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?

AMD EPYC looks very promising. I guess we’ll have to see if it lives up to all the promises and hype. The pricing is aggressive, so we’ll see how Intel responds to that.

Keep in mind that the official launch was yesterday and real products won’t be in the channel for mainstream customers for a few more months. EPYC is not competing against Intel v4 Xeons. It will be going head to head with Skylake-EP/SP Xeon “Platinum” “Gold” “Silver” etc.. CPUs. Intel has the products in their pipeline. AMD is going to force them to be price-competitive and stay timely on release schedules. Prices on Intel offerings are not going to drop until AMD starts gaining market share.

64 cores.... Sooner than you would probably guess.
 
Jarek Zabczynski said:
Anyone think we might get a traditional tower? Maybe just a slimmed down version of the cheesegrater? In space grey of course!
Click to expand...

It has been said that Apple has one in development. I believe it will be a 2019-2020 sort of thing and not something they are going to launch in the near future. There are a lot of things baking in the tech oven that need to be fully cooked in order for Apple to give customers a worthy Mac Pro option.
 
Jeff Kilgroe said:
AMD EPYC looks very promising. I guess we’ll have to see if it lives up to all the promises and hype. The pricing is aggressive, so we’ll see how Intel responds to that.

Keep in mind that the official launch was yesterday and real products won’t be in the channel for mainstream customers for a few more months. EPYC is not competing against Intel v4 Xeons. It will be going head to head with Skylake-EP/SP Xeon “Platinum” “Gold” “Silver” etc.. CPUs. Intel has the products in their pipeline. AMD is going to force them to be price-competitive and stay timely on release schedules. Prices on Intel offerings are not going to drop until AMD starts gaining market share.

64 cores.... Sooner than you would probably guess.
Click to expand...

So, the xeon phi desktop product.
 
L. Langer said:
It has been said that Apple has one in development. I believe it will be a 2019-2020 sort of thing and not something they are going to launch in the near future. There are a lot of things baking in the tech oven that need to be fully cooked in order for Apple to give customers a worthy Mac Pro option.
Click to expand...

They explicitly indicated a modular system, like what I suggested last year. If they do it right you can literally make a processing farm.
 
Wayne Morellini said:
They explicitly indicated a modular system, like what I suggested last year. If they do it right you can literally make a processing farm.
Click to expand...

They said that within the context of the discussion, which was that THEY were having problems updating it.

Modular to them could just as easily mean "a tower or other design that allows US to more easily update components."
 
Wayne Morellini said:
They explicitly indicated a modular system, like what I suggested last year. If they do it right you can literally make a processing farm.
Click to expand...

You're interpretation of "modular" probably varies widely from Apple's.
 
Jeff Kilgroe said:
AMD EPYC looks very promising. I guess we’ll have to see if it lives up to all the promises and hype. The pricing is aggressive, so we’ll see how Intel responds to that.

Keep in mind that the official launch was yesterday and real products won’t be in the channel for mainstream customers for a few more months. EPYC is not competing against Intel v4 Xeons. It will be going head to head with Skylake-EP/SP Xeon “Platinum” “Gold” “Silver” etc.. CPUs. Intel has the products in their pipeline. AMD is going to force them to be price-competitive and stay timely on release schedules. Prices on Intel offerings are not going to drop until AMD starts gaining market share.

64 cores.... Sooner than you would probably guess.
Click to expand...

We've already seen the first respond to the not released Threadripper from AMD with the X299 platform and the i9 (HEDT) intel and that looks hot.
The EPYC platform is on sale now, some of the processors will be later and arrive end of july/beginning august, same counts for the Motherboards. So the mainstream customer will be able to recive it's platform whitin a few(2) months. I've only heard rumors about Skylake-EP/SP Xeon but I think they will have a hard time getting good yields for the 24..32 cores that fit in their TDP envelope, but I hope I'm wrong.

64 cores... with intel 2nd gen 10 nm or 1st. gen 7 nm. AMD should be able to do it now with a double sized size (8 x Ryzen 7) but it's more likely they will do it with GloFo's 2nd gen 7 nm. Time will tell...
 
Moores Law

Intel's move to FinFET addressed transistor leakage at reduced feature size - but dramatically increased the design complexity and manufacturing costs.

When FinFET structures go below '7nm', there is a marked increase resistance, requiring more current drive - increasing the power density, which might even reduce the transistor performance.

IBM 5nm : The jump to slices and stacks of varying thickness sheets of conductive material (Gate-all-around) design is going to introduce a node shrink, some design simplifications, but with markedly lower resistance.

I would anticipate a win-win on moving to this node. Reduction in per-transistor manufacturing cost, and a sizeable bump up in transistor performance.

AJ

Edit : Additional info on why a 'bump' will occur at 5nm nanosheets.

http://spectrum.ieee.org/nanoclast/...anosheets-ibms-path-to-5nanometer-transistors

+) Many companies are struggling with the extra cost of going from Planar (28nm and soon 22nm) to FinFET due to the massive increasing in design complexity & exact requirements of the Fins at nanoscale.

+) All node shrinking thus far has increase complexity (and with it manufacturing expense)

+) 5nm nanosheets will uses exactly the same number of masks as FinFET would have - but the design an simulation costs reverts back to the far simpler PLANER design, along with far more exacting nanosheets thickness (vs variation of FinFET height), far less dark silicon (less current leakage), and higher running frequency.

There will also be a bump with Denmark scaling (power = I^2.R) due to the drop in current drive.

And of interest to those who think that lots of cores is what we need will note that Koomely's law will also have a bump (perf/watt).

Parallelisation of cores:

The 'big problem' with HPC code is parallelisation. Trying to rewrite code from x86 -> GPGPU is 'too involved' and 'to risky' for many groups.

The switch to vectorisation is also complex as lack of standardisation makes the compilers job (of recognising where it can vectorise) rather tricky.

That said - this has been worked on diligently since 2011; And as the fruits of the Mont-Blanc project are more prevalent, HPCs will more readily be able to use cores with a higher efficacy.

Hopefully sometime Before 2021!

https://www.nextplatform.com/2017/06/30/momentum-building-arm-hpc/
 
Last edited:
Timing

Timing

If I'm reading the tea leaves correctly, Apple took their mid-level desktop line - iMac - and created a "Pro" variant that, at least on paper, is a real workstation class device. It's not a "monster" workstation, and typical RedUser tasks are likely to push it pretty hard, but it's release at the end of 2017 allows them to "quell the insurrection" for a while. With some of the time pressure relieved, it would be much easier for them to wait a few extra months for a key component or technology to be available that could extend the useful lifespan of what is likely to be a very expensive offering.

As a potential buyer of both iMacPros and m/MacPros over the next few years, I would advocate for such a strategy. I can drop my $12K-ish on a well equipped iMacPro to hold me over until they release the beast. I'll probably have to access render farms for certain things, but the if the iMacPro lives up to expectations - not that often. Whatever the correct interpretation of Moore's Law might be, I've been in this space long enough to know there are key inflection points for various technologies. If Apple can time it right, they might be able include some significant advancements in the m/MacPro 7,1 which would extend it's useful life - maybe even long enough to ROI it ;-)

Cheers - #19
 
Jarek Zabczynski said:
You're interpretation of "modular" probably varies widely from Apple's.
Click to expand...

Sadly, you maybe right. Their version of modular varies from most people. If they do a normal tower it won't be very revolutionary, and not the most flexible or highest performing. But that's Apple, redefining around reinterpretation of normal based on what they do. I suppose you can look at a tower as modular, especially server like tower cases, rather than just expandable/reconfigurable.
 
So, maybe it will be just a bigger trash can where you can swap things with some expansion on the base.

Somebody was speaking about the non expandability of the IMac Pro and how that will lead to wastage. For desktop publishing, photography in a way, and a numbed of things, it will ha e enough power to keep for the rest of it's life, in uses that don't require expansion for extra performance.
 
Rack mountable please. I would argue that a large percentage of the market for the modular MacPro 7,1 workstation will keep it in a machine room, adjacent ventilated closet, etc rather than listening to it whine all day long. The iMacPro and the 2013 nMP 6,1 cylinder are designed to live on your desktop, so they have to carefully manage the cooling/noise ratio. For the upcoming monster I'd happily live with a few more decibels, in exchange for cooling that can handle challenging tasks like long ass renders.

Cheers - #19
 
Wayne Morellini said:
Sadly, you maybe right. Their version of modular varies from most people. If they do a normal tower it won't be very revolutionary, and not the most flexible or highest performing. But that's Apple, redefining around reinterpretation of normal based on what they do. I suppose you can look at a tower as modular, especially server like tower cases, rather than just expandable/reconfigurable.
Click to expand...

That's the thing though. We don't need anything revolutionary, or thinner, or whatever...we just want power and flexibility. It's a concept Apple refuses to acknowledge.
 
Well, you don't need Apple then. But if you want to do more in less, at less cost, you need something a bit more revolutionary.
 
Designs are topping out where they are because of poor design choices in the last 5 decades. Which means I must have been born at the time coming up to Unix and c for a reason. The choices then has put ceilings on processing density and cost now.
 
Wayne Morellini said:
Well, you don't need Apple then. But if you want to do more in less, at less cost, you need something a bit more revolutionary.
Click to expand...

I don't need their hardware, it's the OS I love. I simply don't care for Windows. If Apple let HP build OSX Ready Z800 series workstations I'd buy one in a snap, so would half the folks here.
 
I used to hink like this concerning the OS's, but I give Windows a break he's days, they have made significant improvements since Windows 7.

Apple made the wonderful connection, which I also designed for, a consumer electronics grade OS (reliability) but many other things with the OS. That's why your DVD, bluray, hifi, PVR works s reliably closed to computer systems. The whole object orientation movement was supposed to.lead to more reliable manageable systems, but they still stuff it up with repetitive prototyping programming.
 
What Apple could do for you guys, is release three motherboards with macos preinstalled, with you interfaces to the hilt for you to come configure how you like. One small itx formfactor, with dual sli support. One largest form factor for workstation server, and one inbetween. They would not be cheap. The small one likely $200+, the mid range one upto $500 etc (comes with OS license).
 
You must log in or register to reply here.
Back
Top