Listen to the audio version:
On a recent episode of the Embarrassingly Parallel Podcast, I walked through my argument to myself on why I shouldn’t buy this iteration of the new Mac Pro. Taking into consideration the hardware on the new Mac Pro in relation to technology trends and upcoming new technologies (such as 4K), I made the argument that the Mac Pro is – despite it being “brand new” – actually at the tail end of a lot of different technologies. Buying an expensive computer designed for professionals isn’t something you’d want to do if it’s already at the end of the line.
But don’t get me wrong – the new Mac Pro is a fantastic piece of design and engineering from Apple, and in a fruit-filled metaphor, Apple took the lemons from AMD and Intel (previous generation graphics and CPU architectures, respectively) and made them into Mac Pro lemonade. In fact, aside from the design decision to remove internal expansion (which is now dependent on Thunderbolt) and the move away from dual-socket CPUs, there is almost nothing about the Mac Pro in terms of design decisions / component choices with which I could take issue. It is, both physically and conceptually, a very compact and well-balanced product.
So, with that high praise for Apple’s bold rethinking of the professional desktop, why do I so strongly want to be able to convince myself to buy this generation of the new Mac Pro?
- Dual workstation-class GPUs
- Up to a 12-core CPU
- Up to 64GB of RAM
- Six Thunderbolt 2 ports & USB 3.0
- Up to a 1TB PCIe x4 SSD
- Heatsink and fan cooling system from the future
- Support for 4K monitors
- IT LOOKS LIKE A DARTH VADER TRASH CAN!
But as I began to pull at the tightly woven threads of Apple’s carefully crafted vision of the future, the awesomeness of the total package started to unravel somewhat in light of the components that Apple had available to work with. Let me clarify that statement: this is not a critique of Apple’s design choices or how creative professionals would use this machine. There have already been a lot of great articles written on the subject (Anandtech and Ars Technica being two stand-outs), and I’d be doing them a disservice by trying to replicate their efforts here. Because as we all know, Replication is Futile.
What I can do is expand upon what Anand Shimpi hinted at in his review of the Mac Pro:
If there’s one graph that tells the story of why Intel’s workstation roadmap is ridiculous, it’s this one. [link] The Mac Pro follows Intel’s workstation roadmap, which ends up being cut down versions of Intel’s server silicon, which happens to be a generation behind what you can get on the desktop. So while the latest iMac and MacBook Pro ship with Intel’s latest Haswell cores, the Mac Pro uses what those machines had a year ago: Ivy Bridge. Granted everything else around the CPU cores is beefed up (there’s more cache, many more PCIe lanes, etc…), but single threaded performance does suffer as a result.
The shiny new Intel Xeon CPU in the Mac Pro isn’t any faster than what ships in the consumer-oriented – and significantly cheaper – iMac in terms of single-threaded performance. And while there are again already a lot of good comparisons between the Mac Pro and the iMac (and which one you should buy based on your use case), I’m not going down that path either. But the point about the older revision of CPU that Intel ships for its Xeon got me thinking:
What else is older than it seems in this new Mac Pro?
With some digging through Wikipedia articles, I drafted a fairly compelling (at least for myself) argument for why the current Mac Pro isn’t as new as it looks. But again, based on what Apple had to work with, I’m not writing off the Mac Pro entirely. I’m just waiting for the Mac Pro – Mark II. So, let’s place the components of the Mac Pro into temporal context:
- Although the Ivy Bridge-based Xeons used in the Mac Pro were released recently (September 10, 2013), the Ivy Bridge family of processors first started appearing on April 23, 2012, almost two years ago.
Furthermore, Ivy Bridge is (mostly) a die shrink to 22nm from 32nm Sandy Bridge, with no IPC (instructions per clock) efficiency improvements.
So, the actual chip architecture is based on Sandy Bridge, which was first released January 9, 2011 – three years ago.
Mac parts reseller OWC demonstrated that the CPU in the Mac Pro can be successfully upgraded, but it’s a hollow victory. New processors to be released by Intel (next generation Haswell-EP) will use an entirely new socket and won’t work in the new Mac Pro.
The Chipset / CPU Socket
Based on the X79 Socket 2011, released on November 14, 2011 – is over 2 years old.
No native USB 3.0 (Apple had to use a separate Fresco Logic USB 3.0 controller to add functionality).
Bandwidth limited on PCIe lanes for maxed-out USB 3.0 throughput.
No new CPUs (besides revisions of Ivy Bridge-EP) for socket 2011.
The Graphics Cards
Tahiti-based GCN 1.0 (Graphics Core Next) GPU architecture was released on December 22, 2011 – over 2 years ago.
Not current with recently released high-end Hawaii GCN 1.1 graphics cards such as the Radeon R9 290X.
XDMA: fixes AMD’s frame pacing issues and increases bandwidth by connecting GPUs over the PCIe bus instead of CrossFire Bridge. Not available on the Mac Pro.
Performance is same as two lower-clocked Radeon R9 280X cards in CrossFire in Windows, but Mac OSX games only utilize one gaming GPU unless specifically optimized for use in the Mac Pro (which is very unlikely).
The Current 4K Situation
AMD’s GCN 1.1 fixes much of the frame pacing issues (seen in multi-GPU CrossFire configurations), but the new D-series GPUs in Mac Pro are GCN 1.0 and might never get a lasting fix for 4K across both GPUs.
While not an issue for normal desktop use, it is unusable for gaming at 4K resolution due to huge variance in frame times, and might impact video playback frame rates.
Currently there is spotty support for 4K monitors on the new Mac Pro (of course, the affordable ones from Dell don’t seem to work).
4K “Retina” display mode currently not supported, and if it were, the screen real estate would be less than that of a 27″ iMac (non-Retina). So despite it being a powerhouse workstation, the screen size for professional projects in Retina display mode would actually feel smaller than a 27″ iMac. Although, there are some theories on how Apple might sidestep this issue with 4K and Retina.
The RAM is less of an issue than some of the other limitations, as DDR3 1866 ECC is fast, error-correcting memory – and one of the only user-replaceable parts of the Mac Pro (pictured is a 64GB Mac Pro RAM upgrade kit from OWC). But DDR3 was first released in 2007, well over 6 years ago.
DDR4 is expected to ship in 2014 along with Haswell-E (the enthusiast desktop consumer line) and Haswell-EP (the successor to Ivy Bridge-EP in the Mac Pro).
DDR4 uses one DIMM per channel (point-to-point), and being quad-channel would work perfectly for Mac Pro with 4 RAM slots – not to mention the power savings and the likelihood (given historical clockspeed improvements) of it being faster than DDR3.
Possible teething issues with any new design (this one doesn’t really worry me too much, though).
Can the new “thermal core” stand the test of time for effectively cooling the CPU and dual GPUs? (Again, not really worried much since this is a workstation-class tool designed to be pushed to the limits for extended rendering).
Software exploitation of dual-GPUs. This is perhaps Apple’s biggest wildcard with respect to the design of the Mac Pro. It’s still too early to tell if this will spread to other applications other than in-house products from Apple (such as Final Cut Pro X, Logic, and Aperture). The move from a dual-socket CPU & single GPU to the new single-socket CPU & dual GPU is a big change for how software (professional or otherwise) is optimized. The first generation of the new Mac Pro will need to demonstrate that this paradigm shift in the CPU / GPU balance will win over the software developers who can make or break the usefulness of the new Mac Pro.
The Old-New Mac Pro
In summary, here’s what you get in a brand-new Mac Pro:
- 3-year old CPU architecture
- 2-year old CPU socket / chipset
- 2-year old GPU architecture
- Limited support for 4K
- 6-year old memory standard
- Questionable software exploitation of dual-GPUs
At this point, I can understand if you take my analysis to be a nit-picking critique of an impressively innovative redesign of a professional desktop. But it’s not for lack of effort or ingenuity on Apple’s part that the Mac Pro’s nascent existence is predicated upon the old bones of workstation-class components. As I stated at the beginning, I’m not criticizing the Mac Pro in general, just this first iteration of the Mac Pro. And only because the available workstation-class options from Intel and AMD aren’t exactly the latest and greatest technologies. (There is also some unfortunate timing with the imminent arrival of DDR4 RAM and the beginning of mainstream 4K). It is not, to be clear, because Apple built a poor first version of the Mac Pro.
So, what will the next Mac Pro look like? Will it resolve all of my qualms with the current Mac Pro? Who knows! Unfortunately, as the quote from Anand Shimpi illustrated, Intel’s release cadence for server CPUs will likely continue to lag their consumer versions in terms of single-threaded performance (due to using an previous CPU version). But the perfect storm of timing issues and older technologies that beset the current release of the new Mac Pro might not conspire so overwhelmingly for Mark II, whenever that might be. And from Apple’s notorious history of sporadic releases of new revisions of the Mac Pro, it’s unlikely that we can predict when a new version would arrive either. So many questions, so few answers! But the speculation and rumors are half the fun, after all.