Apple announced a bevy of new products, software, and hardware refreshes at WWDC 2017; CEO Tim Cook wasn’t kidding when he called it the largest WWDC the company has ever produced. For years, Apple’s approach to traditional computing hardware could charitably be called erratic. For some users — specifically, those who needed AIO devices with good overall specs, but less GPU horsepower — iMacs have been strong choices for years. If, on the other hand, you wanted a powerful GPU to match your CPU, Apple’s options have been much more limited. Its Mac Pro was the only device to offer a plausible graphics solution, and that platform’s update cycle has lagged far behind other hardware.
The newly announced iMac Pro, which we first covered yesterday, is a bid by Apple to change that standing, and an attempt to redefine the kind of performance consumers can expect from an AIO in general. The space-gray (that’s the only color) iMac Pro will start at $5,000 with a 27-inch 5K panel, up to 18 Xeon cores, and up to 22 TFLOPS of graphics compute delivered courtesy of AMD’s Vega GPU. That last figure refers to the GPU’s half-precision 8-bit performance; single-precision performance will target 11 TFLOPS. This implies either a cut-down Vega configuration or a lower clock rate, since the Vega Frontier Edition AMD announced in May will target 13 TFLOPS of single-precision performance.
If we had to bet between the two, we’d bet on a lower clock rate. Squeezing a high-end GPU into an AIO is no mean feat, and the peculiarities of silicon scaling means that a wider GPU with a slower clock rate is typically more power efficient than a narrower GPU with a higher clock rate, all else being equal. With a base price at $5,000, you can bet the configuration will not be the 18-core option with a Vega GPU, but we don’t know anything yet about how pricing will scale, or what exact GPUs and CPUs will be available at the lower price points.
While we don’t know which Xeon chips the new iMac Pro will use, Intel’s current lineup of Broadwell-based E5 Xeons should provide a reasonable comparison point. Right now, the lowest-TDP 18-core CPU that Intel offers is the E5-2695 v4. That’s an 18-core chip with a 2.1GHz base clock, 3.3GHz Turbo, and 120W TDP. We already know that Intel’s upcoming Skylake chips drastically reduce L3 cache in favor of adding more L2, so it’s possible that TDP segmentation will change as well. But a 120W chip in an all-in-one is already a tall order, before you add a high-powered GPU. Then again, AMD may be able to deliver a fairly power-sipping Vega — the Radeon Nano leveraged lower clocks and HBM, and still holds the record as the most power-efficient chip AMD has ever built.
It’s unfortunate AMD wasn’t able to lock down both halves of the iMac Pro design. But given how long product cycles are, Naples likely wasn’t ready when the iMac Pro was under evaluation. Apple is claiming that its new thermal design will improve cooling by up to 80 percent, but that’s the kind of promise we want to see independently evaluated. Increasing cooling capability is relatively easy; increasing it without simultaneously increasing noise and thickness is the tricky part.
The iMac Pro will also feature up to 4TB of SSDs, 128GB of ECC RAM, four Thunderbolt 3 ports, and 10GbE ethernet support. In many ways, it seems as if Apple wants to deliver the kind of workstation performance the Mac Pro has lacked as it languished, sans updates, for the past four years. One wonders what kind of system they’ll ultimately unveil as a replacement for the current Mac Pro, if the new iMac Pro packs this kind of firepower.
The only nitpick we have is the system’s availability. You won’t be able to buy a new iMac Pro until December — a date that’s going to raise questions about what, exactly, is keeping Apple from introducing new hardware more quickly. Is this a sign of continuing Vega problems? But if it is — and that’s far from proven — why would Apple have stuck with Vega in the first place, as opposed to using a different AMD solution or an Nvidia GPU? It’s entirely possible that other aspects of the system design are holding back its launch. We’ll wait to see what kind of availability we see from Vega before drawing further conclusions.