Will there be a new Mac Pro in 2015?

[deconstruct60]

...
Also, I really don't know why people got so "glued" to the TB3... What if next Mac Pro will use Haswell - EP CPUs and will not support TB3?

There is no tight coupling between TB3 and current Intel CPUs; even Intel says so. Skylake is pragmatically needed on many mainstream desktop/laptop designs because there are little to no free PCIe v3 lanes available. In those systems, the TB controller is typically hooked to the PCIe lanes of the PCH ( I/O chipset). The PCH chipsets of the Skylake era are getting PCIe v3 upgrades. That's the "Skylake" coupling. It is not that there is something magical inside the CPU. Using the CPU code name to "name" the coupling is just deeply misguided. That isn't the root cause issue.

Apple could do TB v3 with E5 v3 ( Haswell-EP) in late 2015 (if could get their hands on enough TB v3 controllers). It is just that it is a waste of effort since can do exactly the same thing with E5 v4 (Broadwell-EP) in early 2016 too (when TB v3 controller volume availability won't be an issue and the ). If Apple is waiting for TB v3 to do something, then E5 v4 is the more natural fit along availability timeline.

No other system has put more than one TB controller in it. So the Mac Pro has become the poster child for "bleeding edge" TB usage in some folks eyes. There is a more than decent chance that includes the some decision making folks at Apple.

 

[deconstruct60]

....
It wouldn't surprise me to see Apple drop the GPUs down to PCIe 8x, instead of 16x. This would give them more than enough bandwidth with minimal performance impact.

It is minimal impact for games that hoard data inside of VRAM and can use "cut video" to mask large cache restructuring. For a system which is processing a broad set of OpenCL computations and/or a broader set of non "taking over the whole screen by a single app" it has performance issues.

Apps tuned to PCIe v2 x16 may performance OK on x8 PCIe v3, but 2-3 years down the road is that going to be where most users are in the Mac space? Not likely. There is little to no rational reason to optimize the Mac Pro for the older, stagnant portion of the overall PC industry.

 

[deconstruct60]

dec
I know GDDR5 is dead, what I was saying is that, since HBM is still in it's infancy an improved Tonga (and I mean Tonga since it's the latest design, GCN 1.2) would be a good bet.

It isn't a good bet. GCN 1.2 is explicitly indicative that have already tweaked the basic core design twice. What large gap do you think they have missed in the previous two tweaks. Going to a third isn't likely going to produce much that was punted earlier because it wasn't implemented correctly in time for release.

if look at the 300 "rebadge" sequences covered in the anandtech article here

http://anandtech.com/show/9387/amd-radeon-300-series

in most cases they aren't using the "max cores". Some of these tweaks to higher clock rates come as offsets to fewer cores. The basic design is missing bandwidth head room. They have added compression and have more and better driver tweaks coming down the pipeline. However, the hard coded design premises that GCN 1.0 started out with are a limitation at this point. Iterating a third time on 28nm probably isn't going to buy much.

That is one reason why just throughing more VRAM as an "next iteration" improvement option. With GDDR5 that is a slippery slope since that tends to mean more power too ( or at least consuming the power saved by refinements elsewhere.)

The design team would even be the same.
I guess that way Fury wouldn't stand out that much though.

It isn't Fury that is the primary issue. How do they pay for a substantially different design team with money they don't have? Doing incremental optimizations to a largely design and almost completely debugged design can be done with a much smaller team on a much lower budget.

But who uses more than 2 ports anyway? Some will say yes, but I'd guess a minority. And with everyone claiming the death of TB cause of USB-C, well, that could be the way to go. Don't get me wrong, I believe in TB.

The folks claiming the death of TB primarily are in part basing that on TB ports primarily being used solely for legacy DisplayPort mode. if vast numbers of TB ports just do DP only and USB Type-C can do DP only. Then there is no differentiation. USB Type-C is somewhat competitive in single usage modes, but in multiplexed usage mode it has distinct handicaps. So the question is more so "what is plugged in directly" more so than "more than 2".

In part, the Mac Pro has 6 ports do deal with mini-DP displays. If the Mac Pro gets loaded down with 2-3 legacy displays, then there are still more than one TB ports left for devices even if have to assign one TB port to high speed networking ( FiberChannel/Bonded Ethernet/ etc.). The TB panel on the Mac Pro is covering the functionality that mainstream PC GPU card edge's provide. The custom card in the Mac Pro decouples the interface edge ports from the card.

There are likely substantial number of Mac Pro users using more than 2 ports, since multiple display set ups are commonplace. The question is whether Apple will move more of that graphics card edge functionality to other places on the "back" panel other than the TB and HDMI port sections. With USB Type-C connectors conceptually this could be partially be moved into the USB subsection also.

For two of the TB ports I think they will. I'm not so sure about four . It depends upon whether Apple wants to leverage doing 5K displays over TBv3 or not. With only one controller ( two ports ) the Mac Pro is limited to just one TBv3 5K display. With two you can drive display which is probably more than enough for the vast majority of folks.

The current Mac Pro launched with some emphasis that it could handle three 4K ( biggest at the time) display. Sliding to just two 5K isn't much of a retreat. Sliding back to just one is. If Apple is punting on single cable 5K over TBv3 then it doesn't matter. Just having 6 fully capable DPv1.2 sockets ( whether TBv3 or USB Type-C DP alternate mode) can do 3 5K displays (if the GPU doesn't groan to a halt under the load.)

At this point, I wouldn't bet against a 5K Display Docking station from Apple. Maybe Apple is going to punt external 5K until DP v1.3 is widespread. Maybe, but I wouldn't bet on that at this point in time. I think DPv1.3 is going to roll out slower than many folks think. Intel's TBv3 slide deck is so hyper about single cable 5K on TB that it seems likely they have an implementor out there trying to do it ( or minimally intensely asking for the capability). Given the small number of Display Docking Station vendors out there, that puts Apple at the top of the suspect list.

If Apple really, really, really wants a x4 PCIv3 SSD sooner rather than later (and Tv3 5K is not a priority ) and most users are actually miniDP consumers, then perhaps they would do something like

4 Type A USB ports
2-4 miniDisplay ports (smaller if HDMI port(s) gets its own source)
2 TBv3 ports. ( overload with USB Type-C connectivity also)
1-2 HDMI port

That way folks could still use their legacy Mac focused Displays without an adapter. No adapter for the USB ports, and a denser concentration of data focused devices to the TB ports. It still leaves door open for single TB 5K Display Docking Stations.

TB as only "external PCIe" has problems though. That is is the disconnect of folks declaring TB "dead" with USB Type C. That isn't what it is. TB never was just "external PCIe". The bundling of usage/modes is even higher in TBv3 since the USB controller is being bundled in also. Given up TBv3 ports also means giving up independent controller USB port bandwidth too. Even USB only fans will find utility in that additional bandwidth (while ignoring the other features TB provisions).

I don't really see very many folks complaining that the current SSD is "slow" in the current Mac Pro. I think there are far more Mac Pro folks doing "sneaker net" with fast external drives than folks who want to hoard maximum TB internal to a single box. More high end independent USB/TB bandwidth is probably better. Faster external RAID and sneaker net box connectivity is a more pressing need.

Dual SSD slots will pragmatically only help a new Mac Pro design if one of those slots is standard M2. Two pragmatically Apple proprietary slots isn't going to help over the very long term as the supply of alternatives is likely to remain limited.

 

[deconstruct60]

.....
By the way, a year and a half ago I posted a mockup of a redesign of the 6,1 to support dual CPUs:
....
It's a crude graphics job, but I just made it taller so that a second CPU and set of DIMMs could be put above the first.

Way more than crude. DIMMs are 135.55mm (about 5.25 inches) long. Double stacking DIMMs vertically would just about double the height. If keep the height and width proportional then blow out width also. It is as goofy as putting 4 wings on a basic Wright Flyer design (up from the original 2 ). It shows a disconnect from what the design's core principles actually are.

A smaller desktop footprint than the Mac Mini is nice but necessary. A grossly larger desktop footprint than the mini is missing the boat. The new Mac Pro is designed as a desktop computer. A real one, not a floor standing model broadly classified as a desktop. A single seat airplane and a 200+ passenger airplane are not even remotely close to being in the same design class either.

 

[td2243]

I haven't been here since WDCC and it's nice to know I haven't missed anything.

 

[AidenShaw]

Way more than crude. DIMMs are 135.55mm (about 5.25 inches) long. Double stacking DIMMs vertically would just about double the height. If keep the height and width proportional then blow out width also. It is as goofy as putting 4 wings on a basic Wright Flyer design (up from the original 2 ). It shows a disconnect from what the design's core principles actually are.

A smaller desktop footprint than the Mac Mini is nice but necessary. A grossly larger desktop footprint than the mini is missing the boat. The new Mac Pro is designed as a desktop computer. A real one, not a floor standing model broadly classified as a desktop. A single seat airplane and a 200+ passenger airplane are not even remotely close to being in the same design class either.

The MP6,1 is just slightly under 10 inches tall. How could adding 5.25 inches double that?

 

[PunkNugget]

Ahhhh, just build a Hackintosh and be done with it. Sure you'll only have a 32 core "cap," but it'll sure be faster than any current Trash-Can Pro out there…

 

[ManuelGomes]

dec, Like I said before, I believe Apple is waiting on Broadwell and TB3 wide availability, although I believe they get first (and most) silicon form Intel, both TB3 and CPUs, but still I don't believe we'll see the nMP before early next year at best, although I'd believe they'll present it or start talking about it in the Fall.
Why are we even talking about games on the nMP? This isn't really a gaming machine, although it would be nice if it fares well at it!!
I don't see them toning down the lanes for the GPUs, not now or ever, they bet a lot on it to cap it now.
I guess you missed my point regarding Tonga or it's derivative. I wasn't saying they should have tweaked Tonga into a new core, much the contrary. They should have used it for the top 300 series as well, doubling the silicon. Which in fact is a lot of what Fury is all about really. So, imagine Fury core without HBM but GDDR5. Fury would still be the top player cause of the new HBM and toned down double Tonga would become 390 series, possibly with full 4096 cores (which could be a problem and compete with Fury) and 512bit mem bus, like Hawaii. But the 8GB would also be a problem for Fury, a card with the almost same specs and more mem looks a better deal for most. Maybe that's why they didn't do it. It would make sense though at this time to have the top dogs with GCN 1.2 and the latest tweaks, at least to be able to say it's a new GPU rather then rebadges.
It could look like this:
D710/390X - 4096 cores, 512b, 8GB - super/dual Antiqua XT
D510/390 - 3072 cores, 384b, 6GB - super/dual-Antiqua Pro
D310/380 - 2048 cores, 256b, 4GB - Antiqua XT with mem bus capped to 256b but with compression
Like I said, AMD would loose the punch with Fury, since the core GPU would be almost the same, and have a problem regarding mem size that some users might not understand.
Still, Fury sold well so I guess it could be less of a problem than we think.
Benchmarks still show that HBM has it's benefits.
Like I said, since the core is very much the same, no new team was needed, the same that tweaked the basic core (not the mem interface) would do, since no big differences exist.

Regarding the ports, you pretty much summed it up. I would get rid of the HDMI but I guess some people would need them. Specially since DP 1.3 will be here later rather then sooner. But since AMD only has HDMI 1.4a for now as well I really don't know if it does any good.
Don't really count on a second SSD, much less M.2, they won't go for it.



Punk, trolling are we? 🙂
Good that works for you anyway. Some prefer not to go that route...

 

[deconstruct60]

I guess you missed my point regarding Tonga or it's derivative. I wasn't saying they should have tweaked Tonga into a new core, much the contrary. They should have used it for the top 300 series as well, doubling the silicon.

Tonga is a dubious starting point. From the Anandtech article I linked before.

"... Unfortunately this also means that we’ll be waiting for another day to see what a fully-enabled Tonga would be like. AMD has to date not shipped a fully enabled chip; it has since become clear that Tonga does in fact have 6 memory controllers (for a 384-bit bus) as opposed to the 4 we see active here, ... "

It is not fully enabled in its current state. Copy-and-pasting a double size version isn't likely going to work well. One of the contributing factors here is probably the compression they added in. It has a smaller VRAM footprint but at some point you have to uncompress do work with the data. There is a pretty decent chance they have overloaded the internal bandwidth internally somewhere (e.g., in/out of the cache). They probably need a new internal connectivity lay out and a refactored cache hierarchy.

Which in fact is a lot of what Fury is all about really. So, imagine Fury core without HBM but GDDR5.

In that Fury needed a new internal connectivity due to higher data bandwidth requirements. Yes. But tagging that width ( 1024) with GDDR5 is quite highly dubious because going to lead to power problems. Super wide but relatively very short distances on package can work. Off package and longer distances ... not so well. Even Titan X backs away from

Fury would still be the top player cause of the new HBM and toned down double Tonga would become 390 series, possibly with full 4096 cores (which could be a problem and compete with Fury) and 512bit mem bus, like Hawaii. ... It would make sense though at this time to have the top dogs with GCN 1.2 and

Hawaii is a Iteration on Tahiti ( better memory management). I doubt Tonga could finesse its way past Hawaii. Hawaii/Tahiti throw alot of brute force but were always tracked to get the most out of GCN 1.x

The problem with AMD is more so the GCN 1.0 versions that are still lingering in the line up. GCN 1.1 and up are the path to shared memory management (OpenCL 2.0) and a few other tweaks (True Audio is nice but I'm not sure Apple is going to leverage it much since it is proprietary).

Tonga was probably in part a proof of concept for things they wanted to work on for Fiji (Fury). It wasn't 100% perfect, but if they took those lessons and folded them into Fiji and the next gen with HBM2 then that works. I think your mental model of the hardware design pipelines length is too short. Tonga just emerged last year. There is no good reason why they'd be redesigning Tonga before it finished. A redesign likely would stretch into late 2015.... which won't help now and doesn't do much for 2016.

The other problem is that the 512bit bus like memory demands aren't going to work well with AMD's APU designs. There is no GDDR5 there at all. AMD is competing not just with Nvidia. There is Intel integrated competition too. That consumes some resources. AMD can't pursue every possible path of development.

 

[koyoot]

All of the GPUs that will land in next gen Mac Pro are already on the Market. Grenada, Fiji, and whatever the Tonga iteration now is.

Manuel, you completely forgot that Fiji IS Tonga technology with twice core count and High Bandwith Memory. Why invent the wheel again?

 

[AidenShaw]

dec, Like I said before, I believe Apple is waiting on Broadwell and TB3 wide availability, although I believe they get first (and most) silicon form Intel, both TB3 and CPUs, but still I don't believe we'll see the nMP before early next year at best, although I'd believe they'll present it or start talking about it in the Fall....

Any facts to back that up?

I get prototype systems from my suppliers long before production. The big boys field test and refine their stuff against Intel's early chips. Once the chip goes to production, there is another round of "release candidate" systems for testing - since the "big boys" will have to wait anyway for Intel's production to ramp up.

While Intel is ramping, we're doing system validations against the early production chips.

Intel doesn't give Apple priority access. Apple might use a chip that nobody else wanted (so that it looks like an exclusive). Or Apple might skip a field trial of a chip and ship early production samples - since their sales of pro machines is so small the sample chips are enough.

 

[PunkNugget]

Punk, trolling are we? 🙂
Good that works for you anyway. Some prefer not to go that route...

You actually consider what I said to be "trolling?"

Here's the definition of - trolling:

"make a deliberately offensive or provocative online posting with the aim of upsetting someone or eliciting an angry response from them."

Now I know that you're making a comment with a "🙂," but what I'm doing is no where near trolling. Now saying something like:

"The current "Trash-Can" (piece of crap) Pro out there can't even compare to my amazing machine that was built in 2012 and is still considered to still be faster and far more upgradable compared to anything that Apple can come up with."

Now THAT'S Trolling. - LOL !!! 😀 I'm just messin'.😎

Seriously though, Apple really should've put out something a while ago, since every App company out there; especially Adobe and other's like them have been supporting Windows for sometime now and Apple was known for making kick-butt products that schooled Windows based PC's (when it came to performance). That's no longer the case. It's even gone to the point where Mac OS X is capped at 32 cores, and there are processors that go way beyond that point now. Of course they're basing it off of their current machine. But once Apple decides to put out a Dual CPU machine, their will be some 17 year old kid that shows you how to put together a $1,700 to $2,500 PC and install the newest Mac OS X that will out perform the same Mac Pro that will cost (with comparative PC parts) about $7,500 or more.

What apple should do is build a Dual CPU setup that is able to install at least 3 x GPUs (like the Titan X's that push CUDA cores for apps like Photoshop, Premiere Pro and the like), and be able to allow you to water cool your system (as water cooling has been around for over 15 years now). If they could do that I would definitely look at possibly purchasing a set up like that. Who knows they might go that route and that would be very nice.

Here are a few examples (on YouTube) that show off some of this amazing talent that home PC builders are putting together using PETG Tubing (it's a much cleaner look) that Apple can put together:

A 60 sec vid that shows you how a PC is made:

Go to minute marker 8:58 and see the finished product:

And this masterpiece. Just watch the whole thing:

Apple could easily do this and be a part of something amazing and that's not trolling… Just my thoughts…😎

 

[Dark Void]

I love the design of the pre-2013 G5-esque Mac Pros ... too bad it's out of my price range.

 

[PunkNugget]

I love the design of the pre-2013 G5-esque Mac Pros ... too bad it's out of my price range.

I still like them too, I only wish that they could do something to make them work with multiple GPUs and have them water-cooled with a side window option…

I actually did a build (found here) read Post #7:

http://www.insanelymac.com/forum/topic/287857-new-macmod-2013-build-lemon-lime-twist/

that used an all aluminum PC case (by Silverstone Tek) that I used and that case (although a bit longer) could easily fit 3 GPUs. They could model it after this vertical config as it's actually better as it draws in cool air from the bottom of the case and expels warm air out from the top. I'm sure that they could use a 10.5"x12" Supermicro Mobo to fit Dual CPUs. That would be an amazing machine. I used it for a bit as a Hackintosh but soon turned it into a gaming machine. So I went back to using windows on one machine and my other one (The Hackinbeast) is still using Mac OS X 10.8.5.

https://www.mediafire.com/view/?7iw9act9kwo1d72

https://www.mediafire.com/view/?a1x9i26yd1gtr17

https://www.mediafire.com/view/?58i1e2n7m111ek3

https://www.mediafire.com/view/?2f80c17gsk07xc7

https://www.mediafire.com/view/?js2eteq6dgcwmty

Now obviously it doesn't need to have this type of tubing, but the basic concept of the setup is what I'm talking about. Again, they would have a slammin' machine if they went after making one that you can upgrade over time...

 

[ManuelGomes]

dec, again I said Tonga with 512b mem bus would be better than Hawaii. Not 1024b, that would be a no go for sure, power issues and cross-talk, too many traces on the PCB, more layers, added cost and complexity, you name it. No, that was not the way. But with the compression feature embedded it would almost be like it was that wide.
They're all derivatives of one another really, Tahiti, Hawaii and Tonga. Tahiti lacks a few good features though.
What I was saying is that since they based Fiji on Tonga, or developed simultaneously which would be more likely, why not use Tonga instead of Hawaii? Since the design of Tonga and Fiji is the same, same people would (were?) working on both, resources would be the same, final product (300 series) would be better and actually new, better for AMD when it comes to image and marketing. OK, Hawaii is an older and possibly more stable design, but it's still "old" by today's standards, and I believe AMD didn't come out of this yet another rebrand with a good or better image, much on the contrary.
And I'm not saying Hawaii is not a good chip, it is, but it shows some weakness from AMD not to be able to innovate much at the time. I know their (little) money must be well spent, and compromises must be made.
All in all, Fiji was a good and safer bet.

koyoot, you still think or feel Fiji will be on nMP? Would it be an option to the "regular" Tonga and Hawaii based cards? Would Apple do that? Nahh
But if the signature is in El Cap, maybe we do get it in the next lineup.
I didn't forget about Tonga and Fiji. In fact, if you read carefully it is exactly what I've been saying all along. What I'd have liked to see was a super-Antigua like Fiji, instead of Hawaii, it would make more sense to me. But that's just me.
I would be like Hawaii but only updated to GCN 1.2, based on 2x Tonga XT but at only 512b mem, instead of 768b which would be too wide. 512b already is... at least borderline.

Aiden, I can't back that up of course, it was just a though. I'd imagine at least with the Alpine Ridge they'd get first silicon since they're the main supporter (almost the only one, in fact). And I'd say Intel wants Apple to keep interested in TB, although Intel seems to gain momentum again with USB. If Intel really wants to keep TB alive (and I'm not sure they do in the long run) they need to care for those who were willing to bet on it in the first place, and that was Apple. If Apple comes out with a new design, that might give it a push. If not Apple, who will battle for it? A couple of Taiwanese PC makers? Just to say they can do it too?
CPUs is a different story and surely there are other bigger players that can get samples in advance and at larger quantities.

Punk, it really was just joking, don't take it personally please. I inserted the 🙂 exactly so that you wouldn't take it to heart. My bad, sorry.
I wouldn't count on Apple going the dual CPU route again, unfortunately. I'm all with you when it comes to availability of a beast Mac, that people can customize at will, but you know Apple is not that kind of company. They're control freaks and I understand why, can even agree to some point. Imagine the support nightmare!! Not from some people here that are used to tweaking their systems, but they would open a can of worms really. Compatibility problems, people would ask for their support for the most various configurations, no way that would fit the Apple way.
That's the PC world (OK, some will argue the compatibility problem is no more, but really?).
Nice systems there by the way. I've given up making my rigs already, maybe I'm getting old...

 

[Stacc]

dec, again I said Tonga with 512b mem bus would be better than Hawaii. Not 1024b, that would be a no go for sure, power issues and cross-talk, too many traces on the PCB, more layers, added cost and complexity, you name it. No, that was not the way. But with the compression feature embedded it would almost be like it was that wide.
They're all derivatives of one another really, Tahiti, Hawaii and Tonga. Tahiti lacks a few good features though.
What I was saying is that since they based Fiji on Tonga, or developed simultaneously which would be more likely, why not use Tonga instead of Hawaii? Since the design of Tonga and Fiji is the same, same people would (were?) working on both, resources would be the same, final product (300 series) would be better and actually new, better for AMD when it comes to image and marketing. OK, Hawaii is an older and possibly more stable design, but it's still "old" by today's standards, and I believe AMD didn't come out of this yet another rebrand with a good or better image, much on the contrary.
And I'm not saying Hawaii is not a good chip, it is, but it shows some weakness from AMD not to be able to innovate much at the time. I know their (little) money must be well spent, and compromises must be made.
All in all, Fiji was a good and safer bet.

koyoot, you still think or feel Fiji will be on nMP? Would it be an option to the "regular" Tonga and Hawaii based cards? Would Apple do that? Nahh
But if the signature is in El Cap, maybe we do get it in the next lineup.
I didn't forget about Tonga and Fiji. In fact, if you read carefully it is exactly what I've been saying all along. What I'd have liked to see was a super-Antigua like Fiji, instead of Hawaii, it would make more sense to me. But that's just me.
I would be like Hawaii but only updated to GCN 1.2, based on 2x Tonga XT but at only 512b mem, instead of 768b which would be too wide. 512b already is... at least borderline.

Aiden, I can't back that up of course, it was just a though. I'd imagine at least with the Alpine Ridge they'd get first silicon since they're the main supporter (almost the only one, in fact). And I'd say Intel wants Apple to keep interested in TB, although Intel seems to gain momentum again with USB. If Intel really wants to keep TB alive (and I'm not sure they do in the long run) they need to care for those who were willing to bet on it in the first place, and that was Apple. If Apple comes out with a new design, that might give it a push. If not Apple, who will battle for it? A couple of Taiwanese PC makers? Just to say they can do it too?
CPUs is a different story and surely there are other bigger players that can get samples in advance and at larger quantities.

Punk, it really was just joking, don't take it personally please. I inserted the 🙂 exactly so that you wouldn't take it to heart. My bad, sorry.
I wouldn't count on Apple going the dual CPU route again, unfortunately. I'm all with you when it comes to availability of a beast Mac, that people can customize at will, but you know Apple is not that kind of company. They're control freaks and I understand why, can even agree to some point. Imagine the support nightmare!! Not from some people here that are used to tweaking their systems, but they would open a can of worms really. Compatibility problems, people would ask for their support for the most various configurations, no way that would fit the Apple way.
That's the PC world (OK, some will argue the compatibility problem is no more, but really?).
Nice systems there by the way. I've given up making my rigs already, maybe I'm getting old...

Unfortunately you can't just make up GPUs that AMD doesn't have. As much as you or Apple might like, its not going to happen.

Realistically, these are the possible GPUs:

Tonga, 2048 shaders, 256b, 2,4 GB VRAM
Tonga, 2048 shaders, 384b, 3, 6 GB VRAM (this is a rumored config)
Hawaii, 2816 shaders, 512b, 4, 8 GB VRAM
Fiji, 4096 shaders, 1024b, 4 GB VRAM

All of these are available in cut down versions as well (as in, reduce the number of shaders). These are AMDs GPUs and will likely not change for the next 1-1.5 years. Fiji is certainly a compelling chip, but is limited in VRAM, which is an important component to a professional GPU. I think we will see Tonga on the low end of available options, the only question is Hawaii vs Fiji on the high end. The benefits of Fiji are great, fast, power efficient, but the limited VRAM and reduced double precision compute performance may make it infeasible.

One of the great selling points of the D700 (Tahiti) was it had fantastic compute performance for a consumer card. Both AMD and Nvidia have had to cut some corners to eek out more performance from the 28nm node, which meant focusing more on gaming performance than compute performance. Unfortunately both Tonga and Fiji suffer due to this. In fact, it wouldn't even surprise me to see the D700/Tahiti reappear in the updated Mac Pro, despite being 2 generations old.

Just for reference, the current Mac Pro is:

D300: Pitcairn, 1280 shaders, 256b, 2 GB VRAM
D500: Tahiti, 1536 shaders, 384b, 3 GB VRAM
D700: Tahiti, 2040 shaders, 384b, 6 GB VRAM

 

[cerberusss]

Ahhhh, just build a Hackintosh

I'm an iOS developer and with the dead-slow Swift compiler, I really have been thinking about this. Problem is, if a problem occurs, I would always wonder whether it's the hardware.

I really like Xcode, but it isn't exactly the pinnacle of stability. Would I introduce another problem factor to consider, I'd surely get restless. And I've got team politics to worry about. Suppose I ask for a team member's help, you can imagine the reply "have you tried it on a Mac?"

 

[koyoot]

You will not get ANYTHING apart from the parts that are already available. Grenada, Antigua, Fiji. Thats all of you choices.

AMD got rid of the old GCN 1.0 cards in current lineup. There are only 1.1 and 1.2. Why? Simple: OpenCL 2.0. Also, its not that Fiji was based on Tonga. It was Tonga based on Fiji. AMD had preproduction mules of Fiji even before Tonga. Tonga was testline for Delta Color Compression. Also, there will be no other Tonga, or whatever its called right now Chips, apart from those, which OEMs bought on their exclusivity. The same situation as Tonga in iMac 5K. Will we see it in Mac Pro? I dont think so.

Everything starts to look like we will see new Mac Pro in 2H of this year.

 

[PunkNugget]

I'm an iOS developer and with the dead-slow Swift compiler, I really have been thinking about this. Problem is, if a problem occurs, I would always wonder whether it's the hardware.

I really like Xcode, but it isn't exactly the pinnacle of stability. Would I introduce another problem factor to consider, I'd surely get restless. And I've got team politics to worry about. Suppose I ask for a team member's help, you can imagine the reply "have you tried it on a Mac?"

Well with any PC build you're always going to have minor issues. The way to eliminate those issues is to continue asking lots of questions. You're best bet is to do what I did and get in touch with someone who understands programing (as you do) and discuss stability issues with the MOBO/RAM/CPU/GPU and will it run the most current Mac OS X out there? There are quite a few MOBO/RAM/CPU/GPU configs that would be easy to put together. The person to help you out with all of this, is RampageDev (RD) and his site is, well… http://www.rampagedev.com.

Me, on the other hand I like a challenge as I try to be ahead of the curve when it comes to these builds and having someone (like RD) to help along with the process makes it much easier for me to accomplish. But once it's done, then I don't have to worry about the system anymore except just using it for what I need.

I wasted months trying to figure out what Kext files and extensions and whatnot that were going to make my hackintosh work. Then I found RD and he fixed all my issues within less than 2 hours and had me up and running in no time. I've been using my Hackinbeast now for over 3 years. Now it's time for a new build.

 

[cerberusss]

There are quite a few MOBO/RAM/CPU/GPU configs that would be easy to put together. The person to help you out with all of this, is RampageDev

That is quite interesting... he offers paid support as well. Which is exactly what I'd need for a care-free experience.

 

[PunkNugget]

That is quite interesting... he offers paid support as well. Which is exactly what I'd need for a care-free experience.

Well, you will have to build it yourself and he offers the help to get Mac OS X working on that machine. Then if things go a bit wacky; like say your SSD frying on you a two to three years down the road, then he would help when it comes to the reinstallation of your OS. But like anyone else, he would need contributory funds in order for him to continue to help your needs. At least that's what I do. He never demands the funds, I just freely give because I value what he's able to do. You would have to get in touch with him to go over all those details.

To me it's far better than spending $9,899 for a top of the line MacPro that could easily put together myself (which to me is most of the fun and challenge) and spend only $3,300 (1/3 the cost) to get the same exact machine. What I like to do is spend $6,500 and get a machine that is literally twice as fast in processing and rendering power. Not only that, then you get to have bragging rights when you're killing it with your GeekBench Scores. It gives me that extra special feeling inside knowing that I have the TRUE top of the line Mac OS X machine that I built and had others there to help you with that process. There's a level of accomplishment and pride that you feel knowing that you were able to contribute toward a faster built machine that others can imitate from. Then it becomes a community thing and you're able to share ideas together and tweak what you have to make it better. But that's my take on it.

 

[ManuelGomes]

Stacc, AMD doesn't have that particular GPU, but I'd say they could have had it more or less easily if they wanted to.
They have Fiji, they made Tonga, they could have made something bigger without much additional effort.
But I'm not gonna argue over it anymore.
Tonga XT is in fact full 2048 and 384b wide, AMD released it for Apple in the form of R9 M295X in riMac.
That would be a possible solution for the nMP.
DP is certainly a major setback in newer generations of GPUs, for some professionals. And although I admit sticking with Tahiti would be possible, I find it unlikely that Apple will go for another round with it.
I guess we'll have to wait and see what the future holds.

Punk, just out of curiosity, whenever there's an OS X update, is it a peaceful installation or do you keep an old version running and don't bother to update?

 

[Stacc]

onga XT is in fact full 2048 and 384b wide, AMD released it for Apple in the form of R9 M295X in riMac.
That would be a possible solution for the nMP.

FYI, the M295X does not have a 384b memory bus, otherwise it would have 3 or 6 GB of memory, not 4 GB. There have only been rumors that part of the memory bus has been disabled, but as far as I know it isn't fully enabled in any currently shipping version of Tonga. The M295X does have the full 2048 stream processors enabled though.

 

[PunkNugget]

Punk, just out of curiosity, whenever there's an OS X update, is it a peaceful installation or do you keep an old version running and don't bother to update?

It depends. Like right now I could easily upgrade to Yosemite 10.10.+ but because the software is written to support multiple monitors (3 x 2560x1600 30" monitors), like what I currently have running, then I have to stick with what will support my 3 monitors. That is Mac OS X 10.8.5. In Apple's mind (I guess) having multiple monitors is not useful. NOT TO ME. I like my "real estate." Because everything is right there in front of me and I don't have to switch back and forth to other "pseudo" screen, like they went over on their 2015 WWDC with El Capitan. When I do that, then I tend to forget what is not in front of me and that caused me issues. That's why I went with the 3 x 30" monitor setup. After watching what El Capitan can do, it was nice, but again you should have the choice to use multiple monitors. You need at least a GTX 580 3GB GPUs with the size monitor I have that has 3 GB of RAM in each card. Now, you have two new powerhouse GPUs that recently were released:

1) 980 Ti's with 6 GB RAM / 2816 CUDA Cores ($679.00) (I purchased the SuperClocked one - actually two)
2) Titan X's with 12 GB RAM / 3072 CUDA Cores ($1000.00)

you now have a rendering powerhouse of a machine with either card that you choose to go with. And yes, they both work in Mac OS X.

Just to let you know I ALWAYS have two to three SSD backups. So I will Carbon Copy Clone (CCC) my SSD first on two separate SSDs (just to make sure that I have two backups) which is not expensive since you can get a 240GB now for $100.00 (that's cheap). Then you can do the upgrade. But before I do any upgrades I always send an email or Skype message to RD to ask if the new update is compatible with my machine. He usually knows this information within a week or two of any Mac OS X release. Hope this helps...

 

[displayator]

I was re-reading the Anandtech article about Thunderbolt 3 and I noticed that they mention that the thunderbolt 3 controller is not incompatible with Haswell-E so in theory if Apple wants to release a Mac Pro with Thunderbolt 3 they do not need to wait for Broadwell or, god forbid, Skylake Xeons.

"Intel is making it clear that at a technical level Skylake and Thunderbolt 3 are not interconnected, and that it would be possible to pair Alpine Ridge Thunderbolt 3 controllers with other devices, be it Broadwell, Haswell-E, or other products."

As for the suitable Xeon chips, the Xeon (v2) in the current nMP are 1600 series on the 4,6 & 8 cores and 2600 series on the 12 cores. According to the Intel Ark and the latest intel roadmap leaks, the Haswell (v3) of the 1600, 2600 and even 4600 series are already "launched" since either Q3 2014 (1600 and 2600) or Q2 2015 (4600).

From the leaked roadmap, Broadwell(v4) seems slated for launch (1600 & 2600 series) beginning of 2016. Skylake (v5) Xeon are planned for what looks like 2'nd half of 2017 so I really hope they are not waiting for that. For the Intel roadmap I hope we'll know more at the August's 18-20 Intel Developer Forum.

My hope for the release of a new "new Mac Pro" is Haswell(v3) & TB3 with El Capitan (Metal on OSX, with Adobe After Effect demo at 8x improvement tells me they might be targeting the Pro market again).

But Metal on OSX might also allow Apple to say: "look we just nX'd the performance of your existing hardware, you can now wait a little more for your next nMP", so obviously the following window of opportunity is next year with Broadwell, but Intel is never early, and Apple didn't launch the nMP as soon as the v2 were available either, so it would likely be a while between the Intel release and the Apple release (and a super very long while between the nMP 1.0 and the nMP 1.1/2.0).

I have no idea about what they might do about the PCIe lanes / thunderbolt 3 / GPU part connundrum. It is however likely to be pragmatic and unsurprising. Or maybe they'll shorten the Trash can form factor and only have an "ash tray" with two CPUs and some RAM, on 3 sides (CPU CPU RAM), and only have external GPUs connected over Thunderbolt 3 like the spec allows???? Does anyone knows if two CPU means more PCI lanes available?

If we're dreaming they might go CPU, GPU and a XEON phi instead of the headless GPU (though I don't know who or what takes advantage of the phi).