The New Mac Pro - it's not for me

[johnhackworth]

Hi,

Since I came back to the Mac around ten years ago I've always owned one of Apple's 'pro' systems. From a G4 PowerMac, a G5 PowerMac through to my current Mac Pro. However, I'm unlikely to move from my current four-year old Mac Pro to the new one - which is disappointing since there's nothing else in Apple's current range in Europe which would provide a significant increase in performance.

My issue with the next-gen Mac Pro is that it's target audience seems to be a subset of 'pro' users - namely those who demand stellar GPU performance. I realise that I may be a niche user however I've very successfully used my current Mac Pro for Windows development. For example, it's allowed me to virtualise an entire network of Windows servers in a single machine. I care about CPU and storage performance. I don't really care about GPU performance. The next-gen Mac Pro's bill-of-materials seems to include two high-end GPUs which are likely to contribute in large part towards its selling price. This is unfortunate and I don't really want to spend a large chunk of money on hardware which I'd never use.

Apple's conception of a 'workstation' seems to be focussed on graphics or video professionals. It seems to have abandoned those who care mostly about CPU performance and expandability.

Just my tuppence.

Julian.

 

[iSee]

You're jumping the gun a bit here.
Wait for it to come out, find out what it costs.
Get more info on what it can do.

 

[theSeb]

You're jumping the gun a bit here.
Wait for it to come out, find out what it costs.
Get more info on what it can do.

Wow. A sensible post.... I haven't seen many of these around here recently. No alarmist assumptions? No insults to those that disagree? No making up your own criteria to prove how you are right? Outrageous.

 

[Lostanddamned]

Hi,

Since I came back to the Mac around ten years ago I've always owned one of Apple's 'pro' systems. From a G4 PowerMac, a G5 PowerMac through to my current Mac Pro. However, I'm unlikely to move from my current four-year old Mac Pro to the new one - which is disappointing since there's nothing else in Apple's current range in Europe which would provide a significant increase in performance.

My issue with the next-gen Mac Pro is that it's target audience seems to be a subset of 'pro' users - namely those who demand stellar GPU performance. I realise that I may be a niche user however I've very successfully used my current Mac Pro for Windows development. For example, it's allowed me to virtualise an entire network of Windows servers in a single machine. I care about CPU and storage performance. I don't really care about GPU performance. The next-gen Mac Pro's bill-of-materials seems to include two high-end GPUs which are likely to contribute in large part towards its selling price. This is unfortunate and I don't really want to spend a large chunk of money on hardware which I'd never use.

Apple's conception of a 'workstation' seems to be focussed on graphics or video professionals. It seems to have abandoned those who care mostly about CPU performance and expandability.

Just my tuppence.

Julian.

Care about storage performance and expandability, but don't think that the six Thunderbolt 2 (at up to 20Gb/s) and four USB 3 ports are enough? That is up to 36 Daisychained Thunderbolt devices and then there is the ability to use hubs with USB devices. How much are you looking to expand by?

There haven't been any numbers assigned to the actual power of the processor yet, beyond 12 cores and the fact it will be using a Intel Xeon E5, so it will be very difficult to say what the processor performance is as yet, beyond being capable of up to twice the floating point operations of the current generation.

You are counting your eggs before they've been laid, never mind hatched.

 

[Pakaku]

To me, it's not even worth it in graphics unless they can offer an NVIDIA option... I've had better experiences with them, and CUDA would be a shame to lose.

Otherwise I don't see the appeal in updating to a new machine quite yet.

 

[Moonjumper]

The next-gen Mac Pro's bill-of-materials seems to include two high-end GPUs which are likely to contribute in large part towards its selling price. This is unfortunate and I don't really want to spend a large chunk of money on hardware which I'd never use.

There was a lot of "up to" references in the specs revealed for the new Mac Pro. There will probably be cheaper GPU options.

And there was also mention of GPU power being used for general processing, so it won't be wasted.

Currently there are too many unknowns for most (but not all) people to make a definitive decision.

 

[pertusis1]

There was a lot of "up to" references in the specs revealed for the new Mac Pro. There will probably be cheaper GPU options.

And there was also mention of GPU power being used for general processing, so it won't be wasted.

Using the gpu for general processing is interesting. Where did you see that idea?

 

[GermanyChris]

You're jumping the gun a bit here.
Wait for it to come out, find out what it costs.
Get more info on what it can do.

It wouldn't matter if it's $.50 if it doesn't meet the requirement real or perceived

 

[fastlanephil]

Using the gpu for general processing is interesting. Where did you see that idea?

Phil Schiller at 2013 WWDC. at 5:08

http://www.youtube.com/watch?v=xzwWisemVCQ

OpenCL

http://en.wikipedia.org/wiki/OpenCL

 

[tamvly]

Jumping the gun, perhaps. With lots of virtual machines I would think that lots of individual devices (dedicated volumes for VMs) and lots of memory would be keys. I wonder if you'll care about GPU performance if the price is right and the memory limits permit.

I also wonder why you'd use a MP if you could run a bunch of VMs on a Linux box far cheaper. Given networking who cares what OS they run on?

 

[iSee]

Wow. A sensible post.... I haven't seen many of these around here recently. No alarmist assumptions? No insults to those that disagree? No making up your own criteria to prove how you are right? Outrageous.

oops, I forgot this was macrumors... hope I don't get banned... won't happen again, I promise.

Using the gpu for general processing is interesting. Where did you see that idea?

It's an emerging trend. As GPUs evolved to display 3-D scenes of ever higher quality and resolution, the GPU makers started making them with more and more processing cores because graphics processing tends to be something that can be done in parallel quite well. Over time, to meet the demand for more performance and more flexibility they added more cores and made them more sophisticated. Eventually they made the cores programmable. Each core is much simpler and slower than a typical CPU core, but there can be a lot more of them... hundreds or even thousands.
Even before they became programmable, clever programmers were figuring out how to formulate certain computationally intense problems in a way that could be fed to a GPU to process in a massively parallel way. Now that these GPU cores can be programmed they can be used more easily for a larger class of problems.

The real trick to harnessing this power is to have a processing problem that can be solved in a highly parallel way. It is not necessarily easy. or feasible or maybe even possible for some kinds of problems. So this power often isn't easy to use. From a software engineer perspective, you may need to rethink the entire way you approach a problem to take advantage of it.

Software developers are really just getting started with this, so it's not clear how far this approach will go. By making dual GPUs standard in the new Mac Pro while not providing an option for dual CPUs, it's pretty clear that Apple thinks it is the way of the future. We'll see.

(Personally, I think this is the right approach. To saturate the computing power of a single intel CPU, you have to max out 12 cores. If your problem scales to 12 parallel threads, you might except that it could scale to a lot more... enter the GPU. But I'm a newb to GPGPU, so don't put a lot of weight in my optinion.)

Currently there seem to be two APIs for writing software for GPGPU with some traction: CUDA and OpenCL. CUDA has been around longer and seems to have a fair amount of software. However, it is a proprietary thing from Nvidia and only works with Nvidia cards. OpenCL is open and works with GPUs from multiple vendors and, quite interestingly, also works with CPUs. It seems to be less mature than CUDA, seems less well supported, but also seems to be catching up fast. But I don't see how CUDA can survive, long term. Being stuck on Nvidia cards seems like a dead end, no matter what the early lead is.

It wouldn't matter if it's $.50 if it doesn't meet the requirement real or perceived

Right.

 

[ActionableMango]

My issue with the next-gen Mac Pro is that it's target audience seems to be a subset of 'pro' users - namely those who demand stellar GPU performance.

Actually, many of those who demand stellar GPU performance are also unhappy with the new MP due to limited selection (AMD only) and inability to upgrade using PCIe cards.

 

[Sean Dempsey]

Hey everyone.

I just want to say that I don't think Dr. Pepper is for me. So I made a post to say that.

Sean.

 

[phoenixsan]

Would not....

for me too.....but based more in cost factor than another considerations. Is not only the Mac Pro itself, add the Thunderbolt tech I would need...Hard to justify expenses over $3500 even for productive setup....

No internal expansion bothers me a little, too....More GPU power will be nice. But I would prefer various chip configurations, if possible....

 

[dmax35]

Hey everyone.

I just want to say that I don't think Dr. Pepper is for me. So I made a post to say that.

Sean.

LMAO! Great reply. Love it!

 

[MacVidCards]

Hey everyone.

I just want to say that I don't think Dr. Pepper is for me. So I made a post to say that.

Sean.

Try the New Kool-Aid. It's sweeter yet less filling.

 

[theSeb]

Try the New Kool-Aid. It's sweeter yet less filling.

I am getting quite tired of reading this from you. Are you not tired of it yourself? It's a bit old now.

 

[MacVidCards]

I am getting quite tired of reading this from you. Are you not tired of it yourself? It's a bit old now.

Then by all means put me on your "ignore" list, as I have done for several argumentative types here.

 

[theSeb]

Then by all means put me on your "ignore" list, as I have done for several argumentative types here.

I don't want to do that. You post many useful and interesting things. But your posts since the Mac Pro was announced have been full of insults and repeats of the same tired, old material. Just because someone disagrees with you does not mean that they are kool-aid drinking paid Apple shills. Maybe it's time to sing a new song?

 

[GermanyChris]

oops, I forgot this was macrumors... hope I don't get banned... won't happen again, I promise.



It's an emerging trend. As GPUs evolved to display 3-D scenes of ever higher quality and resolution, the GPU makers started making them with more and more processing cores because graphics processing tends to be something that can be done in parallel quite well. Over time, to meet the demand for more performance and more flexibility they added more cores and made them more sophisticated. Eventually they made the cores programmable. Each core is much simpler and slower than a typical CPU core, but there can be a lot more of them... hundreds or even thousands.
Even before they became programmable, clever programmers were figuring out how to formulate certain computationally intense problems in a way that could be fed to a GPU to process in a massively parallel way. Now that these GPU cores can be programmed they can be used more easily for a larger class of problems.

The real trick to harnessing this power is to have a processing problem that can be solved in a highly parallel way. It is not necessarily easy. or feasible or maybe even possible for some kinds of problems. So this power often isn't easy to use. From a software engineer perspective, you may need to rethink the entire way you approach a problem to take advantage of it.

Software developers are really just getting started with this, so it's not clear how far this approach will go. By making dual GPUs standard in the new Mac Pro while not providing an option for dual CPUs, it's pretty clear that Apple thinks it is the way of the future. We'll see.

(Personally, I think this is the right approach. To saturate the computing power of a single intel CPU, you have to max out 12 cores. If your problem scales to 12 parallel threads, you might except that it could scale to a lot more... enter the GPU. But I'm a newb to GPGPU, so don't put a lot of weight in my optinion.)

Currently there seem to be two APIs for writing software for GPGPU with some traction: CUDA and OpenCL. CUDA has been around longer and seems to have a fair amount of software. However, it is a proprietary thing from Nvidia and only works with Nvidia cards. OpenCL is open and works with GPUs from multiple vendors and, quite interestingly, also works with CPUs. It seems to be less mature than CUDA, seems less well supported, but also seems to be catching up fast. But I don't see how CUDA can survive, long term. Being stuck on Nvidia cards seems like a dead end, no matter what the early lead is.



Right.

Do you often buy things that don't fill a need either real or perceived?

 

[MacVidCards]

I don't want to do that. You post many useful and interesting things. But your posts since the Mac Pro was announced have been full of insults and repeats of the same tired, old material. Just because someone disagrees with you does not mean that they are kool-aid drinking paid Apple shills. Maybe it's time to sing a new song?

I am genuinely sorry if you are offended.

But I really do believe that Apple or a PR company of theirs has unleashed a squad of disingenuous folks working on bullet points.

The arguments they pose make no sense and yet they repeat them endlessly.

This particular Kool Aid joke was in response to Dr. Pepper and I was referencing the "New Coke" as it appeared in 1985 or so. Maybe you just aren't old enough to remember. Be glad.

Ultimately, it is a free forum and the PR hacks will say what they want and I will say what I want. (As long as all is within TOS here at MR)

 

[ValSalva]

Care about storage performance and expandability, but don't think that the six Thunderbolt 2 (at up to 20Gb/s) and four USB 3 ports are enough? That is up to 36 Daisychained Thunderbolt devices and then there is the ability to use hubs with USB devices. How much are you looking to expand by?

This all sounds good in theory but how is one to realistically manage all the cables and power supplies for these daisy chained devices?

I will hold off judgement until the Mac Pro is released and reviewed but it feels a little like things are moving backwards when it comes to storage. Internal storage and the lack of power bricks and Thunderbolt cables seem like the bright future but now they are somehow the past. But any final judgements are premature.

 

[Dj Lee]

Try the New Kool-Aid. It's sweeter yet less filling.

yes this!!!

 

[Glen Quagmire]

The arguments they pose make no sense and yet they repeat them endlessly.

Perhaps the new Mac Pro fits in with their requirements. Why do you think that your needs and desires are the same - or should be the same - as everyone else's?

I have some misgivings about the new Mac Pro, but that doesn't mean that anyone who disagrees with me (either thinking it's perfect or thinking it's completely unsuitable) is wrong or that their arguments make "no sense".

It's possible to disagree or have a proper discussion about this thing without being insulting or rude. Just because someone is enthusiastic about the new Mac Pro doesn't mean they are a shill or are being paid by Apple to flood the forums with gushing praise.

 

[haravikk]

Apple seems to be betting on improved support for OpenCL based processing, which isn't necessarily a bad direction to take as GPUs can handle some impressive processing work. The question is whether OpenCL will (or can) occur for certain areas, as while it may make a lot of sense for video editing I'm not sure how useful it is for complex program compilation. That said, I'm not all that familiar with OpenCL in practice; it's something I keep meaning to experiment with but most of my work has been with Javascript or Java lately

My point though is that OpenCL potentially gives you access to three processors, it's just that two of them are more specialised.

Regarding storage performance though, I think the new Mac Pro is pretty well covered as USB3 is fast enough for RAID-0 HDD arrays of up to four disks, and you could happily have multiple if you want. And Thunderbolt can happily handle much larger and/or faster arrays. In some respects the new Mac Pro is more expandable with regards to storage, it just sucks if you've used RAID cards up until now.


So I dunno, it could still be a very good choice, we'll just have to see for sure