If you had to choose: Thunderbolt Vs Upgradeable GPU + PCIe slots?

[VirtualRain]

the apps are mainly dependent on ram,cpu,&gpu.. raid setups, hard drives, pci cards etc-- that stuff doesn't really matter for app performance (at least the applications i'm familiar with)..
i'm pretty sure the new mac will out perform the current one when it comes to running software

the battle or whatever would have to be less application based (even though that's the main importance imo) and more about storage/data moving workflow.. (though again, i don't really see how the new mac would lose this battle either.. it will lose when cost comes into the picture because it's going to cost additional money for the external enclosures needed with the new mac)

Agreed.

But a simple comparison would be simply to try and equip a 5,1 Mac Pro with the same gear as the new Mac Pro... Dual W9000 FirePro GPUs, USB 3, and a PCIe SSD that tops 1.2GB/s... (The sad truth is that this is impossible thanks to the slot arrangement in the 5,1. And affordable? I doubt it).

 

[Donar]

There is one way to end this on going debate.
Have the nMP go app for app with a 5.1. The nMP will be the "high end" model with all the TB enclosures and cards needed for the "battle". The 5.1 will be configured by Slughead including any Raid setup, PCIe cards or what is needed for the "battle".

A shootout between MacPro 5.1 with a 2009 grade chipset and a MacPro 6.1 seems a bit unfair to me. Funnily enough i think a 5.1 MacPro equipped with Dual X5690 Xeon (6 Core 3.4 GHz) could smoke the 6.1 CPU performance wise.

Edit: According to Geekbench the 6.1 wins, but the score is just 2.8% higher. After official release of the 6.1 we will see if numbers increase in favour to the new system.

Edit 2: I found some GB results that favour the old cheese crater equipped with dual X5690.

 

[slughead]

Check out this link: http://www.techpowerup.com/reviews/Intel/Ivy_Bridge_PCI-Express_Scaling/23.html

Average, it is about a 12% loss in performance for a 680 (The AMD card was around 7%).

Given their selection is valid, that's 12% off the high-end card from First quarter 2012. That's as much speed as you can squeeze out of PCIe 2.0 4x/TB2--forever. It is now nearing the 4th quarter of 2013 and the next generation of cards is significantly more powerful. Not only that, but as pointed out by other users, OpenCL can and will tax the card significantly more than just regular gaming--making the bottleneck much more apparent, possibly approaching 75-90%, if it indeed starts pushing PCIe 3.0 to the 8GBps-16GBps range.

Again, any solution that runs over TB2 will have to be faster than dual FirePro to be an actual "upgrade." Basically, you're saying nMP users will be happy trapped with 2012 GPU speeds in a computer they're buying in 2014. Again, the machine they buy will never be able to beat that benchmark.

What if the best card you could run in a Mac Pro you bought this month (and every year that you own the computer thereafter) was a GTX580? This is the dilemma nMP owners will face.

Also, the Mac Pro is coming with AMD workstation cards. While I'm sure the pros will want these cards more, some are going to want GTX or AMD gaming cards (tons of people add a GTX card to the current Mac Pro). In this case, I think Thunderbolt 2 won't pose as a huuge bottleneck for adding a gaming card. Is it future proof? Probably not. But it will still give gamers with the new Mac Pro enough horsepower

My gaming PC with a GTX 780 is already significantly faster via PCIe16x 3.0 than anything the new Mac Pro could ever run over TB2. I am also skeptical that the nMP will even have Crossfire, but that remains to be seen. If that's the case, GPU power for gaming especially is going to be pretty anemic.

 

[flat five]

Again, any solution that runs over TB2 will have to be faster than dual FirePro to be an actual "upgrade." Basically, you're saying nMP users will be happy trapped with 2012 GPU speeds in a computer they're buying in 2014.

i'm not tech savvy enough to know what's actually possible down the road but i don't think bandwidth is important at all when imagining plug-in gpu modules for openCL based apps..

just using still image renders for example, a gpu can run full bore and only generate a few_hundred_kb of useable data in a minute.. transferring that data to the master could be done at dial-up speeds.. (but yes, there's a bit more communicating which needs to happen so i don't think using dial up is too wise..2gigabyte/sec will definitely be plenty though)

again, i really don't know if that's strictly fantasy world stuff.. i also don't know if future displays will have onboard graphics but that too is an idea that's out there..

i guess my question is why is it even necessary to worry about thunderbolt bandwidth/gpus? you'll have 2 in the mac pro already.. what situation do you foresee plugging in a third+ gpu and bumping into bandwidth problems?

 

[slughead]

A shootout between MacPro 5.1 with a 2009 grade chipset and a MacPro 6.1 seems a bit unfair to me. Funnily enough i think a 5.1 MacPro equipped with Dual X5690 Xeon (6 Core 3.4 GHz) could smoke the 6.1 CPU performance wise.

Edit: According to Geekbench the 6.1 wins, but the score is just 2.8% higher. After official release of the 6.1 we will see if numbers increase in favour to the new system.

Quiet you! We must compare the 2014 Mac Pro to the 2009 featureset, otherwise the nMP fans will be super bummed when it falls short.

i'm not tech savvy enough to know what's actually possible down the road but i don't think bandwidth is important at all when imagining plug-in gpu modules for openCL based apps..

You might want to talk to your friend deconstruct60, he seems to think OpenCL will push the bandwidth requirements past 8GBps, even on a FirePro W9000.

https://forums.macrumors.com/posts/17462575/

 

[slughead]

---- nvm

 

[EpicBlob]

Given their selection is valid, that's 12% off the high-end card from First quarter 2012. That's as much speed as you can squeeze out of PCIe 2.0 4x/TB2--forever. It is now nearing the 4th quarter of 2013 and the next generation of cards is significantly more powerful. Not only that, but as pointed out by other users, OpenCL can and will tax the card significantly more than just regular gaming--making the bottleneck much more apparent, possibly approaching 75-90%, if it indeed starts pushing PCIe 3.0 to the 8GBps-16GBps range.

Again, any solution that runs over TB2 will have to be faster than dual FirePro to be an actual "upgrade." Basically, you're saying nMP users will be happy trapped with 2012 GPU speeds in a computer they're buying in 2014. Again, the machine they buy will never be able to beat that benchmark.

What if the best card you could run in a Mac Pro you bought this month (and every year that you own the computer thereafter) was a GTX580? This is the dilemma nMP owners will face.



My gaming PC with a GTX 780 is already significantly faster via PCIe16x 3.0 than anything the new Mac Pro could ever run over TB2. I am also skeptical that the nMP will even have Crossfire, but that remains to be seen. If that's the case, GPU power for gaming especially is going to be pretty anemic.

Check out the e-gpu forums. Even on pcie 2.0 x1, a higher end card will still perform better than a card that is already bottlenecked. My 660 performs with about a 20% loss, but if I upgrade to a 770 I will still see a boost in fps. The difference won't be as big, but it isn't like once a card is bottlenecked, it is impossible for it to pass it (not really sure how it works but that's how it is).

There's then of course the difference between a gaming card and a workstation card. Same specs won't matter at all; a gaming card will win at gaming and a workstation card will win at video editing, rendering, etc. And if the two cards don't come in crossfire, adding in a gaming card to the nMP via TB2 will have an even bigger boost for gaming.

As yes, your 780 in a desktop will ALWAYS beat out something that isn't at full speed. My 2012 Mac Mini and 660 probably won't be able to beat out a current Quad Core Mac Pro, but can it beat my system in price? Size? Power usage? These things will always come at a higher price. It is always more cost effective to build your own decked out computer, but people now a days are making size a much bigger factor. More and more consumers are willing to sacrifice a bit of power for better portability and a smaller form factor.

 

[macuser453787]

Oh yay! A poll! I'm sure as Apple sees this they're going to realize how wrong they were, apologize, and redesign the entire thing.

LOL that is funny!!! I keep thinking about how Apple will respond to market interest in nMP or lack thereof. Either way it goes, that should certainly get their attention.

 

[macuser453787]

So just in case I got my specs wrong and TB2 is 20Gb/s and not 40Gb/s

Looks like TB1 is 10Gbps = 1.25 GB/s

and TB2 is 20 Gbps = 2.5 GB/s

Anybody correct me on this if necessary but I just looked it up as Tesselator's info sounded a little off the mark.

Based on what math? PCIex16 3.0 = 15GB/s and PCIex16 4.0 = 30GB/s and Thunderbolt 2 = 20GB/s

Ergo, clemahieu's info is also off the mark. Not 20 GB/s, but 20 Gbps (which apparently comes out to about 1.6 GB/s, which is odd because usually Gbps/8 = GB/s, right?). Anyhoo, AFAIK Wonderspark's calculation in his previous post is correct.

PCIe is definitely faster than TB2 (that is, depending on lane width and revision). Specifically, PCIe 2.0 4x is in about the same speed neighborhood as TB2, but PCIe 2.0 16x is 16 GB/s, or about 6.5 to 10 times faster than TB2, depending on whether TB2 is 2.5 GB/s or 1.6 GB/s. Reminder: Current MP PCIe is 2.0.

A new MP with PCIe 3.0 would have been even faster: PCIe 3.0 16x = 32 GB/s or about 13 to 20 times faster than TB2, and I assume that PCIe 3.0 4x would be along the lines of 1.25 to 2 times as fast.

 

[slughead]

Anybody correct me on this if necessary but I just looked it up as Tesselator's info sounded a little off the mark.



Ergo, clemahieu's info is also off the mark. Not 20 GB/s, but 20 Gbps (which apparently comes out to about 1.6 GB/s, which is odd because usually Gbps/8 = GB/s, right?). Anyhoo, AFAIK Wonderspark's calculation in his previous post is correct.

Correct on all counts. Yes, these two people got it way wrong (not surprising people who are pro-thunderbolt have bad information). By the way, real world benchmarks will likely put TB2 at around 2.1GBps maximum.

In another thread, Tesselator's "36 GTX680 plugged into a single mac pro for ultimate CUDA!" is torn apart when it was found that CUDA/OpenGL can sometimes use even more bandwidth than gaming, requiring the full bandwidth of PCIe 3.0.

The nMP is going to be obsolete before it comes out. TB2 is going to be a really cool feature for high-end laptops... but PCIe is up to 8 times faster. A single 8x PCIe 3.0 slot has more bandwidth than all the Thunderbolt 2 ports in the nMP combined. 16x has twice that.

There are quite a few workstations out there that support 80 lanes of PCIe 3.0 running at full bandwidth--that's the equivalent bandwidth of Forty Thunderbolt 2 ports, or 13 new Mac Pros worth of Thunderbolt channels.

 

[AidenShaw]

...

There are quite a few workstations out there that support 80 lanes of PCIe 3.0 running at full bandwidth--that's the equivalent bandwidth of Forty Thunderbolt 2 ports, or 13 new Mac Pros worth of Thunderbolt channels.

You're expecting the fans to do arithmetic?

If you compare the vaporware Mini Mac Pro to the vaporware Dell or HP dual socket announcements - it's clear that the Mini Mac Pro isn't even in the same league. The Mini Mac Pro will support orders of magnitude less memory, orders of magnitude less expansion, orders of magnitude less I/O bandwidth.

But, it really is a much faster Mini Mac!

Apple has eliminated the "Mac Pro", and updated the "Mac Mini".

 

[liquid stereo]

Could not agree more

Yes.

Yes.

Yes.

Yeah, my political science instructors would have something to say about the "neutrality" of the poll.

But tilted or not, the poll is a no-brainer to me, because I'm one of those for whom the current tower configuration is just plain stupid and useless - a huge waste of space, aluminum, and plastic. I don't need or want third-party graphics cards, and I'd much rather have my backup and mass storage drives outside the chassis - that's what networks are for. I'd rather have the CPU up on my desk, so the fan doesn't suck up carpet lint and dust bunnies (that I have to vacuum out before swapping drives, upgrading RAM, etc), and so I don't have to crawl around on my belly under a desk to plug cables into the ports.

While some use terms like Maxi-Mini derisively, this is the first Pro that's made me drool. (But then, I loved the Cube, too.) The form factor is right for my needs, the likely performance (thanks mostly to all that solid-state storage) is very attractive. OK, like a Ferrari, Maserati, or Porsche, it's far more horsepower than I really need (Aperture and InDesign). iMac or Mini with Fusion Drive would be the "sensible" solution. On the other hand, the thing will be so fast that I'll probably get two more productive years out of the thing before it becomes painfully slow when running the latest software.

If this hadn't come along, I'd have continued dreaming of a new 27" iMac. So, Apple has my number, and I suspect the number of many others whose job description and salary justifies a Pro, "if only it had a small footprint." Regardless of the poll results, I think Apple's going to find two new Pro users for every one they lose.