...
Let's not forget too that the existing Mac Pro, although a marvellous piece of engineering, was designed in the G5 era when chipsets would practically burn a hole through the ozone layer! These days they are much more efficient, as are no-doubt any new cooling processes Apple could use for their CPU's. I for one would welcome a flatbed design with a removal hood.
Actually not. mid-to-high range GPU PCI-e cards run hotter now than they did back in the G5 era. Also the CPU package options at the upper end of the core counts aren't significantly cooler than those old G5s. As long as the "core count war" continues there is no huge savings on power in Mac Pro class hardware. "Cooler" cores just allows more cores to be packed into the same package at the same power levels.
There is a very large difference in power savings at idle, but not under workload.
What has started to go down is mainstream desktop and mobile processors. That is partially because the 'core count' war has been capped. On Intel offerings the cap is 4. Additional transistors and "cores' are being thrown at graphics but the x86 count has been capped. That has lead to significant incremental drops in TDP over time for a count of 4.
There are some tweaks Apple could do to improve cooling but much of the Mac Pro's volume is still motivated by the parts in the workstation class that are likely to be used in a Mac Pro. As long as Apple holds the design objective that the Mac Pro is relatively quiet under load this amounts to what the bulk of air that can be moved (basic thermodynamics). They can either it faster in smaller container volumes or slower in the currently large sized (and unobstructed volume.).
The bigger legacy design assumptions that need to be examined are:
1. GPUs don't suck down more power than CPUs. If targeting high performance that role has shifted.
[ Apple could inhibit Mac Pro's competitiveness but capping GPU power but that is increasingly likely to kill its sales numbers. ]
2. 5.25" bays not used as much as a much higher need for 2.5" bays.
Trading at least one 5.25" bay for 4 2.5" bays would be a huge step foward.
3. CPUs aren't just single function packages anymore. Increasingly the I/O is being integrated into the CPU package. Two Xeon E5's doubles the amount of PCI-e lanes available. The Mac Pro should leverage that somehow in the two package configuration.
Not necessarily more PCI-e slots but "wider" ones. ( 4 x16 configurations instead of just 2 x16 and 2 x4. ). [ Some PCI-e switches and a wider daughtercard socket could still allow the basic board to be reused... Just not as wide in the single package config. )
4. There is no XServe so there is no need for gratuitous rack hostile handles for artificial product differentiation.
5. High speed wireless requires 3 (or more ) antennas.
All of those are much larger physical changes than where the CPU TDP levels have moved to.