been pinging Apple tech support and asking some very pointed questions about OpenGL, performance, and throughput. About six weeks ago they got a call from Apple and they asked if they wanted to come in and have a chat….
The team from the Foundry was allowed to work with the new Mac Pro at Apple, albeit shrouded in a large cabinet so they couldn’t see the actual form factor or know the actual specs. According to Greasley, “it is absolutely the best out of the box performance I’ve ever seen. So in terms of not being a custom bit of hardware, it’s incredible.” The thing that is quite interesting that people have been glossing over in discussing the new Mac Pro is the 1GB/sec data throughput from storage into memory, says Greasley. “That just makes a huge difference…it feels like a really balanced system. It’s not as if the GPU is faster than the RAM can keep up with it…it’s effectively a system designed for 4K…it kind of blew me away, actually.”
MARI Product Manager
In the WWDC keynote, they showed a still from PIXAR’s Monsters U — and those were the test case assets that The Foundry team pulled over to try out the new hardware. “It was full animation, lots of textures, lots of data,” in the scene according to Greasley, “and we were playing back the animation at 60 frames per second in raw per-frame animation.”
The hardware requirements for MARI aren’t all that advanced; the baseline card is a state of the art card from three years ago. But the main issue The Foundry is seeing in MARI is data throughput, especially at the high end facilities. This is because MARI processes everything in the GPU and you’re accessing more data than you have RAM, so you hit throughput bandwidth getting from the file system to RAM to the GPU and back. Having fast access to flash memory and even Thunderbolt 2 will, in theory, help make processing and interactivity even more efficient.
The presentation by The Foundry will be available on the Apple Developer web site.
We really have no idea at this point as to pricing, but looking at the parts included it’s unlikely that it will be cheaper than current Mac Pro offerings. The inclusion of two GPUs as well as the flash storage will most likely take away any cost benefits from less materials and a smaller size.
Takeaways from the announcement
It’s clear from the WWDC announcement that users looking for an old-school MacPro update are disappointed. At least in the Twitterverse, the lack of internal expandability has upset a lot of artists and editors. Even with the options for external expansion such as Sonnet’s Echo Express expansion line or Magma’s ExpressBox 3T, the expansion isn’t appropriate for them.
For those users, thankfully they have fairly easy access to other options. For true tech-savvy folks who like tinkering and overclocking their GPU, making custom hardware builds, and more, there’s a fairly vibrant Hackintosh community on the web. For others, Windows systems are an option. The Adobe suite as well as Media Composer run incredibly well on Windows hardware and within the apps it feels the same as working on OS X. MARI, NUKE, and other Foundry products are also available on Windows.
As far as further shortcomings in the upcoming Mac Pro, since Apple is building for the future, our feeling is that 10GigE would have been a solid base upon which to build from a networking standpoint. Granted, current implementation can be a bit sketchy, but if you’re truly touting cutting edge performance for professionals, network collaboration is critical.
We do also have concerns regarding the form factor and large scale installs of the new MacPro — such as a machine-room install of multiple systems feeding rooms around a facility. While today’s MacPro isn’t the easiest tech room install, the form factor does lead to sensible rack mounting on a large scale. Storage of the new Mac Pro will be problematic to say the least, as they don’t seem to be incredibly friendly from a large installation standpoint. And while I poked a bit of fun at the tangle of cables that will come off the back, this will also prove problematic for installs. Great design and functional design aren’t mutually exclusive.
Potential use cases
So what about the systems and how they might fit into facilities? A quick check in with several sysadmins and engineers today found that most graphics or vfx artist OS X workstations have only a decent graphics card, lots of RAM, and a large amount of storage (built-in RAID or attached enclosure). Artists at the “next level up”, so to speak, have some kind of broadcast monitor as well. These two users are probably the sweet spot for the new MacPro — though the lack of After Effects raytracing capability may be a deal killer for many. It seems as though the following would suit many artists quite well as a baseline system:
- Promise Pegasus R6 Thunderbolt RAID
This would provide local storage performance via the thunderbolt RAID and could even take advantage of the HDMI output for use as a calibrated broadcast monitor output in many situations. If it’s not enough, the following accessories could be added which could provide monitoring and I/O:
This seems like a very sensible, if not great, workstation for a great number of artists — whether they be motion graphics designers, vfx compositors, or even some editors. Autodesk Smoke on the current shipping iMac line is fast — faster than the flagship Flame software even. If it is fast on the iMac, it’s gonna be even faster on the upcoming Mac Pro. In fact, just last week, Autodesk’s Stig Gruman alluded to the fact that a system was coming out that would be great for their Smoke product.
What if you’re editing and/or working in a collaborative environment on a variety of footage? Editors and artists doing color grading might be examples of this. A key consideration in this area might be the need to connect to a fibre channel network — or offload processing of R3D files to a RED Rocket card. Take the next step up and go with an external Thunderbolt to PCI expansion chasis to provide access the following:
- Echo Express expansion chassis
- RED rocket card
- ATTO Celerity Fibre Channel card
- Fusion ioFX
I would suggest that this setup would be approaching the “high end” of workflow in the professional vfx and post field. The cards in the chassis alone cost around $10,000. Having three Thunderbolt 2 controllers (six channels total) means that you can hook RAID storage off a separate controller from the expansion chassis. While the improvements in Thunderbolt 2 (double current gen) break down some of the previous barriers to performance. The expansion chassis use case does fall down in a large scale workflow environment, where you might have dozens or even a hundred workstations. For starters, there will simply be a considerable cost in outfitting all of these with expansion chassis
For instance, with current shipping Thunderbolt products, the Fusion ioFX sees less than optimal performance due to the fact that the interface is limited to 1 Gigabyte per second. But with Thunderbolt 2, the full 1.4 Gigabytes per second for the ioFX should be attainable. The drawback to using the ioFX in this way is that the technology is not hot-swappable. You must plug in the Thunderbolt cable before powering up….and disconnect the ioFX after powering down. Hopefully, this can be improved the in the future but the main takeaway should be that throughput becomes less of an issue with the new hardware.
We have a lot of time to ponder the new hardware
It’s important to spend some time examining what you as an artist truly do on your box and not jump to quickly dismissing the new hardware from Apple. And thankfully, since the Mac Pro won’t be shipping for at least six months, you’ll have time to do so.
It’s certainly not for everyone – machine room installations, maxed out primary workstations that need many slots, and more. This new offering won’t provide a new path for those types of installs. And while I poked a bit of fun earlier about the tangle of cables, this certainly something the could be problematic when added to the fact that the Thunderbolt connector itself is not the most secure connection.
But upon closer examination, the hardware is nowhere near as bad as the initial (over)reactions from the online world convey. In fact, I’ll go further: it’s definitely not bad. In fact, it’s great for what I need to do as an individual artist. Get beyond the misguided/misinformed information posted by many on the net, and you’ll likely find it is actually a positive move forward for a majority of creative artists who are currently using OS X. And you don’t have to take my word for it, you already read what some of the software manufacturers have to say about it.