OS X 10.8.3 beta supports AMD Radeon 7000 drivers, hinting at Apple's new Mac Pro - Page 3

The cube was an interesting even desirable machine but Apple really screwed up with pricing. Badly! In fact the Cube was so poorly priced that it turned me off with respect to Apple for a very long time. I eventually replaced my Mac Plus with a series of Linux machines because of it. The proverbial straw that broke the camels back so to speak as I just could never see a price justification for that machine nor for many that came between the Plus and the Cube.

The Cube want a bad idea though. Something similar today probably would work well unless it was big enough to house the capability of the Mac Pro. Notably the Mini is a far better "Cube" that the Cube ever was. So a modern day Cube would need to be able to house both high performance CPUs and GPUs. That at the very least means a bigger box. Price does kill the Mac Pro even though many don't want to admit it. The problem is that the Mac Pro is really only attractive to people that configure it in rather high performance configurations, which means they are by definition less sensitive to pricing. At the low end, for people looking for really good midrange performance and maybe a decent GPU the machine is a joke and way over priced. Thus the terrible sales. I really don't see the problem either. For the most part, that is for the majority of users the Mac Pro is one big box of dead air. The only thing to argue with is the need for a PCI-Express slot. Such a shrunken machine needs at least a couple of slots as external connections would never be fast enough nor have the reliability some users want. As for bulk storage it simply doesn't belong inside a CPU box anymore, there are now multiple ways to interface such hardware and still maintain the performance required. The two aren't mutually exclusive! You can have slots in a small box and t the same time target performance per dollar. The problem is that certain segments of industry will always need some sort of applications accelerator that can only really be leveraged in a high performance slot. Terrible isn't the word for it. It is down right highway robbery. We need to look deeper though and to try to determine where those high prices come from and what can be done to reduce them. I still see the first order of business should be to rip everything out of the box that can't be justified for making a high performance module. Thus anything SATA related must go, that is almost a third of the box and motherboard right there. Apple does have configuration issues that make the low end machines very in appealing.

As to Ivy Bridge, I just don't see it in a new 2013 Mac Pro. Frankly it really isn't worth waiting for, at least it doesn't justify pissing off your loyal customer base for. Instead I see something from the Xeon Phi line up going into the machine. Not so much the accelerator chips already released/announced but rather a Main CPU Phi that has been rumored about. In other words a chipset that allows Apple to implement a dramatically different Mac Pro and something they might see as justifying good margins.

To put it another way, they need something that makes people say wow. Something that changes the mindset as to the Mac Pros value. Frankly if they rolled out yet another Ivy Bridge based Pro machine, in the same mold as the current Pros, I don't see a lot of NEW users rushing to embrace the new machine. New being the key word as to remain viable the new Pro needs to suck in many more new users. The same logic more or less applies to an Ivy Bridge based machine. Or it will when a stable of Ivy based Xeons is out. Things like TB and other technologies are really pushing us to dramatically different Mac Pro architectures. The thing is what is taking so long? Hard question to answer but nothing available Ivy Bridge wise really seems to justify the long delay in a new architecture. To this I agree, but I really can't see anything compelling in Ivy Bridge Xeons either. Think about it, Apple risked many customers by releasing that "bump" machine a few months ago. Does Ivy Bridge justify that? Nope! At least I can't see anything so compelling in Ivy Bridge that I'd risk my customer base waiting for it. This is why I expect something different, who knows Apple could be partnering with Intel on a Xeon Phi specific for Apples needs. All I do know is that they must have something compelling up their sleeves to justify all the foot dragging and non updates we have gotten.
I was thinking February. I suppose another half year doesn't mean much if you haven't done a real update in four years but the customer base is getting itchy. In any event you would think that we would be hearing leaks or rumors rather soon. .

 

You don't consider Nehalem -> Westmere a real update?

It still contained nehalem options. They had the bottom 2 options on nehalem through WWDC. Now it's one Nehalem and the rest Westmere. To be fair, this was typical.You didn't have a single package cpu from westmere appropriate to the low end configurations. The difference is that the others charged significantly less for their base configurations. It's never possible to make a perfect comparison, and whenever you start adding cto options to any computer, the price increases are substantial. You essentially pay for the part and for the configuration.

 

Chipsets aren't really changing again prior to 2014 at the workstation level. Intel is arguably on a skewed release cycl here, but it's what is available. I agree they need to absorb new users. Typically it takes significant growth that can be leveraged well to grow a product line. Completing a task in 25 minutes instead of 30 isn't enough to move a large portion of users. Taking something from a couple minutes to real time has a much greater impact in encouraging either new sales or upgrades. When you're talking about something that can change the way you work, that drives workstation sales far more than the annual bump where users will either wait out generations or wait for a significant "breakthrough" generation. I find the complaints about size and it should be smaller to be a little silly. Rack friendly might help a few people, but people say a lot of things. They imagine something cool, but this doesn't mean they'll buy it. Show them something that can really advance the way they work and they'll either embrace it or grudgingly upgrade due to the perceived need not to fall behind their peers in efficiency.

 

Thunderbolt wasn't really designed to be friendly to this line. It adds a high IO port where it wasn't previously possible, but they can't route it through integrated graphics with E5s unless something eventually changes there. It won't happen with ivy bridge where they're upping core counts and sticking with the same chipset. Perhaps a board revision with haswell or broadwell could take on integrated graphics rather than more X86 cores. Otherwise they'd need to take a different approach if it's seen as a must have item. I've tried to explain that looking back at the last decade, workstation boards are always used as long as possible. It gives the manufacturers better amortization of costs on a lower volume part and some amount of stability/predictability. Ivy E5s will have the same chipset as Sandy E5s along with the same socket just like Nehalem to Westmere. My other point was that if it's intended as a short or minor release cycle, you may not have some of the cheaper options with Ivy, necessitating the use of some Sandy options to fill in gaps in the line.

But if they make it bigger than the 2000 Cube we're back to the NeXT Cube. And that was, as cool as its design was, a bit too big; you really needed to make room for it. Regular workstations are 'more easily fit' . And will a new Cube then take PCI cards? What would the differentiating factor be to a MP? Na, I think they'll release a same or similar case design - exterior, and they'll surprise us with the inner workings. Perhaps 10 or 12 perpendicular HDD's. Or none, simple a few SSD connectors and your data can be either TB or FC connected. The ODD can be omitted; DVD Replication Services want DVD's delivered on tape anyway.

But aren't PeeCee's with similar specs similar priced? I thought the MP was the same as a Dell. Config depending, obviously. Maybe that has changed; the MP has become more expensive, just like iPhones, Mac Mini etc.

But looking at the review from AnandTech back in 2010 were they compare prices to Dell & HP it's not that bad: (maybe times have changed; I certainly am not getting a Windows PC)

http://www.anandtech.com/show/3969/apple-mac-pro-mid-2010-review

Compare their single package options. Most of the time the differences tapered off as you upgraded with Apple. On the PC side the quad core options were always cheap. Apple wanted to protect imac territory in that regard. They could have bumped the $2500 model further to make up for the high cost of entry. If you were looking at a distinctly single package model with the same parts in the Dell, you'd have to go down to a T3500. It still had drive bays, slots, etc. It was using the W3530 like Apple, only you could configure one around $1200. They were cheaper on the 6 core variant too. They all kind of tapered together when you looked at dual models. Sometimes Apple would be higher, sometimes the others were higher. It's mainly the single model that was grossly overpriced even though they removed all expensive dual socket parts from it. The only thing they shared was a common backplane.

If they do redesign the case then please make it a working road unit by fitting into a rack without retrofitting!!!

 

You often take your servers on the road, do you?

 

That's called a Mac Mini. Intel Core i7-3720QM 2600 MHz (4 cores) = 11595 Geekbench score.

 

In comparison the 2012 Mac Pro Intel Xeon X5675 3070 MHz (12 cores) at the top of the heap is only 22271.

 

If Marvin's fantasy was the future then you can just connect 2 of them together via TB and his "vision" is met.  No need to castrate the Mac Pro at all and remove all the slots that actually makes it the most versatile Mac available.

 

F**k performance per dollar.  Performance AND reliability are key.  There's no way workstations ever beat consumer desktops on performance per dollar.  The design criteria is different which is why we spec machines with Xeon, ECC RAM and yes, slots.  Because different workstations have different needs and having one clean box with faster internal PCIe bandwidth is better (and more reliable) than a nest of TB cables with lower bandwidth.

 

Frankly the only thing you can actually remove from inside the current Mac Pro design to save space with little impact are the drives.  While it's nice to have the space for a 4 drive internal RAID you get better performance out of the 6 drive Promise Pegasus 6 thunderbolt RAID array @ 800MB/sec vs the 550MB/sec with Apple's RAID card.

 

Remove the 2 optical bays and the 4 drive bays and include 2 SSD stick slots for a RAID 1-able boot drive and you'll have a smaller (and quieter) footprint. 

Edited by nht - 12/5/12 at 7:46am

 

We do.  Inside 14U Pelican cases.  Heavy bastards when you have to move them by hand but one survived a helicopter crash...okay, "hard landing".  Fortunately no injuries to the crew and the servers DID boot but we deadlined them anyway.  The case itself was pretty deformed so it appeared to absorb most of the impact which is the intended design.

Most of what I referenced in terms of performance per dollar was related to better leveraging gpu capability whenever possible. If more features can scale better than X86 when run this way, the lower models would benefit considerably. It's not like the 12 core machines can carry the line on their own. I don't really think the thunderbolt raid is that amazing compared to what you can buy for slightly more. The Apple RAID card was always a piece of trash. When you're north of $1000 either way, you can probably find a better RAID solution from Areca  + Caviar Black drives or RE4s if you're building a large RAID 5. They have better supported HBAs than that Apple cto option. Apple is relying on convenience and visibility to market that thing, even if Promise also makes enterprise grade hardware.

They'd at least have to drop from full-length slots to half-length for a Cube and would require a custom GPU. The easiest route for them to go is to take a little off the top of the current one but although not everyone cares about the design, the people making it do. They have to figure out a way to make removing the optical drives not just leave a giant hole in it.

It could be somewhere in the middle. The Sandia CPU cooler is much smaller than the giant heatsinks they use in the current Mac Pro and it could allow enough room for PCI slots in the more compact 8 x 14.5 x 14.5 form factor.

Does this fit in with Apple's long-term plans though? I think Apple wants everyone to eventually use an iMac or a Macbook Pro. If we assume that's the case, they'll likely try to wean Mac Pro buyers onto them. It's true that eventually this means there's no difference between an iMac/Mini and a Pro but I don't see that as a bad thing. I think weaning is better than them just discontinuing it.

It certainly wouldn't alienate all Mac Pro customers nor leave it castrated, it would just focus on performance over expansion options.

Apple historically takes the path of least resistance with the Pro so they could even do the same thing they did with the old-style MBPs. Just drop in Ivy Bridge and leave it with opticals, same design, same price points. I think it would be a pretty lame update after 3 years but that would indicate a stronger liklihood of dropping it. A significant redesign suggests continued investment.

They can do whatever they want. If they don't build it the way people are asking, there really aren't any negative consequences. The only option is to buy an alternative.

I'd at least like to see a modern Cube replace the entry quad with a 6-core Xeon. That makes expansion a premium with the higher models.

There's a reason I don't think something like that will happen. The cheaper model does carry a portion of the volume for the line. They're trying to encompass the entire range with mostly shared hardware. It's just that the low end models avoid some of the costly parts, essentially expensive cpus + chipset. There are some machines from other oems that can be configured single or dual even though they use the same thing. I really really agree that a 6 core at the starting position would pump some needed value in there. What we have right now is fairly underwhelming. It's just that Apple's trend with the line has been toward higher margins rather than lower, so I remain skeptical. I'm also skeptical on Wizard's idea of a Xeon Phi. Nothing intel has shown publicly aligns really well there, and I would assume Apple will try to control design costs. The one thing a radically redesigned case would do is tell people that they're still interested in the line. The inspired confidence could be a good thing, but I think it would still need some way to grow. The obvious method would be to grab Windows customers that are otherwise purchasing Dell or HP workstations.

Maybe all they will do is at a 7970 and make the macpro a bit thinner.

 

This was in reference to Marvin's comment: "I think the Mac Pro's area of emphasis needs to stop being expansion and start being performance per dollar."

 

The Promise Thunderbolt RAID is high end prosumer that can replace pro level fiber channel gear using consumer parts (Thunderbolt + SATA drives).  I think you can put SAS into them if you want.

 

The $1000 8 Bay Areca is slower than the Promise at 220MB/sec sustained reads and 210MB/sec sustained writes and is unpopulated.

 

http://www.arecaraid.com/product_info.php?cPath=28&products_id=58

 

"The R6 really stretches its legs when we measure sequential read performance in RAID 0. It falls just 50 MB/s short of the 1 GB/s barrier at a queue depth of 16! If redundancy is more important to you than blistering speed, RAID 1E operates effectively as three RAID 1 arrays, which is why sequential read performance falls to ~315 MB/s. RAID 5 is a fair compromise, enabling block-level striping with distributed parity. We see sequential reads top out around ~780 MB/s in that arrangement."

 

http://www.tomshardware.com/reviews/my-book-thunderbolt-duo-pegasus-r4-2big,3222-6.html

 

Apple is pushing Thunderbolt, not Promise.  That Promise has the only thunderbolt RAID I know of is why they get shelf space on the Apple store.

 

There's no need for a Cube form factor.  There's no need to go to half length cards.  There's every reason to include full length cards for folks that need high end GPUs.

 

Refactoring the space after the removal of all the drives but keeping full length slots isn't as difficult as you make it out to be.

 

 

 

Because without slots you might as well not bother.  The number of folks that need Xeon+ECC+OSX without wanting a slots isn't that large.

 

 

 

Xeon Workstations and pro-video cards typically do not beat consumer CPUs and GPUs on pure performance.  The emphasis is on accuracy of the results and reliability of the hardware under long term heavy loads.

 

 

 

Or they could start making them in the US.

 

 

 

I have always stated that this is another lame attempt to foment for an xMac.  I don't care if you want Apple to make a new cube with a few slots.  That would make me happy and I would buy one.  

 

But it pisses me off when this desire translates into "Apple should castrate the Mac Pro people use to make the xMac I want to buy because I don't give a shit about their needs".

The price is quite high for the standalone board:

http://newsroom.intel.com/community/intel_newsroom/blog/2012/11/12/intel-delivers-new-architecture-for-discovery-with-intel-xeon-phi-coprocessors

"The Intel Xeon Phi coprocessor 5110P is shipping today with general availability on Jan. 28 with recommended customer price of $2,649. The Intel Xeon Phi coprocessor 3100 product family will be available during the first half of 2013 with recommended customer price below $2,000."

That's for 60x 1GHz cores, 225W total:

http://ark.intel.com/products/71992/Intel-Xeon-Phi-Coprocessor-5110P-8GB-1_053-GHz-60-core

You couldn't even put one of them in the current Mac Pro alongside a decent GPU. However, they could go the custom route. I don't think Apple would want to run Linux processes on it anyway, they'd want to run OS X processes natively. They could get a custom chip from a 2013 model, likely with fewer cores and put it on the motherboard. GPUs are still much cheaper though.

Growth is created by shrinking things down. The high growth market is now mobile. Some buyers want the biggest box with everything packed inside so they don't have to buy another machine for life but that sort of thing creates zero growth for Apple. PC manufacturers went this route years ago packing quad-cores into towers with 12GB of RAM, massive drives, GPUs with lots of video memory and cutting margins down. They are now struggling badly because they have no market left. Those machines aren't popular and they don't have enough profit.

Internal expansion only appeals to a very small specialised market so I think performance per dollar and compactness are the best ways to go. I still don't expect massive growth but it's a clear enough message that a transition is happening without the rug pulling.

We don't know that for sure.

It's not about the xMac at all. That was an attempt to get an iMac without a display at a lower price point than the iMac. As I said, it's not castrating the Mac Pro, you would still be able to connect PCI peripherals externally, especially with half-length slots. At worst it's chopping off one ball and you can still have kids. They just might not be that attractive.

You can say Apple's current design is the result of not giving a shit about some people's needs - people who need 8 GPUs or 4 processors. One person's compromise will never be the only viable one but Apple has to make the choices from a business perspective.

 

That doesn't make any sense.

 

Oh yeah...I can see a bunch of guys lining up for that.

 

There aren't a lot of options for pro buyers and reducing capability in what is the most truck like of macs is exactly the wrong way to go.  

 

The mini is already an 80% solution for what you are asking for.  It would be a 90% solution if it had 2 more RAM slots and drivers for GPUs over Thunderbolt.

 

There are NO other options for full length, full speed cards other than the Mac Pro.  That's a unique capability.

I know that the reference to Phi right now results in people looking at Intel current co processor boards.    however Phi is part of Intels initiatives into high performance computing part of that involves buying in to InfiniBand.    There have been rumors and reference to Intel integrating InfiniBand into a Xeon like chip and marketing that as part of the Phi family with the Phi family become a growing HPC initiative on Intels part.    So Apple and Intel could be dreaming up something different along the lines of a "Phi" part.   

 

The possibilities here include a Phi coprocessor chip with fewer CPUs which would directly impact costs and still provide good performance boosts to many apps.    The other possibility is a "Phi" main processor chip with InfiniBand built in.   Information on the chip with InfiniBand built in is hard to come by right now.    

 

However lets say such a chip did exist and Apple was interested in it, think about what that would mean for the Mac Pro.   A machine with InfiniBand built in, with supporting OS additions, could easily become the high performance computing module that many of us have been thinking about.   It would provide for a small machine that could be bonded together with other machines into compute clusters at relatively low costs while still supporting traditional GPUs and maybe even Phi coprocessors.

 

In the end every machine is a compromise. Be it a metal lathe in the factory or a sports car on the highway, they all have real limitations that make them less than ideal for some uses. What people need to realize is that something has to change with respect to the Mac Pro or we won't have a desktop Mac for professional users any more. To address that Apple needs a Mac Pro that appeals to a wider array of users. If people can't grasp that then they are part of the problem with the future of the Mac Pro. I obviously can't say what Apple has up its sleeves for next years Mac Pro, however they need to address the rather pathetic performance value of the Mac Pro especially at the lower end. if they don't it will go the way of XServe and a bunch of other failed products at Apple

You really need to get over this idea that if they change the mac Pro it will result in reduced capability.   I mean really it doesn't have to be that way.

The Mini is not the solution we are asking for.   Mini is an entry level device that frankly shouldn't even come into this discussion.

 

As to the cards, please explain to me why we need that massive tower to support PCI Express expansion cards!   It is asinine to associate that waste of space with the need to support these cards.                                              PCI Express can be easily supported in a much smaller box and has been on countless other platforms for years.

Not quite. They could build an 8" Cube with 2 half-length x16 PCI 3 slots. One would have a custom GPU in it and the other would allow you to connect a PCI breakout box at 128Gbps, like this one with 16 slots:

http://www.magma.com/expressbox-16-basic

One of the cheaper ones would give you the functionality you get in the current Pro and for anyone using a single half-length card, they can use the slot directly. I know you're going to say 'rat's nest' again but only for a few people who would have that setup.

It also means that if you had a group of computers for a variety of work, you could pack 4 red rockets into a single PCI enclosure and pass it between workstations. You can have one filled with GPUs for computation, you could have one filled with audio DSPs, all switched with a single plug. It means if one Cube goes down too, you just plug your $30k+ equipment into another one within seconds.

For people who buy for performance, they get a more affordable and attractive box without being lumped with a "big iron" system.

I like the idea of supporting Thunderbolt through the lineup and I think neglecting Thunderbolt in the Pro will divide the market for high-end peripherals but this could be a route to go first. When Thunderbolt reaches 50-100Gbps, they can replace the slots with ports.

 

 

If you remove all the slots as is suggested then yeah, it's reduced capability.

 

 

 

Entry level?  Lets see which macs out perform the $899 Mini:

 

Mac Pros.

XServe.

Top end 2012 MBP Core i7s (by 5%).

Top end 2012 iMacs Core i7s (by 7%).

 

http://browser.primatelabs.com/mac-benchmarks

 

It out performs the 2011 Core i7-2600 27" iMac, the 2011 MBP i7-3615QM and the MBP Retina Core i7-3615QM.

 

In what universe should this machine "frankly" not "come into this discussion" about compute performance?

 

 

The tower isn't massive because it has 3 full sized slots.  I'm also not asking to keep the large size...just the slots.  I do expect any solution to be equally quiet and robust. 

 

There's no reason not to do full length cards and if you leave sufficient gap for a double wide card then no specialized GPU required.

 

I dunno why you want a cube.  A slim tower design slightly larger than the Alienware X51 can handle 2 full length cards.  The X51 only handles one 9" card but it's a pretty slim tower.  

 

It's really not the 3 slots that makes the Mac Pro so large.  Making them half length will not make the machine all that much smaller.  Just removing the optical bay will slim it down significantly to normal tower dimensions. 

It makes the minimum chassis size 12", which is larger than it needs to be for everything else. The 14.5 x 14.5 mockup I did would fit full length slots with a significant size reduction over the current one but I still prefer the Cube.

Isn't the X51 too narrow for a full height card? They'd also have to put the hard drives in the other way - I think HDDs are still needed for bulk storage in the Pro, otherwise it ships with less space than the iMac.

An 8" Cube allows full 4.2" height PCI cards but half-length. Like I say, if you need full-length cards, they can go in a PCI box, no big deal.

One main reason for the Cube is it creates a more iconic computer design. The Mac Pro is just a tower shape like any other. It looks better but it's not sigificantly different. I'm well aware people go on about it not being about the aesthetics but the reality is, we all like looking at attractive things. Cubes wil sit on people's desks instead of under them and people will want them. They'll efficiently fill server bays like the Minis do at Mac Mini Colo.

Half-length is 6.9", full-length is 12.2". The half-length one will fit in an 8" Cube chassis, the 12" ones will require a run-of-the-mill design because you can't make it 12" wide.

I'd say 2x 3.5" (these are 4" wide) drives at the top up to 4TB each, same install method as the current one. 1" high.
Below that, two 6.9" long, 4.5" high PCI cards, one double-wide, 1" fan at the front. 3" high bay.
Below that, a single CPU (up to 10 or 12-core - 10-core CPU alone should cost ~$1800) cooled by a Sandia-like CPU cooler, 4-6 RAM slots and the PSU, which would be maybe 600W (150W CPU, 300W slots), narrow enough to allow airflow from the front, over the CPU and RAM and then past the PSU. 3" high bay.

Somewhere in there would go an SSD blade on the motherboard with up to 768GB storage that could make a 4.7TB Fusion drive.

It would be attractive, easily portable, suitable for server use, better performance per dollar and still allow the option for as much expansion as you could ever need. It would have to be 6-core at $2499 and 10-core around $3999. If Ivy Bridge has a 12-core, that would be a further option.

The new cube is the Mac Mini.

 

This is like the optical drive thing. You can still use them but externally. Internally, there would be an option for the best GPUs you can buy and it would be the exception that people look for alternatives. If you need a specialized card like a Quadro or FirePro, they can be housed in a PCI box (remember this is running at x16 speed) and better, you can have 4 or more, which you can't do in a Mac Pro anyway.

If you saw one next to the other, the size difference alone is pretty staggering:



It wouldn't need handles really but it could have the same kind of setup as before - the above kind of design means you can stack vertically. With a 10-core Ivy Bridge, it would be faster than the old one and cheaper. Like I say, you could do everything you did with the old one except full length cards go outside but higher power ones would have to anyway.

 

LOL...that's smaller than the area of where the CPUs currently live.

I figure the next Mac Pro, should it be redesigned, will look a fair bit like the Mac Mini. 

 

It won't look like a pizza box Performa or anything but when you look at it, you'll immediately think, "Hey, I can see where the Mac Mini comes from."

See you're impressed already. One of the current CPU bays is 6" tall and 8" long (including fan), This would have a CPU bay of 3" tall and 4" long but the bulk of what uses that space currently is the giant heatsink, which is many times larger than the CPU:



The CPU underneath looks like this:



The heatsink doesn't need to be that big:

http://www.extremetech.com/computing/131656-the-fanless-heatsink-silent-dust-immune-and-almost-ready-for-prime-time

"In the Sandia Cooler, the heatsink itself is the fan. It is a cast metal impeller that floats on a hydrodynamic air bearing just a thousandth of an inch (0.03 millimeters) above a metal heat pipe spreader, powered by a brushless motor in the middle. The end result is a cooler that is very quiet and 30 times more efficient than a fan-and-heatsink solution. The prototype (shown above and in the video below) is 10 times smaller than a commercial state-of-the-art cooler, but has the same cooling performance."

It has a diameter of 3.9" and height of 1.1", which fits in its 4" x 3" space nicely and leaves about 4" depth for the 4-6 RAM slots.

I think forcing the use of third party options like PCI boxes where they weren't previously required is beyond stupid, but we're never going to agree on that. Cheap eSATA boxes and firewire drives have been available for many years, yet a lot of them were terrible.  The current card implementation in the Mac Pro is okay. GPU death does seem to be one of the first points of failure on them, but I can't see this as an improvement. This is one of those areas where Apple and Intel seemingly dropped the ball. The tech sites suggest that Apple got a first year exclusive on thunderbolt. All that kind of thing does is limit third party development. Pushing cards off into a small generic box really is a terrible idea, whether or not you recognize it. It's just not possible for them to accommodate the same range of cards. Something like this makes more sense. The thing is tested and sold as a unit.

 

 

 

It's best for these things to be tested as a unit. If pricing drops on external boxes, between an expensive graphics card and the box, the box is going to be the low margin item. I don't expect much in terms of quality. If it's handled like this where you purchase a unit, you'll likely see more consistent results, assuming they can sell enough to overcome high fixed costs.

 

Yow, that's an ugly daughterboard. Why doesn't my daughterboard look that bad?

 

I'm really liking the 8" cube idea, though. Though wouldn't it be 7.7" to fit with the current trend?

I know this was an option from third parties in the past but perhaps an Apple built expansion box for cards (that is as fast as using slots internally) could come from Apple allowing a smaller box for the mothership? That way those not requiring as many slots can buy in and also expand should their needs grow. Cooling the mother ship would also be helped since those cards can throw off some wicked heat.

 

And the discussion has come full circle…

 

 

 

BTW I meant the 27" iMac.

As fast isn't possible, and this would be more expensive than the traditional route. It's not much of a cost saving measure. I just said if you're going to do it, the unit should be optimized as a whole. Don't expect native eGPU support to originate under OSX either. Macs never received much in the way of after market gpu options. Once you take away the mac mini users who base their purchases on cost effectiveness and the 27" imac users who settled for what was available, you're somewhat limited on people who will pay potentially $600 for a gpu upgrade. I don't foresee it being too much less if it's a gpu + box in one. They'd have a certain amount of development costs to overcome, and there's no reason to go for a product with razor thin margins there. The only reason they sell low end cards under Windows is to help overcome the development costs of their higher margin products. Keep in mind there's a lot more to it than just sticking a gpu in a box. If you want to be able to plug it in without rebooting, that alone requires some amount of firmware tweaking. That's something you'd normally only find in servers. Even then I don't think it's that common.

You can't build an external box that can operate as fast as an internal. If nothing else the distance makes it impossible but the extra lanes commonly used in PCI Express boards makes it impractical. Beyond that as motherboards run at higher and higher clock speeds they get smaller to actually achieve those clock rates. Bus interfaces get standardized around fixed electrical specs to isolate the cards or expansion slots from the other buses that continue to evolve to higher clock rates.
It is interesting that effectively you propose a solution that is equally ineffective in the opposite direction. External slots are simply in effective for many uses. Apple needs to find a sweet spot that serves as many users as possible, the goal being to actually enhance Mac Pro sales. One thing remains clear, Mac Pros future hangs upon stopping the slippage in sales and turning the product around, that means finding a better balance between price and features.

To an extent I'm often suggesting the same thing, that is a more modular computer. The problem is you can't give up the most important things that make the Pro a professional computer, that is the slots. Slots mean that users can configure the machine to their needs. The trick is finding the right allocation of slots considering the realities of external buses, hardware built into the motherboard and other initiatives. Lets face it a box that is priced out of the market is of no interest and likewise a box that can't pull its own weight in professional settings is of little use. Right now the current Mac Pro design is a relic from the past in my opinion. It is sort of like the mainframes of the past, yeah you can find mainframes these days but the use is far more specialized than in the past. The problem is Apple isn't structured to support the mainframe mentality nor is the customer base willing to accept mainframe like pricing. This is why I keep coming back to the idea that the Mac Pros replacement must be designed to appeal to a broader array of users. Increasing sales to the point that the Pro actually supports an engineering team to maintain it over the years is vital.

1. OSX only supports 96GB RAM
2. What do you want, a machine to fit it all in, or a pretty machine?

There's a good case study. Link a machine that likely costs into six figures fully configured sold by a company with multiple bankruptcies over the past 15 years. It's not really SGI anymore. I didn't think whatever company owns them now would still use that tarnished brand name. It's a rackmount form factor, not really designed for use as a single user workstation.

Obviously Apple would need to upgrade the RAM limit in OS X. As for #2, you can have both with the right design
Edited by z3r0 - 12/15/12 at 8:57pm

Base price for the Octane 3 was $8,000:
http://www.maximumpc.com/article/news/sgi_unveils_personal_supercomputer_octane_iii

and it fits under your desk.

SGI bankruptcies were due to poor management and marketing.