Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

New MacPro predictions? Hopes? Fears?

So that is you getting ready to hit the ground in that cartoon Paul, with Marc about to hit you with that odd football shapped rock, to make sure. :)
 
Wayne Morellini said:
So that is you getting ready to hit the ground in that cartoon Paul, with Marc about to hit you with that odd football shapped rock, to make sure. :)
Click to expand...
No, I am the football in that cartoon. I'm always getting jerked away by unrealistic anticipation, especially from Apple.
 
I hope for the 5k screen with built in graphics. That would be the smart way of having all the current trashcan owners to buy their way around poor / not working graphics. And to me that would also unleash more power to laptops in a smart way. I think having a but TB enclosure with fast graphics in a box does not make sense, like the PC laptop gamers do. Much better to stick that graphics into the gaming screen sort to speak.
 
I wonder how this will work. People complain that external GPU boxes have too little bandwidth, but this would be what you face with a MAC Pro. Though people spin around like things are breaking apart over this, the truth is you just program differently to get more benefit. The issue for MacPro, is now you have two data areas separated by a limited data rate connection, so getting them to work together on some things will be difficult. However, with future programming you do some things on one and some things on the other as you process (but you want to save your work).

I wonder if it is just a player function, maybe with an embedded OS functionality with Apple Arm CPU/GPU?
 
I've been musing why Apple have included ARM compatibility into their public Sierra offering - As this is too soon for Apple to have a Chip and eco system to replace Intel (eg no TB), and have reached the following conclusion:

+) Their laptops will include Both Intel & ARM chips!

+) For those of you that sat through the BlackHat seminar - Slightly 'heavy' session:
https://www.blackhat.com/docs/us-16/materials/us-16-Krstic.pdf

I think Apple need a secure enclave (which Intel does not provide) - from their mobile ecosystem.
And THIS is why they have ARM hooks.

+) This enclave will proved the backing they need for fingerprint verification (replacing user login credentials),
ApplePay verification too.

+) AND enclave biometrics could also be used to encrypt data in APFS (semi speculation on my part).

+) And to provide a completely protected bootup method for the OS (as was presented in BlackHat).

Apple will beat Intel in IPC using more transistors (i.e. area per core than Intel).
Intel cores will work at higher frequencies (Intel are more concerned about area and transistor count, smalller = less distance to go = higher frequency)
ARM A73 is a wimpy ARM core that goes higher GHz than Apple due to it size (smaller = faster).

And so if Apple needed a chip to replace high end Xeons is would actually use a lot more transistors than Intel (but not get as hot / not run at such a high frequency)
I don't forsee this happening at TSMC 10nm (not enough transistors for an Apple replacement )
It would be possible at TSMC 7nm - but this would mean the earliest replacement for XEON would be in 2018H1.

And Apple would need to stop using Thunderbolt ....

If Apple Do produce another Mac Pro .. I am back to thinking it will be Xeon again ... Intel 10nm + 12 core + 4Ghz.

AJ

Edit : Hmm : Perhaps I need to qualify Intel CPU sizes vs ARM.
 
Last edited:
Not as far fetched as it sounds Antony. I concluded a second chip on motherboards maybe decades ago, and since then there has been some attempts. The second chip provides a second pre-ooerating system, to do basic things and play media while consuming massively less power then the Intel Chips back then. Such a thing could still be useful, but the arm running the normal OS, and booting x86 for x86 performance binaries when they need more performance. The arm costing very little. However, maybe it will be Arm only Air, as you said before. I think Arm would be particularly competitive against this level of x86 versus Xeon. It gives developers a target to start supporting Arm before general roll out over successive years.

I would have thought the junctions would have been a more limiting factor in speed than distance but maybe Jeff will know?

I think they can achieve higher performance now with less transistors and less heat on Arm against Xeon, but the x86 binary translation maybe a bit of an knock against that. I know what happens in chip design and chip manufacturing processes, is they make it either low powered or high frequency. I know somebody that could probably do 10ghz Chips today, but officially does maybe still sub 1ghz ultra low powered Chips through his unique design methods, which are industry leading. I imagine he is earning good money with the military contracts though on high performance versions. Anyway, that is why Arm was using rediculously large process sizes years ago, they were designed to be much lower powered. But now Arm can use higher speed. But the higher you go the less you get per watt, and the harder it is to cool, affecting array size. So, backing off on the Ghz, which Intel has done for more than a decade (because high speed chips were actually over clocked to get there) allows you to put larger amounts of transistors together per watt = more performance. As well, having smaller core design = more parrallel GHz (and simpler cores could go faster but only so much with overockkng on top of that, so it is a matter of picking best balances as a smaller core can mean less functionality. Now Arm traditionally were very light and small from a limited instruction set and functionality, but now are little monsters full of nearly as many potential instruction options as Intel, but your ISA design effects how many transistors and speed the circuit can work at (or in Arm's case towards power savings). However, I guess that Arm has the smaller advantage overall, but if you look at a i7 processor, the integer instruction decoding section is a small part of it compared to other parts of the processor, so your savings are limited. Now, the small top end Arm core, how limited is its functionality compared to an i7, and the other trick is how lower unit of work are the features to save energy and size? However, with my design I am sidestepping this stuff with smaller design with high unit of work density and low power, and high frequency. How far do you think the distance between 4k transistors is compared to millions Antony?
 
So is it 8k proxy, what intermediary format?
 
Can't find HP z1 G3 on their website. Is there some date they use for releasing products coming up?

http://www8.hp.com/au/en/campaigns/workstations/z1-g2.html

It is strange the 4k 23.4 inch version but no large version for proper use. But here the old large touch version is there, not pulled.
 
Notice the magical fairy dust like and sparkles in the slowmo, that surrounds the release of many Apple products? Is this a coincidence.
 
If you see Cook roll out a big Nvidia GPU like this on stage under a massive sheet you may well know which GPU is in the Apple display.
 
My prevoius statement about chip area vs top speed was not exacting enough.

Apple-A10-Fusion.jpg


The ARM A72 (As used in the Kirin) delivers better performance per square mm than the Apple Hurricane.

The Hurricane provides better IPC than the above ARM chips And is comparable to the Broadwell (big blue rectangle in the background)

Without 'Signal boosting' on the core, there is voltage drop as signal travels around the chip. Apple don't use this (LLC) circuit (as far as I am aware), but Intel do.

When LLC (Load Line Calibration) is Not in use, the Area (or distance that signal has to travel) vs top chip speed is intrinsically linked.

AJ
 
Yes, disappointing the calibration and multiple tap (but what process and speed are each design above).

But the most important things is,what do you think of the Nvidia mythbusters integrated GPU display proposal above? :smiley:
 
Actually, I wouldn't mind having access to one of these things, and when people answer something we can consult it, and it prints "You're Wrong" and tell them "There, it renders the answer pretty quickly". When they protest you ask them to try another question, and ask them to answer the meaning of the universe, and then press the button again.

(Actually, I've got to find one of these things to make an animated gif of it printing "You're Wrong" for use on the forum.
 
Intels' offerings :

intelcoresb7jai.png



For x86 : Intel performance increases in CORE are limited by:

+) WATTAGE of a CORE: When Intel moves to 10nm, and power per goes down by a factor of X : This will correspond to increase of X in performance.
I would expect this jump to be (guess) 30%. (roughly the same 30% from 22nm to 14nm)

+) PACKAGING losses : When Intel upgrades to an equivalent of TSMC like (InFO) : Their Chips will increase by (guess) 15%

+) IPC : Intel ISA is very very heavily optimised. I don't expect a magic improvment here. I guess that 5% in optimisations per generations.

And so for Apple get a Meaningful improvement in speed from an Intel Xeon - they really need a 10nm Xeon + 6 x Generation updates from Intel.
Ivy Bridge (22nm Xeon in Current CPU in nMP) -> Haswell (14nm) -> Broadwell -> Skylake -> Kaby Lake -> Coffee Lake -> Cannon Lake (10nm)

This would probably give a (back of the envelope) performance increase of Cannon Lake Xeon of (1.3 x 1.3) x 1.05^6 = 2.3 x Speed increase for the same Power
If Apple don't hold out for a 10nm ... they will (another back of the envelope fudge calculation) get a 1.7 time speed up (on 14nm) - Assuming that Intel do NOT roll out 'advanced' packaging for consumer chips (as there is not competition forcing their hand).

AJ
 
Years ago, before I stopped reading up on things die to health, there was an initial article on the amount of power savings if each generation shrinking. From the last Nvidia article I read a few years back, it will dissapear completely around 7nm. So each generation the power savings should he getting less and less, so generation to 10nm shouldn't be much. However, technique can make a lot of difference. So, we see different processes at the same size with very different characteristics (some times the line length spacing is less than another for instance, and packing is more dense in a circuit). The x in performance and the x in energy saving are different and variable. Now, what has svtuslly been gapoeing over the last decade or so, is they have been dropping overclocking overhead to save power. Hence we often are still under 5 ghz for 10 years. Even FPGAs are catching up (without that 5Ghz FPGA tech of 12 years ago that the military snapped up). So, are we really getting lower power simply by running the chip closer to it's natural speed (likely around 1ghz or more at a guess, if you look up the data you should find it). So will 7nm turn out to be 2-3ghz. I don't know, not keeping up, could it be closer to 4ghz, maybe some mild overclocking.
 
You must log in or register to reply here.
Back
Top