MP 7,1 Waiting for Mac Pro 7,1 A1991 (no more)

[deconstruct60]

But why?

Honestly that where the "solution in search of a problem" aspect is with throwing extremely customized iPhone (or even iPad Pro ) SoCs at Macs. That is a bunch of 'could do that", but the 'why' is quite thin.

Making a Mac specific ARM SoC has a huge 'why' economically. Almost zero justification there.

There has been some movement around T.2 taking some load off the CPU and doing co-processing.

Not really much movement at all. Several mainstream SSD Flash controllers have been able to do self encryption for more than several years. All Apple did with enabling offload, self-encryption drives is to gain parity with what was already available on the competitive Windows for years previous. Apple's solution is a bit more secure ( so being a bit different and late is somewhat excusable ), but this really isn't as much a movement as to "keeping up with the Joneses"

If the T2 is primarily a security co-processor you do not want to put random user apps/drivers onto the T2. As it is the T2 has "a lot" going on and relatively complicated stack for a security processor. Piling more and 'random' stuff only will open more vectors to be exploited.

Similar issue with the T1-T2 driving the touchbar ... that was more a Intel GPU output limitation and that really isn't an "offload". The T-series GPU drives the screen by copying a framebuffer from the main memory.

The T2 taking audio/video inputs is as much driven by security as it is "offloading". ( again 'random' software augments can't get at the raw audio video which makes them more secure. ). There is some workload shifting there but it tends to be on the way in ( picture/video pre-processed into usable form. HEVC or natural language recognition done or eventually FaceID. ). Doubfult this is going to mutate into more general compute offload.

Apple buying up parts of Dialog for power management. More of that may get weaved into future T-series but that processor workload isn't an x86-64 one.

But I'm not sure why you'd farm a bunch of work out to little ARM CPUs when you have a big honkin screaming fast Xeon CPU right there. Even if each little ARM board was as fast as a few of the Xeon cores, a single Xeon CPUs these days still has 20-30 cores alone.

If they were magically hooked up into a NUMA, flat memory space. You'd have something similar to what Intel tried with their "x86 GPU" and Phi series. But if that is a completely different memory space you run into same issue as discrete GPUs run into.

However, 4-6 iPhone SoC slapped together on a single card... that's largely a waste of time. Each of those with their own memory is more like a cluster of Raspberry Pi's than being anywhere near competitive with any of the GPUs for GPGPU kinds of workloads. Almost everyone's GPUs beat ARM if it comes down to just doing "embarrassingly parallel math" at high perf/w workloads. Apple putting an empty x16 PCi-e v3 slot in the next Mac Pro would do far more to enable that path than any hand waving folks are doing around stuff multiple iPhone SoC into a Mac Pro.

Such a setup would make sense on a low end Intel box where you needed ARM CPUs to beef up a slow Intel CPU. But on a Mac Pro why would you want to bypass the speed of the Xeon for much slower and smaller ARM chips?

it doesn't even make sense there if they are all independent SoC without a shared, flat memory space.


Pushing every expanding workloads down to the T-series makes very little sense. It is at odds with the core objectives and it probably wouldn't buy a whole lot even if tried. Even if Apple switched to an iPad Pro SoC still leaving the t-series decoupled and more secure would give higher security if going to enable external drives (and normal "laptop" like multiple boot options ).

 

[AidenShaw]

And the benefit for end users would be?

Having to buy new versions of all of your softwares?

 

[deconstruct60]

Yeah, that is also true, since Intel is currently stuck at 14nm. We will probably see the better picture when Intel 10nm shows up on holiday next year.

There are now rumbling of a "Comet Lake" release coming in 2019 for U-series and S-series (at least). Well see if that's their 'started over with new design rules for 10nm" solution or yet another iteration of 14nm to fill the gap until 2020. I suspect it is the latter. Intel make squeak out some Y-series 10nm solutions next holiday (and some ASICs and some maybe some cell modems) , but I wouldn't bet on them doing most of there line up.

But it showed Apple's variation of ARM architecture has lot higher performance ceiling than previously thought. It will be difficult to expect another large scale increase in IPC, but if they manage to scale up to higher power target and achieve decent increase in clock speed, we will definitely see the switch on Macbook line first.

I don't think there is much rocket science there. Apple cranked up the cache sizes and did some stuff to mitigate some memory latency.

IPC is limited by just having 'von Neuman' style code. There is only so far you can go in code before run into a branch or unfinished results. When start grossly cheating around those inherent limitations then get stuff like Meltdown and Spectre problems. Apple ARM isn't going to magically jump far out in front of Intel/AMD because ARM code has much of the same encoded cap on IPC as every other design. Apple could start doing more threads, but single threads have a cap were there isn't tons of slop there to clean up past most modern implementations.

Well, I wouldn't worry too much about it right now since it is 2 years away at earliest.

Apple could take the MacBook into an iBook within 2 years if they wanted. Apple could push the iOS devices into the a similar form factor space as the Macs. Apple already has a highly successful OS on ARM. macOS is needed there why? ( "macOS will get move 'love' from Apple if it is on ARM" isn't really a good answer. It is more self serving from mac 'fanboy" perspective than strategically salient to Apple's future. )

Splitting the relatively small Mac line up into a ARM 'half' and x86_64 'half' is a dual edge sword. There are downsides ( development complexity) as well as upsides ( maybe longer battery life ... if not farted away by thinness anorexic disorder.).

Apple is at least 2 years from parity ( in functionality as well as speeds on drag racing single thread benchmarks) in the lower half of the Mac laptop space. But they are more than 2 years away from the desktop space ( especially if weave in AMD options in 2 years. The likelihood that both Intel and AMD are both screwing up at the same time is relatively low. )

 

[Biped]

IPC is limited by just having 'von Neuman' style code. There is only so far you can go in code before run into a branch or unfinished results. When start grossly cheating around those inherent limitations then get stuff like Meltdown and Spectre problems. Apple ARM isn't going to magically jump far out in front of Intel/AMD because ARM code has much of the same encoded cap on IPC as every other design. Apple could start doing more threads, but single threads have a cap were there isn't tons of slop there to clean up past most modern implementations.

Isn't the argument for risc style arch in some (small) part, that with the x86 you have instruction set bloat, with some sets bordering on vestigial, but some functions in the older sets actually being more performant than same newer implementations. Result being that to really optimize, you would want to fully profile before compiling against every sufficiently different platform ? Whereas with risc you are already far more ahead in the optimization game ?

 

[Slash-2CPU]

Isn't the argument for risc style arch in some (small) part, that with the x86 you have instruction set bloat, with some sets bordering on vestigial, but some functions in the older sets actually being more performant than same newer implementations. Result being that to really optimize, you would want to fully profile before compiling against every sufficiently different platform ? Whereas with risc you are already far more ahead in the optimization game ?

That argument maybe made sense in the 586/Pentium Pro vs PowerPC era. Now, x86 processors are basically RISC under the hood with microcode translating complex ops into smaller ones. For that matter, x86 using bigger, simply-encoded instructions is generally better with memory and cache bandwidth when moving those instructions around.

 

[Biped]

That argument maybe made sense in the 586/Pentium Pro vs PowerPC era. Now, x86 processors are basically RISC under the hood with microcode translating complex ops into smaller ones. For that matter, x86 using bigger, simply-encoded instructions is generally better with memory and cache bandwidth when moving those instructions around.

Fair enough, I was reading something recently regarding sse / avx translation penalties during certain vector code ops, that translation process involved saturating the CPU with load/unload memory ops to the detriment of being marginally better than using linear code. I'll have to go dig through the browser history to check the datestamp on it.

 

[0388631]

The x86 vs ARM should be put to a rest for now and focus on what the new Pro will offer. A processor is only so good until the motherboard and what peripherals can be connected to limit the versatility of the machine. That said, I'm going with a slightly downclocked Apple specific Xeon processor. I predict the format will be much like the old cheese grater but more modern. If Apple is serious about offering a pro machine, they'll build it with the price to match.

 

[Slash-2CPU]

If Apple wants to be courageous, make it all CPU-direct PCIe. Cascade Lake-W should have 44-48 lanes. First 12 lanes to Thunderbolt, USB 3.1 Gen2, and 10Gb Ethernet.

A box that accepts 3 or 4 PCIe x16 or x8 modules.

Higher-option x16 module with a x16 to x32 switch to 4 PCIe slots (x16/x0/x4/x4 or (x8/x8/x4/x4).

Higher-option x16 module with a x16 to x32 switch lanes that hosts 8 U.2 SSD’s with available U.2 to m.2 adapter.

x16 module no switch 1 double-width and 2 single-width PCIe slots at x16/x0/x0 or x8/x8x0 or x8/x4/x4.

x16 module with no switch that hosts a single x4 PCIe slot, 2 m.2 SSD’s and, a SATA chip with 4 drive bays and eSATA.

 

[filmak]

Having to buy new versions of all of your softwares?

True.
If they want to build something more complex there is no need to add ARM into the mix, just put a second xeon inside or leave it as an option and it would be fine.
Or they could also produce a PCIe add on card with a T2 chip and several lockings inside and sell it (~1000$) so it would be possible for users to install it and run MacOs on real workstations (selected models) from Hp, Dell etc.

 

[BlueTide]

I don't follow the logic.

The low end is already on ARM. You do know this? Why does that osborne anything? I think we all know that whatever they do with a Mac Pro will be replaced in a few years. That is what this whole thread is about: What will replace the current Mac Pro. ( Or maybe, WHEN will they replace the current Mac Pro.) If people need a Mac Pro, they will buy one and hope Apple replaces it with a better version in a few years, at least LESS THAN 5 years.

Yeah. I'm aware. I hope you're aware that people need to run their (assumedly "pro") software on these things to make them practical. And where we are today with many people already having jumped the ship already if they need to get more demanding work done since current Mac Pro ain't it. If they've suffered this long waiting for a new one and then there is word out that it's going to be a pricy stop-gap for a year or two and then you're sol... Oh my. LOL.
[doublepost=1543770815][/doublepost]

maintaining the slimmest hope that "completely rethinking the mac pro" means a design that is the most slotty of slotboxes, and Apple makes a pci card similar to one that takes 4 m2 ssds, but instad of ssds, each is an ARM processing unit, something like an iPhone chipset, minus the parts that are needed to run a phone. That way you get a xeon workstation with minimum disruption, plus arm processing (if it's all its cracked up to be for pro tasks).

Hmm, that is an interesting hypothetical. What if you could start with Xeon and down the line replace those with ARM due to slots that allow you to do this? Haven't thought of that before. In theory one could market it like "Hey, we know this is an investment, we have ARM coming, but now the software needs Xeons to run. So start with them and upgrade when your apps catch up. Or, if you can use ARM already with your apps, here's our first generation and it really screams with apps X, Y and Z."

I'm sure there are massive complications to that, but... seems like an interesting thought exercise.

 

[pl1984]

LT as himself yet again...
[doublepost=1543519099][/doublepost]I also don't think ARM will be an option in the short run.
I would like to see x86 gone though, for a long time now. Mentioned it before here, and Aiden was kind enough to let the dogs loose on me at the time
The thing that used to be a value add in x86 (OK, x64) - legacy - has been, in my opinion, a major slow down.
Apple had the b@lls to switch and it went (insert something here) well.
Too bad Itanium didn't make it, maybe a desktop variant would be wildly used today.
Maybe if Intel would start with a clean slate (well, it never really is clean) and go RISC this time around. Gone are the glorious days of RISC workstations...
Or AMD start with something new.
Although starting a new platform is always a daunting task, I believe it's now easier than ever. And it's about time.
Patching up forever decades old legacy stuff seems odd these days.
Maybe Apple is secretly developing something brand new, not even ARM based. Probably not though, but they would be the ones with the means to do it.
[doublepost=1543519206][/doublepost]On the 11th we'll see how Intel hopes to keep things afloat with CCL-AP and maybe a hint at Artic Sound.

A major slow down in what way?

 

[askunk]

Please go to 3:00 and enjoy

 

[rrl]

Uh, no joy.

 

[Varmann]

Please go to 3:00 and enjoy

Excellent design.... for repairs.
Finding usable 3rd party parts were often futile though.

 

[filmak]

Please go to 3:00 and enjoy

Thank you.
Nice and exciting years. Modular computers.

 

[Simon R.]

ARM would be stupid. All applications need to be re-written. If they came out with an ARM-based Mac Pro it would be almost software-less for years. And many programs might not even get a port. C'mon even Apple can't be that stupid.

It is however ridiculous that almost two years after they began work on a new Mac Pro, they still have nothing to show. If they had anything almost ready to go, they would have shown it by now. This shows that very little resources are put into this. I am losing faith.

 

[ManuelGomes]

Hi filmak. Sorry, away for the weekend. Couldn't order it these days but will do tonight. I'll get the 11" as well, 13" is too big. How do you feel about yours? Like it much?
Would I get a modular Mac Pro if I could network them both? Joke of course
I'm getting the keyboard as well, should also get some hard (-ish) case for them.
[doublepost=1543833758][/doublepost]pl, I just think that patching up, or layering upon something that's 4 decades old diverges resources that could be put into development of something new, better.
My 2cents anyway.

 

[Ifti]

I am expecting this years CPUs in next years 'new' Mac Pro.
Apple are never bleeding edge any more.

 

[filmak]

Hi filmak. Sorry, away for the weekend. Couldn't order it these days but will do tonight. I'll get the 11" as well, 13" is too big. How do you feel about yours? Like it much?
Would I get a modular Mac Pro if I could network them both? Joke of course
I'm getting the keyboard as well, should also get some hard (-ish) case for them.
[doublepost=1543833758][/doublepost]pl, I just think that patching up, or layering upon something that's 4 decades old diverges resources that could be put into development of something new, better.
My 2cents anyway.

Yes, I like it very much, coming from an iPad 4, there is much of a difference, it is very fast and the screen is big enough.
The upgrade is very noticeable in my case.
I did not buy the keyboard, I like the tablet to be autonomous, (but I got the Apple Pencil 2 ) so, I got this nice leather case from Stilgut.
https://www.amazon.de/gp/product/B07KBSD7HL/ref=oh_aui_detailpage_o00_s01?ie=UTF8&psc=1
Hope you will like the new iPad Pro too.

 

[barmann]

Having to buy new versions of all of your softwares?

Oh no, like post Snow Leo OSX updates ?
Say it ain't so !!
[doublepost=1543848300][/doublepost]

If Apple wants to be courageous, .....

You lost me after the C word .
The last thing Apple should do about the next MP is getting courageous, creative , innovative or even think .
It'd only mean fewer options, fewer ports, and using your thumbs to operate the thing .

 

[ManuelGomes]

Good to know filmak, thanx.

 

[fuchsdh]

I am expecting this years CPUs in next years 'new' Mac Pro.
Apple are never bleeding edge any more.

With the next -W CPUs only a few months away at this point I highly doubt it.

 

[askunk]

So let's get to pricing a bit:

actual tMP are at 2,999 and 3,999, while the iMP is at 4,999.

Let's shape off the 900-1,000$ that takes for the screen. I don't think Apple would sell the mMP at less than 3,999 for a 32GB/1TB base model, with very likely a Vega 56 with 8GB HBM2. I'm shooting in the dark... the mid will get 32GB/1TB (4,499$) with a Vega 64 16GB HMB2 and the top will get either a soon to be available (by Nov '19 it should be close) Vega 2, an Nvidia or a 2 x Vega 64 configuration (5,999), 64GB/1TB.

Whether they would be on PCI-e, with a proprietary slot on a daughter card, soldered (please, no!), it's very likely that nVidia won't be part of the Apple offering. We have talked a lot about this and the only chance would be getting a PCI-e slot (and drivers) or NVidia miraculously building a card on Apple's standards.

Maybe they will come out with a cheaper base offer with a 500 GB Flash drive. I doubt it: they will rely on RAID0 on two drives for each configuration.

We seem all to agree that it is quite unlikely that the mMP will come out with new video cards since there seem to be no new model coming till H2 2019.

Selling a mMP at 3.499 with similar RAM, GPU and Flash drive would be possible only with a contextual reduction of the list price of the iMP, which I think is even more unlikely.

USPs? The T3 (I hope), modularity (let's hope it's PCI-e), FaceID on the new TBD and some freaky design that allows it to run silent with just a little thermal throttle

 

[Joe The Dragon]

1TB flash base (boot drive) is way to much for an multi slot system and no need for raid 0. It's not an ALL in one imac pro or an thin laptop.

Pro need storage choice and maybe an systems with an lower end video card (not all pro workloads need an high end video card)

 

[AidenShaw]

they will rely on RAID0 on two drives for each configuration.

and no need for raid 0

If you're calling a T2 chip with two flash daughtercards "RAID 0" you're not even half right.

Virtually all high performance SSDs (and even thumb drives) get added bandwidth from doing parallel reads and writes to multiple banks of flash. You can say that this is somewhat analogous to how RAID-0 works, but it is not RAID-0.

If you disagree - look up what the "I" in "RAID" means.