...
His biggest argument for the dual CPUs is that he believes GPU rendering's time is up.
That is a bit of an overstatement. A GPGPUs goning to exclusively rule the future world of computation? No. Are CPUs? Also no.
He flips flops a bit in the article. For some folks it is "go get a cluster" or for the Pixar folks that they aren't going to up up their cluster. Frankly if 2 CPUs are 'better" why aren't 4 or 8 or 16?
His argument is more so "cluster license" is too expensive and wants the "cluster' in one box.
That's debatable depending who you ask. But I really don't see Apple making a dual CPU version of this machine in future iterations.
I don't either but far more so for two reasons.
A. Not very enough folks were buying the dual CPU models to make a completely different case and separate engineering track worthwhile.
If that had been the bulk of the Mac Pro market then they would have dropped single CPU and kept the dual CPU design constraint. They didn't.
Peeled off from the single CPU models the duals probably don't meet a minimal unit number threshold.
[ Frankly the whole Apple needs a "halo" workstation is also a bit delusional. Apple is out to sell products that move.... not showboat, posturing partial demo products. ]
B. Intel is creating "dual CPU" as single CPU. The "CPU" is a package of cores. What the real root cause issue is what is minimum number of cores need to be useful/production.
This iteration tops out at 12. The next iteration will likely be 14. The one after that is likely close to 16. The response of "I need as many as I can get" again begs the question of why stopping at 2 packages.
This Mac Pro doesn't have to suck in everybody... just enough this round to get to the next iteration.
He also goes on to mention that Apple would have been better off getting the same performance or better by offering a dual 6 core or dual 8 core model, referencing the low cost ($583) of the 6 core chip.
Apple would have been better off for him as a customer if they are added the 10-core E5 2680 v2 option that is the same price as the E5 1680 v2 (8 core) model. (basically the same product with the latter having two cores flipped off and the clock rates boosted a bit as compensation. )
Just one more BTO option. Possibly on just the upper standard configuration if put the minimal threshold on GPU card at D500.
A 10 core option would have beat the old 12 core Westmere on the CPU metrics. He would have had a somewhat faster machine for the "same old price as the 12 core".
Apple is targeting folks with more mixed usage software. They could have trivially picked up a somewhat wider group with just one more CPU option. However, as the article points attempt to point out that folks are a bit price insensitive. So Apple is probably betting to get a decent fraction to go for the 12 core.
He's right on about the driver issues though. That needs to be fixed.
there are driver and application issues. The Cineabench 11 versus 14 scores mentioned. That is an application issue. Similar on CPU renders that AVX optimized. The hardware is there waiting... it is just not being used.
Apple is a behind the curve on OpenGL (and OpenCL ), but that is hardly new. The previous reviews grumbled about the exact same issue.
He's lost me permanently.
