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.
Mac Pro 2013 Rumored Specs?
- Thread starter Nandifix
- Start date
- Sort by reaction score
Exactly, which is why I emphasized instances on my first post. Anyway none of us know what will happen if anything in this field. I just thought it might be practical to put some 64 bit ARM co-processors in an otherwise x86 box and get modest cost, with mostly existing software. I'm probably wrong and am willing to see actual results as my guide.
Exactly how do you efficiently and effectively run more instances of x86 based software with ARM processors ???? That isn't practical. You are still completely missing the "practical" aspect. Apple could paint purple butterflies on the inside of a mac Pro. What practical use is that?
If someone needed 5 iPhone concurrent simulators maybe ... but that isn't a big market.
P.S. OpenCL code doesn't buy anything because that code can be dispatched to the already available GPGPU cores.
Last edited:
Actually I suggested an x86 box running OSX with xGrid and others controlling several ARM chips running iOS or OSX as the instances. All that runs right now. The only thing lacking is the 64 bit ARM chips at 20nm right now.
I am strongly in favor of GPGPU but I am concerned about software more on that implementation. Needs work.
Do both in a MacPro!
Well, thank God you're here to tell the rest of us how dumb we are, having opinions and all. Hopefully you don't mean to sound quite so superior.
No one has ever provided proof of this. As volume takes time to ramp up, it might be easier to fill a small order from Apple than a large order from HP, but as Deconstruct pointed out, the supercomputer vendors tend to get the first units. I would not suggest anyone use such logic as a guide on when to expect updates.
No software supports ARM on Xgrid.
Almost nothing supports Xgrid at that.
Why would you use Xgrid on ARM instead of OpenCL on a GPU? The use case doesn't make sense. And the GPU is cheaper and more powerful. I'm not sure how you could be concerned about OpenCL and NOT be concerned about Xgrid. Xgrid is way more flakey than OpenCL, and no one even supports it. Not to mention Xgrid requires a new port for each CPU type, while OpenCL works.
Apple moving back to Xgrid for local work would be like them sticking a Zip drive back in.
actually all the substantive buyers get product before the Intel announcement. There isn't much special about that. Some vendors may need more time to build quantity to meet the initial demand surge. ( e.g., like Apple building iPhones for 1-2 months before launch). If the initial demand surge is low and Intel didn't overt restrictions then a relatively small vendor could bolt out of the gate early. The only special treatment there is not getting "slap down" for doing it.
I don't think Intel is allows folks to bolt out the gate early anymore.
" ... Advertisement and sales of Haswell CPUs and Lynx Point products will be allowed starting from June 2, just two days before the Computex 2013 show in Taiwan. ..."
http://www.cpu-world.com/news_2012/2012122901_Intel_Haswell_processors_to_launch_in_May_or_June.html
It is sloppy because there are usually leaks and prevue products at Intel Developer Forum and trade shows (e.g., Intel effectively released slides about their entire E5 motherboard offering before the "official" announcement) but as far as "taking orders for John Doe" that is a no go without permission from Intel.
In 2008, Apple jumped the gun but if anything they got decreasingly unattached from and increasingly trailing to Intel's announce dates from there. Given Apple didn't get an appropriate 3600 series line up to roll out a new Mac Pro in a timely fashion in 2010, it is not even clear they didn't get a "slap down" for doing that. ( I don't think so. Some folks at Intel are confused about the role that Xeon E3 plays in the Xeon product mix and how that does/doesn't mean there should be E5 1600's . )
This isn't about opinions. It is about the facts.
Fact: Apple has cancelled XGrid.
"XGrid Officially Dead"
https://forums.macrumors.com/threads/1412900/
So the likelihood that this is some magical glue that is going to make these ARM solutions relevant is deeply lacking in substance. Apple has already explicitly told folks they don't want to do XGrid like things. To couch that as being a difference of opinion is pure misdirection.
They are OK with someone else perhaps putting something like this together ( e.g., probably would not be adverse to Intel rolling out drivers to their Phi card and making their parallel compiler toolstack work on OS X. ) But as far as that being an "Apple driven" solution... no.
Fact: Name one current or proposed Apple iOS device with no screen output?
Crickets chirping.........
This is not a matter of opinion. iOS as a computational module OS kernel? Why? It is not an opinion that this is not even remotely motivated at all. This is not "think different". It is pull random stuff out the butt.
The primary purpose of iOS is to be a touch GUI oriented OS. It makes some sense to use it on AppleTV as there is still a GUI and large TV buttons to push by remote control. Even a watch running mini apps with possibly touch/voice/button control. But no GUI at all?
Apple Airport router products don't. No GUI. No iOS.
Fact: Apple is not a "I have a hammer so everything is a nail" company.
Massive Apple data centers don't run OS X. They certainly don't run iOS. Right tool, right job. Apple probably will have some ARM cores doing VM duty running front end web infrastructure at some point. Could it be custom boards? Sure google , Facebook and others build/commission custom boards for substantive numbers of their grid computation nodes. Do most of them sell them? No. Would Apple likely sell theirs? No. More than likely would buy "off the shelf" solutions; just like they do now.
"Apple designs some ARM SoC so therefore it has to go into Mac Pro somehow" is a "hammer / nail" , "round peg / square hole" groundless speculation. It is not oriented to solving problems that Mac Pro users have. It is a solution in search of a problem. It is actually far more indicative of thinking inside the box; not outside of it.
----------
The most likely candidate Xeon E5 was announced/released a year ago. So no it isn't. If Apple shipped next Tuesday most likely it would be something that has been shipping and/or announced for almost a year.
Even Xeon E5 v2 (Ivy Bridge) engineering samples have been floating around for a year at this point.
http://news.softpedia.com/news/10-Core-Intel-Ivy-Bridge-EP-CPU-Has-95W-TDP-at-2-4GHz-245709.shtml
BTW my original post said XGrid and other solutions. That would certainly include OpenCL.
As for being headless, of course it would have access to some monitor (or pad) on the network for configuration, but with multiple CPU imaging you often configure one image and implement it along a range of CPUs at once.
As for the have a hammer, see nails everywhere problem, my post certainly fits. I am not Apple. I am a guy posting to a rumors site. Speculation or suggestions to same.
I would think the solutions where you use a GPGPU, or a separate CPU instance, are largely mutually exclusive.
Anyway this is offered for the ether to further ignore or be inspired by. Just don't argue with the poster.
As for being headless, of course it would have access to some monitor (or pad) on the network for configuration, but with multiple CPU imaging you often configure one image and implement it along a range of CPUs at once.
As for the have a hammer, see nails everywhere problem, my post certainly fits. I am not Apple. I am a guy posting to a rumors site. Speculation or suggestions to same.
I would think the solutions where you use a GPGPU, or a separate CPU instance, are largely mutually exclusive.
http://hardware.slashdot.org/story/...chip-brings-big-gains-for-floating-point-appsarticle said:
Anyway this is offered for the ether to further ignore or be inspired by. Just don't argue with the poster.
Last edited:
If we're including OpenCL, we don't need ARM co-processors at all. GPUs are faster and cheaper and do the same things as a CPU cluster (because they are really a CPU cluster). Why would you go ARM? A modern Radeon has a few thousand cores on it.
On XGrid being dead: For local stuff, OpenCL and GPUs are pretty much the replacement for XGrid.