The first cylindrical Mac Pro was announced in June 2013 and released near the end of that year. With the system starting to show its age, there is a good chance we'll see a refresh announced sometime this year.
On the CPU side, the refresh path is fairly straightforward: just replace Ivy Bridge-E with Haswell-E. The GPU side is less clear. While Maxwell would seem to be the obvious selection, this presents some issues; Nvidia has generally been less willing than AMD to create semi-custom solutions of the sort that Apple needs, and equally important, Maxwell simply can't provide reasonable performance in Double Precision computing. The latter is not important for most gamers, but the Mac Pro is a GPGPU-focused system designed for artistic, technical, and scientific tasks. To put this in perspective, GM200 Titan X has lower DP GFLOPS (48) than the lowly Pitcairn (160). Sure, Apple could downplay this in their marketing, and people who care mostly about photo/video editing won't mind. But they will lose at least some sales. Focusing on CUDA isn't going to happen; Apple doesn't want to promote third-party proprietary solutions (their support for OpenCL is not an aberration; look at their position on Flash).
It's possible they could use Kepler; a GK110 (tweaked to fit in the Mac Pro's thermal envelope) with full 1/3 DP performance would beat the current flagship D700 Tahiti in almost everything. But is Apple really going to want cards that are obsolete the moment they're produced, especially since Nvidia guards their markups jealously? That's assuming Nvidia would even produce a custom SKU, which they might not.
I think the strongest possibility is that the next Mac Pro continues to use AMD GPUs. But the GCN 1.0 products in the current generation (FirePro D300, D500, and D700 - Pitcairn, Tahiti, and Tahiti respectively) are no longer good enough for a high-end workstation. In fact, none of AMD's current lineup really fits - with one exception. We have to look to what AMD will be releasing in June. In fact, I wouldn't be surprised if Apple's involvement is the primary reason for AMD's extreme secrecy this time out. Apple is very insistent on zero leaks, and they demand that their suppliers maintain the same policy. If Apple is planning to announce the 2015 Mac Pro in June - two years after the 2013 Mac Pro announcement - then part of the deal might have been that AMD remains tight-lipped until then.
So, what GPUs would be suitable for a 2015 Mac Pro? On the low end, the obvious choice would be to replace Pitcairn with a fully-enabled Tonga. In fact, the Pitcairn->Tonga upgrade was already done once, when AMD replaced the discrete FirePro W7000 with the FirePro W7100 at the same TDP (though the latter was a cut-down Tonga, not full). So the entry-level Mac Pro GPU (let's call it the "FirePro E300") will be Tonga with 2048 shaders, a 256-bit bus, and 4GB of GDDR5 VRAM. This would easily beat the FirePro D300 (Pitcairn) in everything.
On the high end? Well, Hawaii can be ruled out; it runs too hot for the Mac Pro's shared heatsink (even Tahiti stressed that quite a bit when both GPUs were run full blast), and the power requirements are way too much as well. So the next obvious question is whether the upcoming Fiji might fit the bill. Even if the full chip would have too high a power envelope, reducing the shader count and the GPU core clock might be enough to get it to work. I'm going to put my speculation out there: the premium Mac Pro GPU (let's call it the "FirePro E700") will be Fiji with 3584 shaders and 8GB of HBM. (I don't believe the 4GB limit rumors; besides, the Tahiti FirePro D700 had 6GB, so 4GB would be a regression.)
What do you think? Does this sound plausible, or will Apple do something different?
On the CPU side, the refresh path is fairly straightforward: just replace Ivy Bridge-E with Haswell-E. The GPU side is less clear. While Maxwell would seem to be the obvious selection, this presents some issues; Nvidia has generally been less willing than AMD to create semi-custom solutions of the sort that Apple needs, and equally important, Maxwell simply can't provide reasonable performance in Double Precision computing. The latter is not important for most gamers, but the Mac Pro is a GPGPU-focused system designed for artistic, technical, and scientific tasks. To put this in perspective, GM200 Titan X has lower DP GFLOPS (48) than the lowly Pitcairn (160). Sure, Apple could downplay this in their marketing, and people who care mostly about photo/video editing won't mind. But they will lose at least some sales. Focusing on CUDA isn't going to happen; Apple doesn't want to promote third-party proprietary solutions (their support for OpenCL is not an aberration; look at their position on Flash).
It's possible they could use Kepler; a GK110 (tweaked to fit in the Mac Pro's thermal envelope) with full 1/3 DP performance would beat the current flagship D700 Tahiti in almost everything. But is Apple really going to want cards that are obsolete the moment they're produced, especially since Nvidia guards their markups jealously? That's assuming Nvidia would even produce a custom SKU, which they might not.
I think the strongest possibility is that the next Mac Pro continues to use AMD GPUs. But the GCN 1.0 products in the current generation (FirePro D300, D500, and D700 - Pitcairn, Tahiti, and Tahiti respectively) are no longer good enough for a high-end workstation. In fact, none of AMD's current lineup really fits - with one exception. We have to look to what AMD will be releasing in June. In fact, I wouldn't be surprised if Apple's involvement is the primary reason for AMD's extreme secrecy this time out. Apple is very insistent on zero leaks, and they demand that their suppliers maintain the same policy. If Apple is planning to announce the 2015 Mac Pro in June - two years after the 2013 Mac Pro announcement - then part of the deal might have been that AMD remains tight-lipped until then.
So, what GPUs would be suitable for a 2015 Mac Pro? On the low end, the obvious choice would be to replace Pitcairn with a fully-enabled Tonga. In fact, the Pitcairn->Tonga upgrade was already done once, when AMD replaced the discrete FirePro W7000 with the FirePro W7100 at the same TDP (though the latter was a cut-down Tonga, not full). So the entry-level Mac Pro GPU (let's call it the "FirePro E300") will be Tonga with 2048 shaders, a 256-bit bus, and 4GB of GDDR5 VRAM. This would easily beat the FirePro D300 (Pitcairn) in everything.
On the high end? Well, Hawaii can be ruled out; it runs too hot for the Mac Pro's shared heatsink (even Tahiti stressed that quite a bit when both GPUs were run full blast), and the power requirements are way too much as well. So the next obvious question is whether the upcoming Fiji might fit the bill. Even if the full chip would have too high a power envelope, reducing the shader count and the GPU core clock might be enough to get it to work. I'm going to put my speculation out there: the premium Mac Pro GPU (let's call it the "FirePro E700") will be Fiji with 3584 shaders and 8GB of HBM. (I don't believe the 4GB limit rumors; besides, the Tahiti FirePro D700 had 6GB, so 4GB would be a regression.)
What do you think? Does this sound plausible, or will Apple do something different?
Facebook
Twitter