New MacPro predictions? Hopes? Fears?

[DJ Meyer]

Intel just hasn't been able to advance their CPUs well enough.

I think people are underestimating the amount of performance that is being left on the table by Apple's design and laziness in updating it. The highest end Tube available comes with the E5-2697v2. In 3D rendering here's the speedup (approximate) you can get with modern replacements:

E5-2697v4 +40%
E5-2699v4 +71%

You could also go with the dual E5-2699v4 if you'd prefer a 342% speedup. Sure, its not cheap at the high end, but with Apple the high end is no longer even an option.

Not to mention faster memory, GPU, storage, etc etc.

 

[Wayne Morellini]

Mike Lol! Considering how everybody flocked on the MacPro like seagulls on a chip. When I got to see what it was actually like, I felt "but".

Re-edit: Sorry for the familiar tone Mike, I thought you were Marc before (text too small).

 

[Wayne Morellini]

DJ, good points, this is only a relatively low and setup. They need a second, bigger can to do everything you say.

 

[Wayne Morellini]

Jeff, I'm going over to your workstation thread to suggest maybe a new thread on building a cheap blade based extendable processing farm like I suggested for the Mac mini Blade above. A sort of sticky that canbe updated with freshly founded parts and designs.

 

[Mike Halper]

I think people are underestimating the amount of performance that is being left on the table by Apple's design and laziness in updating it. The highest end Tube available comes with the E5-2697v2. In 3D rendering here's the speedup (approximate) you can get with modern replacements:

E5-2697v4 +40%
E5-2699v4 +71%

You could also go with the dual E5-2699v4 if you'd prefer a 342% speedup. Sure, its not cheap at the high end, but with Apple the high end is no longer even an option.

Not to mention faster memory, GPU, storage, etc etc.

There's a difference between theoretical performance and real-world performance. Just because specs and benchmarks show 40% or 71% performance increases doesn't mean you'd actually see that in real use. The problem is that most applications cannot scale that many cores effectively, and so most of the time you'll get better performance out of the higher clock speed CPU's than the lower GHz/higher core count CPU's.

 

[Phillip Lakis]

There's a difference between theoretical performance and real-world performance. Just because specs and benchmarks show 40% or 71% performance increases doesn't mean you'd actually see that in real use. The problem is that most applications cannot scale that many cores effectively, and so most of the time you'll get better performance out of the higher clock speed CPU's than the lower GHz/higher core count CPU's.

Very true. A lot of software isn't coded well above 12 cores.

 

[Blair S. Paulsen]

Very true. A lot of software isn't coded well above 12 cores.

A lot of software rarely uses more than one core - especially on OS X ports, which are typically an afterthought. Moving to ARM, or any other major architecture change, would likely make that discrepancy even worse. The trash can deserves much of it's criticism, but more effort to optimize drivers/graphics libraries/etc by software vendors would help.

To me, the simple fix is to make the tube taller so it can host regular PCIe cards, up the PSU to 650w, bury the hatchet with nVidia to get Pascal cards with optimized drivers inside, move to Kaby Lake CPU family with lots of PCIe4 lanes and you'd have a compact workstation many RedUsers would snatch up in a heartbeat. Sadly, that does not seem likely...

Cheers - #19

 

[jake blackstone]

To me, the simple fix is to make the tube taller so it can host regular PCIe cards

Yes and while at it, add about four corners to that tube
What's the fascination with the tube shape? Majority of cards are square, so why trying to squeeze square peg into the round hole?

 

[Jarek Zabczynski]

Yes and while at it, add about four corners to that tube
What's the fascination with the tube shape? Majority of cards are square, so why trying to squeeze square peg into the round hole?

Think different bro.

 

[jake blackstone]

Think different bro.

Form over substance doesn't mean thinking different...

 

[DJ Meyer]

There's a difference between theoretical performance and real-world performance. Just because specs and benchmarks show 40% or 71% performance increases doesn't mean you'd actually see that in real use. The problem is that most applications cannot scale that many cores effectively, and so most of the time you'll get better performance out of the higher clock speed CPU's than the lower GHz/higher core count CPU's.

If you're more interested in clock speed there are plenty of other modern Xeon's they could have also used. I do a lot of 3D work (hence the context of the post) and the performance scaling is as near to linear as it gets.

 

[Wayne Morellini]

Hmm a square tube with slightly angled sides. Form and function.

 

[Wayne Morellini]

There's a difference between theoretical performance and real-world performance. Just because specs and benchmarks show 40% or 71% performance increases doesn't mean you'd actually see that in real use. The problem is that most applications cannot scale that many cores effectively, and so most of the time you'll get better performance out of the higher clock speed CPU's than the lower GHz/higher core count CPU's.

Being a past researcher in this area I can tell you there is a way, why it isn't being used on the x86 PC/semi PC, is a mystery. I have not kept up with what the outside world has been doing for decades. But people don't seem to know of it, I'll keep it to myself.

Anyway, doesn't the use of these new Open CL, or Open MP, support the distribution of programs accross processing resources? Ultimately, specifically programming to a distributed model is the best way to go.

 

[Marc Wielage]

Hmm a square tube with slightly angled sides. Form and function.

Jesus, just make a 19" rack mount cabinet with enough slots to handle 6 PCI cards. Every big-time audio/video pro in the world would buy that, particularly if it had upgradable CPUs and GPUs. Unfortunately, Apple's priorities and goals are not our​ priorities and goals.

 

[Frank Glencairn]

Ti

I wish MS would release a Pro version on Win10 with a lot of legacy support factors removed, more features turned on, less processes running in the background, no F#$%ing auto updates, and with a focus on stability.

Apple adopts Windows tactic to auto-download Sierra to Macs
http://www.computerworld.com/articl...0-tactic-to-auto-download-sierra-to-macs.html

 

[Jarek Zabczynski]

Jesus, just make a 19" rack mount cabinet with enough slots to handle 6 PCI cards. Every big-time audio/video pro in the world would buy that, particularly if it had upgradable CPUs and GPUs. Unfortunately, Apple's priorities and goals are not our​ priorities and goals.

Marc Wielage for CEO...

 

[Antony Newman]

I think there may be a subtle difference in strategy.

+) Apple have waited some time after availability to ensure that Sierra does not bork systems during upgrades before turning on auto-download.
+) If there is insufficient free space -> no download
+) If the system did have space, but starts running low -> the download is deleted
+) There are menu options that you can take to Not upgrade (and delete the download).

On the other side of the fence - if you Do decide to upgrade to Sierra when prompted by auto-download ... there is No network wait. For many, this will minimise their downtime.

AJ

Apple adopts Windows tactic to auto-download Sierra to Macs
http://www.computerworld.com/articl...0-tactic-to-auto-download-sierra-to-macs.html

 

[Antony Newman]

INTEL : E5-2699v4 : 7.2 Billion transistors : $4,500 to buy
AMD : Polaris 10 : 5.7 Billion transistors : $199 to buy
APPLE A10 Fusion : 3 Billion transistors : Estimated cost to Apple $30

Intel would very much like their CPUs to the limiting factor in what we can afford when buying a new computer.
Intel can dictate this at the moment as they have a monopoly on high end x86. Intel also know that people desperate for more CPU power than their 2 year CPU will sign them another blank cheque.

As unimpressed as I am about what Intel charge for their 7B transistor designs, I am impressed by AMDs 6B costs.
When 4 of Apples mobile phones can compete in performance with a 12 core Intel design considered state of the art in 2013 - I think there is some ground breaking work being undertaken in the FruityCorp.

I don't think it is necessarily laziness that has prevented Apple upgrading to a slightly faster (12 core) chip XEON chip in 2015, but more likely:
+) Their inability to negotiate a deal with Intel (Intel probably want their cake and eat it)
+) Apples desire for Xeon performance/watt improvement within a 140W TDP
+) Intel not wishing to release a 10nm Xeon Part (which probably improved performance/Watt by 50%) as they can keep milking their 14nm process.

MacOs is tuned for up to 12 cores. With the bandwidth restrictions (eg chip to memory), there comes a point where the majority of applications would not see a performance increase with more cores. As XEON chips cannot work flat out without overheating, the point at which the XEONs self regulate and slow down is proportional to the number of cores working.

AJ

I think people are underestimating the amount of performance that is being left on the table by Apple's design and laziness in updating it. The highest end Tube available comes with the E5-2697v2. In 3D rendering here's the speedup (approximate) you can get with modern replacements:

E5-2697v4 +40%
E5-2699v4 +71%

You could also go with the dual E5-2699v4 if you'd prefer a 342% speedup. Sure, its not cheap at the high end.. <snip>

 

[Wayne Morellini]

He, he, yes I'm on that too Antony. But thought of a major issue, but it can be designed around.

Less hopeful people seem to have just negatives a lot I'm saying less hopeful, in the vein of not so hopeless.

But, thanks for the pricing there, rather than the ussual aBS people around here get into. Shows the potential. But the issue is to start the revolution on mass and people develope the software to suite, in between, as long as you can do well compared to the competition, it doesn't matter, you can do spectacular latter. So, those 3, mostly wasted, transistors, can be crafted in more productive ways than being all CPU's, you can include arrays of other processing resources, a thousand fitting in the space of one CPU, each performing a variety of calculation tasks, each at a speed the CPU would do. Now, think sideways a bit more, we know the high end workstation workflows involve specific calculation area, and soacific maths. So you can sell a CPU with an calculation array tuned for different industries So, audio/video, 3D, and cariouse engineering and scientific. So, a lot of what you get out of a CPU or GP-GPU is unseeded, and could be replaced by a lot more specifically c processing power. Now, for less transistors you csnnoutdo the best x86. But people have trudge for brains.

 

[Jarek Zabczynski]

MacOs is tuned for up to 12 cores. With the bandwidth restrictions (eg chip to memory), there comes a point where the majority of applications would not see a performance increase with more cores. As XEON chips cannot work flat out without overheating, the point at which the XEONs self regulate and slow down is proportional to the number of cores working.

AJ

Exactly, which is why I still think a high clocked 6 core dual CPU design is still the best. They really crippled the nMP by giving it one processor. Not to mention fewer PCIe lanes.