You must be right. Apple clearly didn't think of any of this stuff. They don't know how many PCI lanes a graphic card needs. As we know, they never sweat the details and always release crippled products. And of course, nobody will ever release an external PCI expansion chassis that connects via thunderbolt. The only thing to do is keep releasing the same old design.
Also, if you're buying a new Mac Pro, that's a substantial layout of cash. I think spending a bit extra on a thunderbolt-PCI expansion chassis for your old gear won't make it that much more unpalatable of a cost.
Yeah, you keep thinking that "extra bit" isn't going to make it more unpalatable. Only rich folks think that way. The rest of us for whom money actually
matters can't always afford to splurge on every last thing. $350-$400 for a BYOD (0TB) enclosure isn't "just a bit" when you're paying top dollar for the MP already.
Oh, and it doesn't matter how well Apple "thinks" things over with regard to the video cards - you can't just magically make more lanes appear on a single controller. TB2 takes nearly half of them, two more go for the SSD, leaving far less than the 32 lanes necessary to run both video cards at full speed (22 in fact if you factor in the wireless card). But as I've already mentioned, two cards in crossfire mode
can exceed a single GPU in certain tasks. And being that these are workstation cards, those tasks are failry limited, which could mean that these cards were optimized for just such a setup.
And on the subject of PCIe expansion chassis - you do realize that if you use one for a video card that the two video cards already inside the Mac Pro still eat up the same power and dissipate the same heat, right? They don't magically shut down. And not even a TB2 connection can run a video card at full speed. The best it can do for a single card is x2 unless it can synchronize two TB2 connections simultaneously (20 Gb/sec = 2.5 GB/sec).
First of all, did you forget that there are 6 of those 20GB/s Thunderbolts?
Making it 120? And they have to work the same time as there's a possibility for 3 4K screens via TB ports.
Then we go to graphics. 15.75 GB/s if x16, 7.88 GB/s if x8. It makes already 31.5 or 15.76 GB/s.
Then ok, the storage goes around 2.5 GB/s. We go around 160 GB/s.
But how does this stuff work? There are either different controller for the Thunderbolts that they have said nothing, or then it goes like they show and makes us understand it: the Graphics works with the x16 (or higher), and the Thunderbolts just goes with graphics flow. (as they work also in the world of screen connections.) Or then boths and something in the middle making them work great together.
(as you might find out, that the 40GB/s bandwidth for PCIe 3 was on the page of processor, that it means the processors/ it's controller's limit. Oh right, and if the top model comes with 2 processors, it doubles their PCIe controllers bandwidth. I see no problem.)
Each port can do 20 Gb/sec. As another poster below in this post pointed out, I did mistakenly read GB instead of Gb for the "per port", making the total data rate for all six ports 120 Gb/sec, or 15 GB/sec. I've corrected that below. That said, the 40 GB/sec PCIe bandwidth is the maximum period for that bus. Each device on the bus might be able to go above that
internally (I know the FirePro cards definitely can at 264GB/sec+ combined), but the bus itself is limited to 40 GB/sec throughput minus overhead.
Adding a second CPU does not double available PCIe bus bandwidth. It is shared between CPUs. It is only doubled if there is a second PCIe controller for the second CPU to have direct access (a couple PC motherboards offer this configuration)
Look i agree with almost all of your post but if there is one i have learned, its to trust the Apple engineers. I would bet a large sum of money that the cooling system on the new Mac Pro will work very well. The new Mac Pro was not built overnight, im sure the Apple engineers have tested it thoroughly. I know alot of people were sure the new iMac would run hot but it runs cooler then the previous generation!
I'll just remind you of the G4 Cube. Engineers are good. But they don't always hit home runs. And that small space with so many hot components crammed in there is bound to cause the CPUs (and perhaps even GPUs) to be thermally throttled at some point. Thankfully they aren't always running at full speed anyway due to SpeedStep, but when they are and it gets really hot in there, you can bet they'll be throttled down.
I'll give you one thing - not everybody's going to use all of their cores at max load, so for the more mundane of us, it might not matter much with regard to heat. For those that intend to push these machines, it may very well prove problematic. We'll just have to wait and see.
Do you two know your Gb's from your GB's?
Isn't Thunderbolt 20Gb/s?
Each port can do 20 Gb/sec. It does look though like I need to adjust my calculations slightly. At six 20 Gb/sec (120 Gb/sec total bi-directional) that comes to 15 GB/sec or 15 lanes. Thank you for making me look back at the ad again. That still only leaves 23 lanes max, and I had forgotten about the wireless card which is likely mini-PCIe too, but thankfully should only require one lane off the bus. So 22 lanes between two video cards. That still leaves them at x8 each.
