[corresation]

The one strange assumption that most Mac Pro reviews start with is a baseline of two compute cards -- that the Mac Pro is competitive when compared with other machines with two high-end compute cards.

But I don't want two high-end compute cards, and I suspect that many who are trying to convince themselves that they'll benefit from it will gain no value from it.

For many, many workloads, modern compute still represents an iffy proposition (at the price levels being talked about, the Xeon Phi would almost certainly represent a better proposition). With unified memory things might get more workable, but as is it remains a relatively fringe benefit, and it seems odd that the entire value proposition of the machine relies upon it.

[rsynnott]

> But I don't want two high-end compute cards

In which case, do you want a Xeon workstation of any sort? As mentioned later on in the review, you make significant sacrifices for Xeon (startlingly expensive, last-gen cores), and, besides the option for more cores than you can get on an iX, the main thing you get is extra PCIe lanes, which are not actually that useful for most things; one of the few things they _are_ useful for is dual hefty GPUs.

[corresation]

do you want a Xeon workstation of any sort?

ECC memory is a big one. Usually getting a Xeon workstation comes with SMP, though not on the new Mac Pro. Big memory support. Lots of PCI lanes. Usually lots of space to drop in extra storage, a couple of 10GbE ports (the Mac Pro has just 1Gbps ports which is another oddity).

There are a lot of traditional reasons a so-called workstation features a Xeon.

Worth noting that there is a couple of Haswell (therefore AVX 2.0 supporting) Xeons -- the E3 v3s. Unfortunately they're the baby ones so they have ridiculous low max memory, no SMP, and max out at 4 cores. Hopefully the E5 v3s are out soon.

I honestly don't get how the Mac Pro hasn't gotten more mainstream criticism. It solves a problem that I don't remember anyone ever having (honestly a garbage can form factor seems like more of a nuisance than the flexible cubes we're all used to), while bringing a ton of problems to the table, and being a massive sunk cost for fixed hardware that is going to be outdated very, very quickly.

[justincormack]

Yeah 10GbE missing is odd, although its probably a cost thing, and you appear to be able to buy thunderbolt 10GbE.

[abtinf]

For the Mac Pro target market, 2xGPU combined with massive memory bandwidth is a blessing. Final Cut, Photoshop, and other creative software are GPU accelerated.

The Xeon Phi is significantly more expensive, draws significantly more power, and is significantly less useful for creative software loads.

[corresation]

For the Mac Pro target market

Which is what? It's just anecdotal but most of the people I know with Pros got it as a high end development workstation, building iOS apps, etc. Final Cut Pro is pretty much the only app that benefits from the dual GPUs, and even then the gain is relatively marginal over a machine four years old. And that's paying for very high priced "workstation" GPUs (I called them compute cards because that is what they are geared for, though as with all compute cards they are derivatives of GPUs. They're really price ineffective as GPUs), when you can get almost all of the same advantages on a basic ATI card for a couple hundred dollars.

Virtually every review of the Pro seems to be giving it a very soft glove approach for some reason. It is an enormously expensive monument to the dual compute GPU, for marginal gains in most apps.

[interpol_p]

Buying a Mac Pro to "build iOS apps" or even to do heavy Photoshop work is a waste of money. The people that bought them for those reasons did not make an educated decision. A 15" rMBP or iMac is perfectly fast for those tasks.

The Mac Pro is a reasonably priced dual compute GPU workstation. But it's enormously expensive when compared to what most people actually need.

I'd say that if you are doing 4k Final Cut Pro work or writing your own OpenCL software, this machine is for you. In the future it may be suitable for people using GPU-based 3D renderers (I don't think there's any great ones on OpenCL at the moment).

Aside from the above niche target market, the Mac Pro target market doesn't exist yet. Apple seems to be using this machine to push the development of OpenCL and to push the development of "Pro" software for Mac OS X in the direction they want to see it go.

Personally, I think it's a very exciting direction to take pro software. I hope many developers jump on board. Once that happens I think we'll start to see a larger target market for the machines.

[xenophanes]

so you think 2 compute cards is bad. but what about the mac pro approach 2 gfx cards? good or bad? not sure what position you're taking and curious.

[corresation]

Are you debating nomenclature? One of the GPUs is effectively dedicated to compute, while the other is architecturally chosen to optimize compute. These cards are geared to compute and not graphics.

[xenophanes]

Your post was unclear and I asked what you meant. I didn't debate anything. i was extremely clear and explicit about this. i asked a question, said i didn't know, said i was curious, didn't dispute anything you said.

Only a domain expert could be expected to automatically know whether double GPUs is effectively the same thing as double CPUs or not. Apple has put a lot of effort into making their GPU structure effective (in general, not just for a few special cases), and I didn't know if you were saying they'd failed or not. i know parallel processing is hard, but i also know apple has smart people who've worked on it. i don't know things like whether "lack of unified memory" is a problem for apple's design too, or not. it seems completely plausible to me, not knowing the domain that well, that apple could have had something that works well in general, or not – i don't really know and you didn't say, just assumed your reader would somehow know what you meant (which will basically only work for people who already know your point and have no need to read your comment at all).

[icegreentea]

His entire point boils down to "GPU Compute (of all forms) is overweighted in comparisons between the Mac Pro and other workstations".

He got antsy because from his perspective, your question definitely seems kinda out of the air. In his original comment, he's basically implying that any GPU based compute solution (so these can be plain gaming optimized GPUs, the "professional" GPUs, or compute optimized GPUs) aren't worth is for the majority of use cases.

The "second card" he talked about was the Xeon Phi, and he makes the differentiation between it and other "GPU based" compute solutions since Xeon Phi consists of a "large" (sub 100) number of relatively simple, but full blown CPU cores (for example, the current Xeon Phi is based on an old Pentium core, the next gen is supposed to be basically an Atom core). This should, in theory, make it easier to exploit parallelism.