Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Its funny that people on this forum start to call for more optimized software, when I was calling about this this, what... 2 years ago? 3 years ago?

What happens when you start optimize software? You can save TONS of money on hardware. Like this:

cifar10_average.png


Comparison of performance in TensorFlow between GPUs.
Volta V100 is 9000$. Vega is 1200$, and delivers 85-90% of performance of V100.

You can buy 6 of Vega GPUs for the price of single V100. Just by optimizing your software. What will have better impact on your job?

But hey. Nobody cares about money saving on this forum.

Did you read the blog postS? Do you even understand what matrix multiplication is?
The blog post is arguing, that in CERTAIN task the cheaper Pascal or Vega cards are more cost effective. There is nothing about SW optimization, nada, zip...
I'm sure you didn't get a lot of credits for linear algebra, but if you are interested and get off the forums MIT have excellent free courses online.
 
Did you read the blog postS? Do you even understand what matrix multiplication is?
The blog post is arguing, that in CERTAIN task the cheaper Pascal or Vega cards are more cost effective. There is nothing about SW optimization, nada, zip...
I'm sure you didn't get a lot of credits for linear algebra, but if you are interested and get off the forums MIT have excellent free courses online.
Did you even read the graph, you quoted? ;)

And what is difference between TF1.01, 1.3 and 1.6 other than software optimization? The reason why TF1.3 gained performance on AMD hardware is because of software optimization.

Im sure MIT has great courses online, but they do not have anything for reading comprehension that you can use.

I used that blog post to show that you can get tremendous benefits of optimizing your software, and what software optimization(which TensorFlow 1.3 had for AMD hardware) does for your money spent on hardware.

I love you guys, for your being easily offended by things which are not written anywhere.
 
Last edited:
Did you even read the graph, you quoted? ;)

And what is difference between TF1.01, 1.3 and 1.6 other than software optimization? The reason why TF1.3 gained performance on AMD hardware is because of software optimization.

Im sure MIT has great courses online, but they do not have anything for reading comprehension that you can use.

I used that blog post to show that you can get tremendous benefits of optimizing your software, and what software optimization(which TensorFlow 1.3 had for AMD hardware) does for your money spent on hardware.

I love you guys, for your being easily offended by things which are not written anywhere.
TF is library used by code writen to do certain task.

Are you a troll or you are just clueless ?
 
TF is library used by code writen to do certain task.

Are you a troll or you are just clueless ?
FFS.

Let me quote directly everything for you from that site.

AMD(TF1.0.1):
Ubuntu 16.04.3 x64
HIP-TensorFlow 1.0.1
Python 2.7
Driver: ROCm 1.7

AMD(TF1.3):
Ubuntu 16.04.4 x64
TensorFlow 1.3
Python 3.5
Driver: ROCm 1.7.137

No differences in software, eh? Zero effect on performance, eh?

Edit:
Not related to Mac Pro, and Server CPUs but it should show you what is happening inside Intel:
Ddp7YSrVQAAkVHD.jpg:large
 
Last edited:
Edit:
Not related to Mac Pro, and Server CPUs but it should show you what is happening inside Intel:
Here's the link to the more than year-old XMM7560 story....

And more recent (Jan 2018) headlines from Motley Fool:
Intel Corp. Is Now Sampling XMM 7560 LTE Modem

The tech titan seems to be on track with its next-generation cellular modem.
...
Sampling now, shipping later in 2018
Bryant said during the Jan. 9 event that Intel is sampling the XMM 7560 modem to "multiple customers today" and that products incorporating the modem will be available "later in 2018."

https://www.fool.com/investing/2018/01/16/intel-corp-is-now-sampling-xmm-7560-lte-modem.aspx
 
Here's the link to the more than year-old XMM7560 story....

And more recent (Jan 2018) headlines from Motley Fool:
https://twitter.com/TMFChipFool/status/998245933728006144

Check out the author of BOTH: the twitt, and the Motley Fool article ;). The post, in the TT is from forum TheLayoff.

As a side note, non related: https://twitter.com/rs_mux/status/998265236330369024

All of this has materialized based on Intel's promises?

Lake Crest coming 2017, is now coming 2019?
 
Last edited:
All of this has materialized based on Intel's promises?
Roadmaps are not promises, they are best guess estimates.

Missing a roadmap date is not a broken promise - it's a prediction/estimate that wasn't accurate.

Should I start to catalog all of the AMD roadmaps that missed their targets? (Either on date or performance?)

And how about:

https://seekingalpha.com/article/4099410-amd-vega-next-bulldozer

Power Consumption

At 486mm, Vega 64 should be beating the flagship 1080 TI (or at least coming close to it). Something went horribly wrong in the development of Vega. It should not eat power and yet only perform near (or slightly beating) a 1080 vanilla. It makes me think they just pumped power into it to get some degree of performance once it came back from the foundry -- the trade-off being increased noise and power use. As a business, AMD had to eventually ship something which was vastly superior to the nothing AMD was shipping for the high-end desktop market.

But I assume that you'll not want to talk about the failure that was Bulldozer.
 
Last edited:
AMD on Mac Pro 7,1 will NOT happen you make me laugh. It will be Intel Xeon for sure. My guess is 8 core on the low end for $3k, 36 on the high end around $25k decked out.

Perhaps Apple will build something like the Dan Case A4 SFX [0] but without the ribbon cable and instead a daugter card. This would allow for a single PCIe x16 port in a small case. The thermal design would involve air coming in the front and going out the two sides. On one side air blows over the CPUs, on the other side the air blows over the GPU. With Apple engineering they could fit a lot into a case that size and provide many external ports, 6 TB3/USB-C ports, 4 USB-A, dual 10G Ethernet. It would be a 2 sided horizontal design as opposed to the 3 sided vertical design of the trashcan, it would also be potentially even smaller than the trashcan.

Craig Federighi: "But I think we designed ourselves into a bit of a thermal corner, if you will. We designed a system that we thought with the kind of GPUs that at the time we thought we needed, and that we thought we could well serve with a two GPU architecture… that that was the thermal limit we needed, or the thermal capacity we needed. But workloads didn’t materialize to fit that as broadly as we hoped.

Being able to put larger single GPUs required a different system architecture and more capacity than that system was designed to accommodate." (empasis mine)

Mac Pro will come with an Intel processor along with a T3 chip.

Craig Federighi: "We’ve actually see that a little bit already. [With the] T1, I think we called it? But it brings some of the authentication, the Secure Enclave processor, for instance, out of our iPhone SoCs and makes them available on the Mac. So we see a really interesting complementary role for our silicon working with Intel. And we certainly work with Intel on our needs to deliver chips into our Mac roadmap and we see that continuing."

Further confirmation that they are building a small box:

Craig Federighi: "But we didn’t start with a shape and say ‘here’s the fastest machine we can put in that box.’ We actually started with a target for performance and came up with what I think was a very clever design of that thermal core and thermal architecture to accommodate what we thought was the right power architecture. What I think we didn’t appreciate completely at the time was how we had so tailor-designed that specific vision at the time that in the future we would find ourselves a bit boxed in to a circular shape. We were boxed by a circle... I wouldn’t say we’re trying to paint any picture right now about a shape. It could be an octagon this time [laughter]. But certainly flexibility and our flexibility to keep it current and upgraded. We need an architecture that can deliver across a wide dynamic range of performance and that we can efficiently keep it up to date with the best technologies over years."

Read in between the lines and Craig is saying the the Mac Pro will be a box with an Intel processor, an Arm coprocessor, and a single upgradable GPU.

[0] https://www.dan-cases.com/dana4.php
[1] https://techcrunch.com/2017/04/06/t...-john-ternus-on-the-state-of-apples-pro-macs/
 
  • Like
Reactions: Zarniwoop
Craig Federighi: "We’ve actually see that a little bit already. [With the] T1, I think we called it? But it brings some of the authentication, the Secure Enclave processor, for instance, out of our iPhone SoCs and makes them available on the Mac. So we see a really interesting complementary role for our silicon working with Intel. And we certainly work with Intel on our needs to deliver chips into our Mac roadmap and we see that continuing."

Thanks for the quote and the link. There's a lot of extra information in the transcript that the media didn't cover. According to the transcipt, Apple started to design the next Mac Pro on Intel, and it seems to be an obvious choice. Why would they say so if they were designing an AMD custom APU for Mac Pro.

If AMD finds its way to a Mac one day, it wont start with Mac Pro 2019.

And it is pretty logical too, if you think about the designing costs. If Apple designs an APU with AMD, they need to make a lot of copies in order it to become economical. So, if ever AMD custom APU arrives to Apple, it goes to a mainstream machine: iMac, Mini and laptops.

Still, if you read between the lines, Apple is not designing Mac Pro 2019 to make a lot of profit from the hardware itself, but to keep the ecosystem alive where the money comes.
 
Last edited:
Read in between the lines and Craig is saying the the Mac Pro will be a box with an Intel processor, an Arm coprocessor, and a single upgradable GPU.
I somewhat agree to that observation and also wish that to be the case. Something akin to a micro-ITX motherboard inside a mini-tower case, with potential room for drives but not a 2nd PCIe card slot.

The question is then why that design would take years for company as resourceful as Apple to realize. They seem to still have the intention to shove some "clever" form-factor that's not yet seen in the PC case world.
 
I somewhat agree to that observation and also wish that to be the case. Something akin to a micro-ITX motherboard inside a mini-tower case, with potential room for drives but not a 2nd PCIe card slot.

The question is then why that design would take years for company as resourceful as Apple to realize. They seem to still have the intention to shove some "clever" form-factor that's not yet seen in the PC case world.
It could be that Apple's' ARM T[x] roadmap was not planned for a modular system. So, they had to re-think about it. And it takes some time to either re-tool a chip and/or write a new software stack for it.
 
Roadmaps are not promises, they are best guess estimates.

Missing a roadmap date is not a broken promise - it's a prediction/estimate that wasn't accurate.

Should I start to catalog all of the AMD roadmaps that missed their targets? (Either on date or performance?)

And how about:



But I assume that you'll not want to talk about the failure that was Bulldozer.
The roadmaps, and therefore products slipped because Intel cannot put manufacturing right. Is it this hard to accept?

And you dared to call me a fanboy, some time ago, when you are textbook example of fanboy.

Your post is complete flamebait, for those who care about company vs company battle. You want to spin this into company vs company. Im not gonna fall into this, my friend.

Your post is bannable on 90% of forums, because this is trolling. Just so you know.

Who cares about Bulldozer, when all that matters in this industry is future? Intel cannot manufacture any of products on 10 nm process for upcoming 1.5 years. Everybody will be on 7 nm by then, for 1 year over.
 
Something akin to a micro-ITX motherboard inside a mini-tower case, with potential room for drives but not a 2nd PCIe card slot.

Phanteks Evolv Shift

mini ITX, compact footprint, single pci3 x16 slot, room for AIO water cooling for both GPU and CPU, and 4 x 2.5" drive mounts.

And frankly, better looking than the trashcan was, imho (which in practice was grubby silver with an oily fingerprint patina).
 
It could be that Apple's' ARM T[x] roadmap was not planned for a modular system. So, they had to re-think about it. And it takes some time to either re-tool a chip and/or write a new software stack for it.
The problem is timing; if the change of mind happened right before 2017 April roundtable, then a trashcan update should have come before that (as incremental as it may have been). If the iMac Pro was conceived as a solution, whether it be considered stop-gap or permanent, its form factor obviously doesn't address most MP users. It also didn't help that the Mini was stagnated along as well so the lower end of the MP spectrum is also under-served. If Apple didn't purposely engineer this ridiculously long turnover, then their collective competence is in serious question.
[doublepost=1526911309][/doublepost]
Phanteks Evolv Shift

mini ITX, compact footprint, single pci3 x16 slot, room for AIO water cooling for both GPU and CPU, and 4 x 2.5" drive mounts.

And frankly, better looking than the trashcan was, imho (which in practice was grubby silver with an oily fingerprint patina).
This does look like something a large section of this sub-forum will buy. Of course it doesn't address the 2+ GPU bandwidth demands for some use cases, but at least having one open slot with the card of your choice, and reasonable number of internal drive slots that cover most creative workflows. The remaining question is if any outbound PCI lane not in the form of Thunderbolt3 is possible, if not then for some users this is just a trashcan that sucks slightly less.
 
  • Like
Reactions: Manzanito
I somewhat agree to that observation and also wish that to be the case. Something akin to a micro-ITX motherboard inside a mini-tower case, with potential room for drives but not a 2nd PCIe card slot.

The question is then why that design would take years for company as resourceful as Apple to realize. They seem to still have the intention to shove some "clever" form-factor that's not yet seen in the PC case world.

Apple has to build a case, and not just any case a Johnny Ive approved case, a motherboard, a daughter card. They may be using a yet unreleased Xeon from Intel. There is a lot of engineering involved fitting all that power into a tiny case. All of that takes time. They can't update the cheesegrater without doing most of the above work. They can't or wont upgrade the cylinder Mac Pro due to "thermal limitations." iMac Pro was not a stopgap it was Plan A, Mac Pro is Plan B. Mac Mini has gone 4 years with barely a mention -- it is dead. We are getting a hard look at how long it takes for Apple to build a new product from concept to production starting April 2017 and ending sometime in 2019.
 
Apple has to build a case, and not just any case a Johnny Ive approved case, a motherboard, a daughter card. They may be using a yet unreleased Xeon from Intel. There is a lot of engineering involved fitting all that power into a tiny case. All of that takes time. They can't update the cheesegrater without doing most of the above work. They can't or wont upgrade the cylinder Mac Pro due to "thermal limitations." iMac Pro was not a stopgap it was Plan A, Mac Pro is Plan B. Mac Mini has gone 4 years with barely a mention -- it is dead. We are getting a hard look at how long it takes for Apple to build a new product from concept to production starting April 2017 and ending sometime in 2019.
Why does it have to be tiny?
 
Not trying to start a flame War here, but I cannot sugar coat it either. Apple has so undercut it's Professional users by not offering a modern customizable computer that has upgrade and expandability options like the competition on the WinPC side.

Here's an article that is basically a case study in the drip of Apple losing it's Pro users to Windows setups.

https://www.thepixellab.net/mac-vs-pc-as-a-motion-designer-the-end-of-an-era

For those that would scream, "make a Hackintosh", I'd say it's dicey to base my business work on a computer that could be bricked with an Apple update that my Hakintosh doesn't support or requires going to forums to constantly read up on how to handle updates. I don't need additional unbillable work!

I'm pissed. Apple dropped the Mac Pro market with the trash can Mac that didn't offer any expandability or customization and cost way more than comparable WinPC's. I do 3D, video (with compositing and motion graphics) and print advertising so I'm a power user. Macs just seem to be behind the curve for pro users wanting desktops that are modern and expandable and customizable.

Please in response, don't flame but offer tangible information why you agree/disagree and why. let's keep this civil and productive.
Well... I said this several times, but I'll say it again.

After 21 years working heavily on Macs, from film and sound editing, graphic and VFX design, post-production systems design and consulting I stopped recommending and relying on Macs.

Macs lost not only their upgradability, but also (and this is at least as bad) their customisation, therefore not meeting users' needs right from the start. Macs are quickly returning to their closed cocoon, contradicting the recently shared words "technology is more powerful when it empowers everyone". They also became quite more expensive, specially if you're outside de US.

I already listed what I would need in Mac in order to buy it.
Amongst other things, one of my requisites is nVidia GPUs. Regardless of being better or worse, they are the only supported and recommended by the software I use.
And this is part of what I think is happening: Apple prepares their hardware, for their software, which makes sense to a certain point. But, again regardless of their software's value, I for one will not be conditioned by their hardware. The Mac is just a tool, not an end in itself, just like hardware and anything man made. So, if I come to the conclusion that What I need is to change my hardware setup, instead of software I will. In fact I already did it and recommended the same to whom consulted me.

Another thing on the list was the cost. Apple is expensive, except for the newly launched iMac Pro. The problem with this machine is not exactly in it's value, which is great, but again, customisation and upgradability. A part from this model, Macs are expensive and keep lifting.

There was however one thing I didn't put on the list, which for me is more and more mandatory: Support. Also a good warranty. They both a very poor. Apple's guarantee speaks for itself. The competition's blows it way, by far. Don't expect me to buy the so called "best computer on the planet" with a slim 1 year warranty. Even in the EU market, Apple fails to comply with their warranty laws. The "best computer on the planet" must have the "best warranty on the planet" to go with it. That's the real proof of manufacturer's confidence in its product. As for support… just too bad to be true. Most people don't even know the basics of the products their supporting… and again Apple's guidelines are tricky, just like a classic insurance company.

Over the recent years I've seen Apple reshaping itself… unfortunately not for the better, in my opinion.
Apple became a technological jeweller, more interested in (a very questionable) aesthetics in detriment of functionality and stability. That happened since Yosemite, culminating with current OS. In the hardware side, more and more proprietary, custom made, non standardisation and again lack upgradability and repairability has been adopted, which is the worst you can do to protect the environment - in fact is the opposite.

Yes, I'm a switcher… to the Windows side, not pretty, nor perfect, but it meets my needs. Switched to HP - runs wonderfully, cheaper, faster, 3 years warranty, one phone call away and a support that really supports me.
Windows with all it's flaws, doesn't change every year and has very wide compatibility scope, for old software as well. I'm satisfied.

The iMac Pro was something good to be seen. Is a step in the right direction.
But is a very small step, built on an ancient thinking.
Far from enough for me to return, really far.
Apple will have to reinvent itself, including reinventing why it should reinvent. Somethings are just great and don't need to be changed.

I'll be around and… who knows?… may be Apple surprises me again… positively this time.

"La originalidad es volver al origen" Antoni Gaudí
 
Last edited:
The problem is timing; if the change of mind happened right before 2017 April roundtable, then a trashcan update should have come before that (as incremental as it may have been).
In the transcript Craig Federighi says, it was "longer than six months ago" they realised the situation. So, they had more than six months before the 2017 April roundtable to think and plan.
 
In the transcript Craig Federighi says, it was "longer than six months ago" they realised the situation. So, they had more than six months before the 2017 April roundtable to think and plan.

Right, but the original plan was replacing the Mac Pro with the iMac Pro.
 
  • Like
Reactions: ssgbryan
So Apple hardware may be without any substantial change for upcoming... 3 years.
But the MP6,1 is using CPUs that were released five years ago.

Apple could make a huge leap by using the current generation of Xeon-W (like the 18-core iMac) and Scalable Xeons (up to 28 cores per socket).

You seem to be making an argument that Apple should wait for 10nm, rather than shipping the much better (than E5-x6xx v2) current Xeons.

Sometimes your "AMD agenda" leads you to post arguments with a questionable basis in facts. This whole 10nm tangent seems to be a case in point. There's no reason to claim that Apple can't update its (very old) MP6,1 systems before 2021 - much better chips are here now.
 
Last edited:
  • Like
Reactions: filmak
I can see the mini becoming the pro and the new mini adopting the atv form factor

Always pushing for smaller , thinner less is more

Scary thought tbh
 
Why does it have to be tiny?

Apple has a quest for ever decreasing size and ever increasingly simpler computers. Part of it is that consumers prefer a smaller simpler product and part of it is just Apple. The cMP is big and heavy and hard to ship and they aren't ever going to make a computer like that ever again.

Craig Federighi: "I think it’s a strength of the company that we see new technologies creating new opportunities. We tend to try to jump on those pretty aggressively and so you look at that architecture of that Mac Pro, it had great Thunderbolt external I/O and we said: ‘This is a great opportunity to change what had been a conventional build a big card rack and slot a bunch of cards in there.’ We said: ‘a lot of this storage can be achieved with very high performance with Thunderbolt. So we built a design in part around that assumption, as well. Some of the pro community has been sort of moving that direction, but we had certainly in mind the need for expandability. If you wanted a great RAID solution in there, it probably made a lot more sense to put it outside the box than actually be constrained within the physical enclosure that contained the CPU. So, I think we went into it with some interesting ideas, and not all of them paid off."

Craig doesn't see replacing SATA bays and PCIe slots with thunderbolt as a mistake, when he says not all of the interesting ideas payed off he's referring to dual GPUs and thermal limitations that he talked about earlier but the overall concept as a small box with external storage connected via thunderbolt he still approves of. He's totally wrong about RAID and PCIe slots but I don't think he sees it that way.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.