Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP 7,1 Waiting for Mac Pro 7,1 A1991 (no more)
- Thread starter Mago
- Start date
-
- Tags
- macpro 2016
- Sort by reaction score
Perhaps Apple's anemic share of the growing workstation market wouldn't be anemic if they, like, offered a range of workstations?
And did more than, at best, buggy drive by updates to OSX?
The coming Mac Pro is Apple's last chance to pull out of the self-inflicted nose dive into total irrelevancy in the pro content creation market.
Apple doesn't have to compete with the entirety of Dell and HP line up. There are 7 Mac product models. HP and Dell have at least 5-6x product models as Apple does. Which one of those two groups has the business with a better return on investment? Apple has never been about making everything for everybody. Even less so after the return on Jobs.
Apple holds less than 10% of the "classic PC" market with the Market. They don't have to have max volume size. Having 8% is fine as long as that 8% is a healthy, growing subset of the overall market. Bypassing the flat stagnant (or declining) parts of the 90+% only helps Apple; not hurt it.
Leakage where. If go back and look at what Apple talked about in their April "pow wow" a substantive part of the "leakage" has been into other parts of the Mac line up. That Mac Pro is a small single digit portion of their pro market sales.
the iMac Pro will cover a healthy fraction of the current Mac Pro users ( which wasn't). The revised Mac Pro can cover most of what the current Mac Pro doesn't cover and a substantive chunk of the old 2009-2012 era hold outs. Does Apple have to get all of them. No? There is constant inflow from Windows as much as there is outflow to Windows. With less than 10% of the market that swapping is going to be a constant.
There are not higher caches. The caches are coupled to the core counts ( in the odd corner cases where cache looks higher the core is there ... just suppressed.). Lower memory and PCI-e lanes are easy to do ( just turn them off even if function correctly. Not selling defect free wide lane counts. more sell selling access to the feature. ).
PCI-e lane count typically doesn't count as much in the "hot rod" desktop market because the workloads are PCI-e crippled GPU cards anyway with software optimized to tapdance around that limitation. The "more upscale" compete is against the desktop CPU which are capped at x16. x24 is still better than x16. Lower Memory is just fine. ( number of folks past 128GB is relative few with respect to overall market) and again hot rodding the memory clock is the focus as long as behavior not too flakey (not a focus on better data integrity) .
For the most part every time Intel kneecaps functionality they cut the price. Price sensitive folks will take the trade off while folks who need functionality will generally pay. More than a few of these "extreme" processors are going to go into custom builds where the run rate isn't 10-12 1,000 CPU trays per customer per quarter. Those folks are largely buying one-offs (through a retailer to get some volume discount back , but nothing direct from Intel. ). In short, the folks who comprise the market being sold into are different.
for most part this are not primarily binned on failure. There are very likely 4-6 core Intel W parts with all 10 cores that could work just fine. These are binned to fit demand also.
Generally defects occur randomly across the wafer. The overall manufacturing is geared toward producing wafers with very few defects ( intel isn't selling finished wafers to customers ), but "stuff happens". at 14nm can be just several extra atoms off or just one of the few dust particles gets in the light lightgraphy beam or the doping element is only 99.99% pure instead of 99.999% and something doesn't come out 100% perfect. There is not plan to "bake" broken wafers with broken in special way dies.
Multiprocessor systems, arent workstations are just Workload servers borrowed from production to R&D, I do believe Apple will mate the Mac Pro (sp, unlikely more than 32 Cores), with some Server-Farm service application so those requiring mamoth renderings just setup their linux cluster (either own, or on the cloud), most Mac 3-D rendering tools had this from years, I don't know about AI but its the same scenario, VR/AR do not need MP systems for R&D, just the best available GPU setup on the market.
Developers actually do not need Multi Processors Workstations unless they use their workstations as servers instead to hire servers on demand (the way the industry works, at least on 3D rendering, AI, and finite element analysis)
Developers actually do not need Multi Processors Workstations unless they use their workstations as servers instead to hire servers on demand (the way the industry works, at least on 3D rendering, AI, and finite element analysis)
Hopefully Apple realizes that it isn’t as simple as this. Keep sanding off what you perceive to be edge case users and you’ll eventually be left with nothing.
When I stopped buying Mac Pros around 2010ish, I also stopped buying MacBook Pros. Why deal with 2 platforms? A lot of people in my line of work have been buying Dell XPS and 5520 laptops - a first non Apple laptop for them - since they’ve switched their desktop workstation to windows. Some have even gone to Android for their phones although I personally haven’t gone that far yet.
Uh no.
(For the record, based on Apple's comments, I think DP is possible. Not guaranteed, but Apple seems to understand that the Mac Pro needs to hit edge cases. Especially if they make it with a modular daughterboard like the cMP.)
I believe it is folks are more afraid of the price premium of SP processors should their workload really only benefit from a single (the 18-core Xeon Gold with a slightly higher base clock commands a $1000 over the 18-core W series).
On the flip side, for someone like yourself whose software benefits from all the CPUs and cores you can throw at it, that price premium is a non-issue compared to the performance boost offered.
Wow - that's just so far off the mark for what we do that it is laughable. It's an acronym salad (using AI and VR/AR together?).
We do ML (branch of AI), and our production (and some R&D) systems are 72C/144T with 1TiB to 3TiB of RAM, and quad GTX 1080Ti GPUs. Getting an order together for twenty Titan V cards to upgrade. Other development systems are 36C/72T with 1TiB and quad Titan X (Pascal and Maxwell) cards.
Some parts of the workflow need CPU cores, some parts need CUDA cores. If you don't have lots of both, Amdahl's Law will smack you in the butt.
The dual socket and quad socket systems are needed to get the necessary number of cores, DIMM sockets, and PCIe lanes. If we don't need full core count, we'll still buy dual socket systems with fewer cores to get the DIMMs and PCIe lanes.
And, on a semantic note, a single Core 2 Duo is a "Multi-Processor" or "MP" by most definitions. "Multi-Socket" unambiguously refers to a system with two or more sockets.
We give our engineers whatever they want. Some are happy to use our lab systems, some like the public cloud, and some like it under their desk. Some even try to use eGPU setups - but without much success. (Not due to the eGPU per se, but because code that's been tested on an Apple laptop with an eGPU often takes a lot of work to port to a big Linux GPU server.)
The typical case is someone with a laptop on their desk, remoted into a 24 to 36 core 1 TiB dedicated server in the lab. Their "workstation" is in the lab, and the laptop is the "thin client". Wherever they work, they use the remote connection. (Yes, this means that "typical" is someone with a MacBook Pro using SSH to their dedicated $50K server.)
Last edited:
I think everybody see what they want, actually Apple just named moving from dual low power GPU to single high power GPU, and modular macintosh pro, both very open to what you want to see, whatever It comes 2 year later.
Aghhh
I believe most people do ML with python, so porting from a MBP to a Linux cluster should be trivial (as on most cases I know about python+GPU offloading, of course I'm not familiar with ML)
I think the mMP will be focused on AR/VR than ML, and most ML development will continue on this kind of remote setup, whatever ML is moving very fast, maybe even the latest Volta GPU will be obsolete in 18 months on the upcoming ML optimized Tensor Processors.
I agree completely and totally with your profound assessment here.
You don't program, do you?
Volta is an "ML optimized Tensor Processor".
But you're right, in that today's "ML optimized Tensor Processors" will be eclipsed by ones from 18 months in the future. And that today's GPUs eclipsed GPUs from 18 months ago. And that in 36 months GPUs will eclipse those coming in 18 months. "Lather. Rinse. Repeat.".
Which is why the modular Mac Pro should support off-the-shelf PCIe GPUs, and not proprietary form factor modules that only come from Apple. (See the .sig)
Notice how I cleverly brought this back on-topic?
Last edited:
I use run Pypi/SciPy on Jupyter and offloading to GPUs (my own C++ GPU Kernels), I'm very careful to code platform independent but the Jupyter part its transparent on Mac/Linux, my kernels just include the required pre-processors as the SDK reference point out, so I rarely touch my code on local (iMac 5K+GTX1070 as eGPU) or the server (Ryzen 1700 + Pascal GP100)
My apologies.
In our environment, we spend a lot of time fighting version skew issues. (The Apple updated to Python x.y.z, and the Linux systems are running x.y.z-1. Or vice-versa.)
And all hell breaks loose when it's x.y.z vs. x.y-1.z.
I still am very confused by those who use the obviously useless datapoint of “how many nMPs does Apple sell here in 2018” as proof not only that they “can’t” go after the markets they used to with the cMP but that they “shouldn’t” - that’s absolutely nonsensical.
Of course they aren’t selling much in the workstation/pro market....they essentially don’t have a viable product to offer currently, apart from the iMac Pro. Does that mean that they should just give up?
Also, I believe I saw the number earlier of 25k nMPs per quarter, esitamated? Even in its current massively outdated state? That’s actually quite impressive. Now imagine the market share they’d have if they actually had a full-width Mac Pro product line that used current-gen tech and gave a broad base of “edge case” users what they needed.
Again...it seems like the hobbyists are accidentally confusing what they do with what many use high-end workstations for.
Of course they aren’t selling much in the workstation/pro market....they essentially don’t have a viable product to offer currently, apart from the iMac Pro. Does that mean that they should just give up?
Also, I believe I saw the number earlier of 25k nMPs per quarter, esitamated? Even in its current massively outdated state? That’s actually quite impressive. Now imagine the market share they’d have if they actually had a full-width Mac Pro product line that used current-gen tech and gave a broad base of “edge case” users what they needed.
Again...it seems like the hobbyists are accidentally confusing what they do with what many use high-end workstations for.
Ahh, version hell, shared headache, takes time to master a 'version flexible' code, it's my approach, unless its something really good and useful I avoid to adopt the latest stuff, unless its an patch or bug turnaround, that's another big problem to master (and get more silver, not in the bank but hair).
I rely on Anaconda, this is its big strength, only issue pending are the GPU drivers, macOS drivers use to be the most primitives, as Windows use to implement new stuff earlier.
o/
Last edited:
And it seems high-end users are confusing themselves for customers Apple has ever catered to. Again, you could never buy a high-end Mac Pro even in its cheesegrater form. There's no real evidence they've decided that's a market they're actually interested in pursuing.
The idea that "everyone who has lower system requirements than me is a filthy casual" is dumb, insulting, and makes you look like a tool of the highest order.
I don't know why you keep repeating this. From 2007-2010ish the cheesegrater was at or near the top end of what one could buy in a dual-socket workstation from anyone. That's why i bought them. That's why a lot of people bought them. They were even reasonably priced.
It didn't have the most slots or expansion, but then again GPU was not as big a deal back in those days except perhaps for gaming, and nobody buys a mac if gaming is your focus.
Because ‘xmac’ or whatever they want out of Apple. Ironically it makes the whole ‘argument’ against hi end Mac Pros a nonstarter because Apple never made an ‘xmac’, never mind hi end.
Mac Pros never ran with EX processors.
Both camps are ignorant to history. I'm not in either one.
Neither did dual socket workstations from any other vendor. EX was exclusively for server use and only supported in Server OS's.
Your point?
Last edited:
That the -SP line has a lot more in common with the old -EX line than it does with the -EP given how expensive it is to get decent clocks for your cores. You disregard the importance of a GPU in how conceptions of 'power' have changed since the last tower Mac Pro, but also disregard that Intel has changed the lines too where dual sockets come with more compromises.
Again: in a magical world, if you wanted X from Apple, Apple would produce it, but we don't live in a magical world. I and other here are responding to the reality that Apple isn't going to make two radically different SKUs in all likelihood, so a 1S config makes more sense given what we know.
The one who seems to be ignorant of history is you. Apple used to provide hi end workstations that were competitive with offerings from other vendors if not downright superior.
Your signature shows a Mac Pro 5,1 , 6 core variety. Is that a dual processor one ? Or a single one ?
Also if I may ask, was this purchased from Apple around the time Apple was selling these systems or was it bought from someone who was looking to sell it ( say a couple of years after usage )?
i guess we'll have to wait and see. I know Apple has a hard time attracting and retaining engineering talent these days so perhaps it's harder than one would think for a company of their size to keep up with the Dell's and HP's of the world. They are after all a phone company first and foremost.
I don't think the size of the engineering department has anything to do with it, it's much more as mentioned earlier the deliberate desire to keep Apple's team sizes small and keep people involved in a bunch of projects. As far as we know there's no dedicated pro applications or pro hardware teams.
What's your point with this line of questioning?
My point is that none of your four systems go beyond the six core mark ( and if you purchased the 5,1 second hand, likely the tcMP with its higher core count and dual GPUs was also not something you need or wanted ) nor the current iMac Pro ( presumably )
Perhaps you are looking for empty slots in a Mac Pro that you can fill with peripherals of your choice ?