Ahead of our full review of the Apple Mac Pro (Late 2013), we have begun running some preliminary benchmarks in both OS X Mavericks as well as Windows 8.1 (64-bit). As we noted in our
first look at the Mac Pro, our unit is the high-end standard configuration with a six-core Intel Xeon E5 processor clocked at 3.5GHz, matched with 16GB of RAM. As with all Mac Pro models, it comes fitted with dual AMD FirePro GPUs, which in this instance is a pair of FirePro D500s, each with 3GB of DDR5 VRAM. Our early benchmarking has yielded some interesting results, particularly in terms of the Mac Pro's graphics performance -- helping to clarify why the Mac Pro is classed as a 'workstation,' as opposed to a high-end 'desktop' computer or even gaming rig.
In our upcoming full review, we will run the Mac Pro up against our 13-inch MacBook Pro 2.5GHz (Retina Early 2013), and a 27-inch iMac (Late 2013), which we expect to receive from Apple shortly. In the meantime, we have run the Mac Pro up against the MacBook Pro, which it naturally leaves well and truly in the dust. However, while the MacBook Pro might not be much competition, the benchmark results of the Mac Pro raise some interesting discussion points.
For this sneak peak, we have focused on Cinebench R15 and Geekbench 3, which are are cross-platform compatible -- so we have also put the Mac Pro through its paces running Windows 8.1 using Bootcamp. We have also run the Mac Pro through Futuremark's 3DMark, which currently only works in Windows 8 (and iOS and Android), but not OS X -- it yields some interesting insights into the Mac Pro's graphics capabilities.
According to Maxon, Cinebench R15 is a real-world cross platform test suite that is based on its Cinema 4D software that has been used for making 3D content in films like
Iron Man 3,
Oblivion,
Life of Piand
Prometheus. The Open GL graphics test (which was our focus here) assesses the graphics card to deliver a result scored in frames per second (fps). The test pushes the GPU through one million polygons while measuring its geometry performance, along with textures and various effects. As you can see in the graphs below, the Mac Pro with D500 achieves a GPU score of 80.13fps. While it is one of the best results we have seen on a Mac, it will likely be outpaced by the dedicated (although mobile) NVIDIA GeForce GTX 780M graphics with 4GB GDDR5 that is available as a CTO option on the current iMac, along with the FirePro D700 upgrade option for the Mac Pro.
Interestingly though, when we ran the Mac Pro through Cinebench R15 in Windows 8.1 with CrossFire enabled, the Mac Pro blitzed the same test with a score of 106.13fps (CrossFire splits the graphics processing duties of a display in two, with each powering one half of the screen). Both Maxon and Apple have confirmed with
MacNN and
Electronista that Cinebench R15 is not yet optimized address both FirePro GPUs under Mac OS X. However, it is possible that it might not be optimized to use both GPUs for graphics processing, depending on how Maxon feels the two GPUs are best ultilized for running Cinema 4D in OS X on the new Mac Pros.
The GPUs in the Mac Pro are not designed for just graphics performance, but also for use in computing tasks thanks to Apple's implementation of Open CL in OS X Mavericks. When you learn that each of the FirePro D500s has 4.3 billion transistors, you begin to realize that there is a lot of computing power there that can be harnessed for running certain types of calculations, beyond just graphics processing. The Mac Pro utilizes the FirePro GPUs for professional apps like Final Cut Pro and DaVinci Resolve that can use the second GPU for off-screen rendering and computing.
If you're a gamer, you might wish that Apple and AMD work to enable CrossFire in OS X, but general computing is not what the Mac Pro is designed for (even if it remains excellent in that context too). If Apple had chosen high-end AMD gaming-oriented graphics card, a single one of these could out comfortably outscore one of the FirePro D500s in the Mac Pro, or even the higher-end FirePro D700. The AMD FirePro GPUs are designed for both graphics-intensive 3D applications as well as computing tasks, and are based on AMDs Graphics Core Next technology. This means that they are specifically designed for applications that utilize Open CL-enabled systems (like the Mac Pro), tasking highly complex computations between both the CPU and the GPU.
The reason the AMD FirePro workstation graphics cards are so expensive (each of the D500 cards in the Mac Pro, the equivalent of a W8000, sells for well over $1,000), is that they have been designed for 'mission critical' tasks and applications that take advantage of double precision numerics. This doesn't mean that you can't get solid gaming performance from the Mac Pro –- you can, as we have found when playing Windows games in 2560x1080 at ultra-high settings averaging around 60fps on recent titles like
Batman: Arkham Origins -- but it is a side-benefit, not an aim.
In fact, the Futuremark 3DMark benchmark for Windows 8.1 attests to the solid gaming credentials of the Mac Pro -- it also reinforces that although the GPUs in the Mac Pro are high-end workstation cards, they are not going to threaten high-end gaming rigs and their respective GPUs in any way. The 3DMark (which assesses GPU performance across DirectX 11, geometry, illumination, particles, GPU simulations and a combined CPU/GPU stress test) compared the performance of the Mac Pro to scores achieved by gaming rigs. It revealed that the Mac Pro performs similarly to typical upper mid-range dedicated gaming cards.
As a side note, one of the tests in the suite is actually a cross-platform test, making it possible to directly compare the graphics performance of the iPad Air and the Mac Pro. In the common Ice Storm test, for interests' sake, the Mac Pro scores 139573 points, against the iPad Air's score of 15269. Where the iPad Air reached a peak frame rate of 104fps (exceptional for a current-gen tablet), the Mac Pro's dual D500 GPUs (with CrossFire enabled) punched through the same test at a peak of 1235fps -- not unexpectedly, of course, but interesting nonetheless.
The Geekbench 3 results produced by the Mac Pro in OS X and in Windows 8 were, as expected, roughly equivalent -- although there is a slight performance edge in OS X. If you are still wondering what makes the Mac Pro a workstation and not just a regular desktop, the multi-core performance should help to clarify your understanding further.
In terms of outright speed on single-threaded applications that tax just one core of a system's CPU, the new Mac Pro is certainly fast, but not staggeringly so -- we expect that when we compare its performance in the Geekbench 3 single-core test with the 27-inch iMac we have enroute, that it may not be much faster at all. Part of the reason for this is that the current iMac incorporates Intel's 'Haswell' microarchitecture, which is a generation ahead of the 'Ivy Bridge' microarchitecture underpinning its latest Xeon processors. Where the Mac Pro really shines is in powering through multithreaded applications across its multiple processing cores (six in this instance), as you can see from its superb multi-core score of 20683.
Remember, Apple offers eight and twelve core models as well -- for applications in the science laboratories, film studios, and 3D design studios, in particular, the new Mac Pro will excel at getting through intensive and complex processor, parallel processing, and 3D graphics intensive computations with ease. For our full review, we will publish some additional test results that we have run, including disk speed tests as well as ripping and encoding tests. We will also throw Apple's Pro applications -- including
Final Cut Pro X,
Logic Pro X and
Aperture -- at the our Mac Pro as well, which have been optimized for it. Given the overall processing, RAM, storage, graphics, and I/O bandwidths of the new Mac Pro, we expect that it will handle just about anything we can throw at it. Look out for our full review of the Apple Mac Pro in the coming days to learn more about how it performs across a wide range of tasks.
By Sanjiv Sathiah