Nicholas Croft
Well-known member
- Joined
- Aug 4, 2016
- Messages
- 52
- Reaction score
- 0
- Points
- 0
Good times.
Last edited:
Welcome to our community
Be a part of something great, join today!
New MacPro predictions? Hopes? Fears?
- Thread starter Jarek Zabczynski
- Start date
Good times.
Antony Newman
Well-known member
Maybe the reason we are irritated with Apple is that we equate bleeding edge technology with Pros - rather that just 'leading edge'? And that many would rather live life 'twice a bright for half as long'? (To quote Tyrell)
A 75 Watt GPU from Nvidia is ~TWICE as fast as the 35 Watt AMD (R460) GPU in the MacBook.
Perhaps by Apple NOT towing the 'burn as much power a possible' mantra ... giving software writers the nudge to finally make their code more efficient.
... Like spreading ther CPU load of Several cores! (Yes that takes more effort!)
... of taking several GPU operations and coalescing them (Even more complex ... But Apple does this to accelerate their Metal graphics)
Perhaps Apple was ready in 2015 to roll out 10nm XEONs to give a real improvement in performance per Watt .. before the Intel CEO decided that (as there was no competitive reason to ... i.e. 'Yields would be better in 2 years!) that 10nm would be delayed.
Perhaps it is Apple refusing to 'cave into' Intels desire that people by 14nm now that has caused them to take a back seat in MacPro development?
Perhaps it is Apple pushing the leading edge forward - and disassociate our beliefs that we can only work with the bleeding edge?
AJ
A 75 Watt GPU from Nvidia is ~TWICE as fast as the 35 Watt AMD (R460) GPU in the MacBook.
Perhaps by Apple NOT towing the 'burn as much power a possible' mantra ... giving software writers the nudge to finally make their code more efficient.
... Like spreading ther CPU load of Several cores! (Yes that takes more effort!)
... of taking several GPU operations and coalescing them (Even more complex ... But Apple does this to accelerate their Metal graphics)
Perhaps Apple was ready in 2015 to roll out 10nm XEONs to give a real improvement in performance per Watt .. before the Intel CEO decided that (as there was no competitive reason to ... i.e. 'Yields would be better in 2 years!) that 10nm would be delayed.
Perhaps it is Apple refusing to 'cave into' Intels desire that people by 14nm now that has caused them to take a back seat in MacPro development?
Perhaps it is Apple pushing the leading edge forward - and disassociate our beliefs that we can only work with the bleeding edge?
AJ
andrewhake
Well-known member
- Joined
- Oct 4, 2009
- Messages
- 785
- Reaction score
- 55
- Points
- 28
Terry VerHaar
Well-known member
That's an interesting theory but it seems like a lot of people argue just the opposite - working pros can't afford to be on the "bleeding edge" because they have to rely on their tech day in and day out and can't afford down time. And, of course, customers want "bleeding edge" to be perfectly implemented - which it rarely is. So, I guess it depends on your definition of "bleeding edge" vs. "leading edge." And when IBM says that Macs are cheaper in the long run - because of the higher cost of keeping a Windows box running, isn't it important to the "working pros" that they have their machine operating without a lot of down time for maintenance.
Everyone has their own needs and perspective. Apple is Apple. One of their advantages is that they control the whole user experience by controlling hardware and software. They have their issues just like any other company. And, yes, they may not be as inclined to go to that edge when they face backlash from so many customers when things go south.
- Moderator
- #605
Jeff Kilgroe
Well-known member
Apple is rarely bleeding edge, with the exception of securing CPU exclusives for a few systems in the past. They are about balance and efficiency and in those regards, hit a home run with this new MBP. They have a habit of being late to market with the new tech and latest specs, yet tend to do it better (in their own way) than anyone else. I see a lot of people angry over certain aspects of the new MBP's, but what many are really asking for isn't anything an MBP would ever be, or most any laptop for that matter. I totally feel that Apple missed a few key elements with this MBP update. Namely the option of more RAM. Most of us would also appreciate a better GPU. That wasn't going to happen with their intended power envelope for this device. An nVidia GPU would have been much preferred for CUDA support, but at 35W we're not going to see any true gains here. And we should also recognize the good about their decision. The 80GB/s throughput of the 460 Pro GPU and its pixel pipeline is actually superior to many of the nVidia offerings at higher power levels. Not including OpenGL/3D or computational use, it's quite impressive.
This new MBP has already sold via pre-orders more than any other MBP model to date. Prior record holder was the 2012 retina introduction model. Which was also similarly criticized at the time for a handful of reasons. Similar to the 2013 Mac Pro trashcan. Heavily criticized around here and in similar professional circles. And yet, it just keeps on selling.
As for software developers working to streamline and optimize their code to better use the available hardware, I wholeheartedly agree. Adobe could take a massive lesson here. The performance of Photoshop, Illustrator, Premiere, AE, etc.. are all terrible and use few of the resources available in contemporary systems. No reason all of them are not heavily multithreaded to leverage all those CPU cores. Premiere is the only one, and AE on renders. And even then, Premiere is clunky and slow compared to Resolve and now especially compared to FCPX 10.3.
I think that if people are unhappy with the new MBP and don't feel it fits their needs, then they should go buy what does. Unfortunately, there are not a whole lot of alternatives. And most of the alternatives have serious flaws or compromises of their own.
This new MBP has already sold via pre-orders more than any other MBP model to date. Prior record holder was the 2012 retina introduction model. Which was also similarly criticized at the time for a handful of reasons. Similar to the 2013 Mac Pro trashcan. Heavily criticized around here and in similar professional circles. And yet, it just keeps on selling.
As for software developers working to streamline and optimize their code to better use the available hardware, I wholeheartedly agree. Adobe could take a massive lesson here. The performance of Photoshop, Illustrator, Premiere, AE, etc.. are all terrible and use few of the resources available in contemporary systems. No reason all of them are not heavily multithreaded to leverage all those CPU cores. Premiere is the only one, and AE on renders. And even then, Premiere is clunky and slow compared to Resolve and now especially compared to FCPX 10.3.
I think that if people are unhappy with the new MBP and don't feel it fits their needs, then they should go buy what does. Unfortunately, there are not a whole lot of alternatives. And most of the alternatives have serious flaws or compromises of their own.
Wayne Morellini
Well-known member
- Joined
- Jan 3, 2008
- Messages
- 6,157
- Reaction score
- 0
- Points
- 0
I like the Cube, but it was maliciously canned. The tube is just like a replacement for that without admiring it.
So Jeff, I take it you are under NDA and can't discuss the concepts I mentioned.
The future, folks, is blindingly evident, but people, like around here, rather stay safe and conventional, rather than safer with much better models of doing things. Brain dead I know, but that is the steam/gravy train of the conventional straight into the loss making gravey shift (cemetery joke).
Of course if you are running multiple applications Adobe's thing you are alledging, isn't too bad. But the reality is that is often not so much the case in a high performance way
So the solution is to run cores at highest performance to energy ratio, then slowly fill each core with processes before activating the next one. Once filled, then increase energy to next step and slowly fill as you activate the next step up on each core as the last is filled (there is a lot of stuff I'm leaving out of here for expedience). And so forth until full speed on all cores is obtained. Of course that is the simplistic view as code is not written that way and high performance is different, and you would load balallnce and relocate processes in so doing. But (sorry a friend rang and I lost my train of thought) with performance you are going to load up variouse individual cores with fewer programs and likely fewer processes. Simplistically you can say put a program on one coreband a plugin transcoder it is using on another core, but unless you are doing something else, like a filter on that core, only one core is likely to get much use, and the other 8 are just going to twiddle there thumbs. But if you are passing a simplified high performance work flow through multiple programs you can use up a number of cores that way (again simplistic). So the modification of the above simplistic scheme, is to load up and step up individual cores in turn I guess out of sync with the others and greatly, and redistribute light processes off of them, and even all processes as they become devoted to a performance task, or a number of non saturating ones.
But the senario above, of a number of high performance programs on individual cores passing to each other is similar to designing your program as a chain of a number of modulrs/processes that stream data through each of them in a pipeline. Now you are not reliant solely on operating system scheduling to optimise performance. The developer offers simple load distribution options to the operating system.
Now, the above scenario, of a number of program modules working I'm a pipeline can be extended to processes working like that (accommodating more cores) and a more complicated scheme of processes for even more cores. Now you are up to Phi territory. More complex? Depends on how you approach it. Load balancing and page passing are crucial in not tanking performance in these things on the Operating system side, as your cache /memory pages are your smallest practical unit of performance (unless you can pass direct transfers from core to core bypassing memory. Do they have that functionality? Sounds like something they wouldn't. Would require direct lines between cores and/or work areas, eben a simple nearest neighbour grid. More than one line allows them to be dynamically assigned between processes/apps, so that when fewer apps/processes the more bandwidth the remaining get, the more processes/apps, the more lines are divided up. Of course this system is transparent to the program, they just pass to destination, and wait for data. Work areas act as buffers to even out flow).
So forth, you get more practical use out of more cores for a bit more programming skill, designing how things work just a bit differently. But GP-GPU's are another thing again, that over competes with this.
Guys, it is obvious that intel is moving Xenon line to Phi.
So Jeff, I take it you are under NDA and can't discuss the concepts I mentioned.
The future, folks, is blindingly evident, but people, like around here, rather stay safe and conventional, rather than safer with much better models of doing things. Brain dead I know, but that is the steam/gravy train of the conventional straight into the loss making gravey shift (cemetery joke).
Of course if you are running multiple applications Adobe's thing you are alledging, isn't too bad. But the reality is that is often not so much the case in a high performance way
So the solution is to run cores at highest performance to energy ratio, then slowly fill each core with processes before activating the next one. Once filled, then increase energy to next step and slowly fill as you activate the next step up on each core as the last is filled (there is a lot of stuff I'm leaving out of here for expedience). And so forth until full speed on all cores is obtained. Of course that is the simplistic view as code is not written that way and high performance is different, and you would load balallnce and relocate processes in so doing. But (sorry a friend rang and I lost my train of thought) with performance you are going to load up variouse individual cores with fewer programs and likely fewer processes. Simplistically you can say put a program on one coreband a plugin transcoder it is using on another core, but unless you are doing something else, like a filter on that core, only one core is likely to get much use, and the other 8 are just going to twiddle there thumbs. But if you are passing a simplified high performance work flow through multiple programs you can use up a number of cores that way (again simplistic). So the modification of the above simplistic scheme, is to load up and step up individual cores in turn I guess out of sync with the others and greatly, and redistribute light processes off of them, and even all processes as they become devoted to a performance task, or a number of non saturating ones.
But the senario above, of a number of high performance programs on individual cores passing to each other is similar to designing your program as a chain of a number of modulrs/processes that stream data through each of them in a pipeline. Now you are not reliant solely on operating system scheduling to optimise performance. The developer offers simple load distribution options to the operating system.
Now, the above scenario, of a number of program modules working I'm a pipeline can be extended to processes working like that (accommodating more cores) and a more complicated scheme of processes for even more cores. Now you are up to Phi territory. More complex? Depends on how you approach it. Load balancing and page passing are crucial in not tanking performance in these things on the Operating system side, as your cache /memory pages are your smallest practical unit of performance (unless you can pass direct transfers from core to core bypassing memory. Do they have that functionality? Sounds like something they wouldn't. Would require direct lines between cores and/or work areas, eben a simple nearest neighbour grid. More than one line allows them to be dynamically assigned between processes/apps, so that when fewer apps/processes the more bandwidth the remaining get, the more processes/apps, the more lines are divided up. Of course this system is transparent to the program, they just pass to destination, and wait for data. Work areas act as buffers to even out flow).
So forth, you get more practical use out of more cores for a bit more programming skill, designing how things work just a bit differently. But GP-GPU's are another thing again, that over competes with this.
Guys, it is obvious that intel is moving Xenon line to Phi.
Brody_B
Well-known member
- Joined
- Mar 8, 2016
- Messages
- 1,145
- Reaction score
- 16
- Points
- 0
- Location
- Los Angeles
- Website
- www.boldpoppy.com
- Moderator
- #608
Jeff Kilgroe
Well-known member
Y'all can whine about dongles and connectivity all you want, but USB-C is the connector moving forward with USB 3.1r2 and newer. Typical Apple -- they jumped to the new thing at the right time, just left a number of people stranded in dongle land by not kludging in a couple of the legacy ports like the PC guys do. Someone hands me a USB flash drive, yeah I'm going to need an adapter. Everything else, just replace the cable. Apple could have included a couple legacy USB 3 Type A ports, at the expense of one of the TB3 ports. Looking to the future, I'm betting I'll be happier with the TB3 port. ...Then again, I'm probably going to dump my new MBP on eBay the second I can order one with more RAM.
Brody_B
Well-known member
- Joined
- Mar 8, 2016
- Messages
- 1,145
- Reaction score
- 16
- Points
- 0
- Location
- Los Angeles
- Website
- www.boldpoppy.com
Calm down Jeff... I posted the video cause it's funny. I have the new MBP on order and a hand full of adapters for it sitting on my desk. No whining here
I have the old top-of-the-line mbp and would never buy the new one. Why? $500 in added dongle costs! Plus any dongle --even one -- is like design leukemia. This is FUBAR boys. Apple is not thinking about users. Rebel and DO NOT BUY! Send the message they need.
Brody_B
Well-known member
- Joined
- Mar 8, 2016
- Messages
- 1,145
- Reaction score
- 16
- Points
- 0
- Location
- Los Angeles
- Website
- www.boldpoppy.com
Apple had more orders in the first week after the announcement than they have ever before, in the history of selling MBP's. I think it might be a bit futile to send a message by not buying one. Probably best to not buy cause you don't want to. P.S. A set of dongles/adapters don't cost anywhere near $500. Just sayin.
- Moderator
- #612
Jeff Kilgroe
Well-known member
Calm down Jeff... I posted the video cause it's funny. I have the new MBP on order and a hand full of adapters for it sitting on my desk. No whining here
Yes, it's funny.
$500? How do you come up with that?
2012-2015 MBP had: 2 x TB2, 2 x USB Type A, 1 x SD, 1 MagSafe connector, headphone jack with digital I/O extension....
2016 MBP has: 4 x TB3, headphone jack with digital I/O extension...
So, to make it identical to the old MBP you need as the most basic setup: 2 x TB3 to TB2 adapters ($25 each), 1 x USB-C multi-adapter w/ SD card slot, micro SD slot, two Type A ports and USB-C pass-thru ($40).
You can get TB3 to TB2 cables and eliminate adapters, about $20 each. The catch is I don't know anyone shipping them just yet, but a few cable vendors have announced them.
I agree that adapters are not the ideal solution. However, within the next year or so, USB-C will be the norm on most new USB hardware. If it's that big of a deal to you, skip this generation and get the next one with more RAM, better CPU and USB 3.5 will most likely be available, not to mention USB-C/TB3 will be a lot more common.. I already have a handful of USB-C thumb drives, most are actually the SanDisk and Lexar that have both A and C connectors on them. Some of us have been using TB3 / USB-C on other systems for a while.
Wayne Morellini
Well-known member
- Joined
- Jan 3, 2008
- Messages
- 6,157
- Reaction score
- 0
- Points
- 0
Antony Newman
Well-known member
I was just about to buy the latest Shure microphone .. looked at the specs, added it to my basket .. then realised that it did NOT use the latest USB-C style connector ... and changed my mind.
Apples move to this connector is going to light a fire under accessory and gadget makers, and help accelerate the industry onto the USB-C standard that is going to be around for a very long time.
If Apple DID include 1 x OLD style non-reversable USB connector ... it would have NOT had the same effect on 3rd party companies.
As it would have been straightforward for Apple to include legacy connectors, this must be an example of Apple willing taking a 'hit' on sales to push the industry forward.
AJ
Apples move to this connector is going to light a fire under accessory and gadget makers, and help accelerate the industry onto the USB-C standard that is going to be around for a very long time.
If Apple DID include 1 x OLD style non-reversable USB connector ... it would have NOT had the same effect on 3rd party companies.
As it would have been straightforward for Apple to include legacy connectors, this must be an example of Apple willing taking a 'hit' on sales to push the industry forward.
AJ
- Thread starter
- #615
Jarek Zabczynski
Well-known member
- Joined
- Apr 3, 2007
- Messages
- 2,410
- Reaction score
- 4
- Points
- 38
- Location
- Los Angeles
- Website
- www.jarek.com
I guess we can expect any update to the MacPro to only have USBC/TB3 ports. At one point the rumor was 10?
Antony Newman
Well-known member
I am not sure which new 'platform' the 10 x USB3 chipset will be used ... But if it is a Purley, I will definitively sit up and take notice.
It would allow 1 or 2 XEONS
And an impressive jump up from 60GB/S (4 channels in the nMP) - which is believe is a memory IO bottleneck (prevents efficient scaling as core count goes up) to something like 12 channels (6 per CPU) which would be insanely good (up to 3x more bandwidth) for HPC work.
But I don't see Purley in an iMac.
With the server world expanding and the PC market shrinking ..... maybe Apple's taste of enterprise will get them fishing in a bigger pool?
Pic above compares the size of the nMP Xeon socket with a Purley socket.
AJ
It would allow 1 or 2 XEONS
And an impressive jump up from 60GB/S (4 channels in the nMP) - which is believe is a memory IO bottleneck (prevents efficient scaling as core count goes up) to something like 12 channels (6 per CPU) which would be insanely good (up to 3x more bandwidth) for HPC work.
But I don't see Purley in an iMac.
With the server world expanding and the PC market shrinking ..... maybe Apple's taste of enterprise will get them fishing in a bigger pool?
Pic above compares the size of the nMP Xeon socket with a Purley socket.
AJ
Last edited:
Wayne Morellini
Well-known member
- Joined
- Jan 3, 2008
- Messages
- 6,157
- Reaction score
- 0
- Points
- 0
He, he. So that's it for the future Phi.
There is something wrong with that picture. The CPU can be integrated onto a memory module which greatly speeds up memory access and width without stepping down process size. Has Intel decided on a memory integerface for their x-point memory? A processor+chipset that basically presents a series of something lower latency and higher speed then current PCIe is needed (even optical) and doable. A dram module like that is also possible. However, conventionally they don't have a memory process suitable for microprocessors yet. But maybe somebody will make a 3D memory process suitable for both using substrate capacitance. Then things will change again.
But watch this sort of socket, interesting things are apparently coming.
There is something wrong with that picture. The CPU can be integrated onto a memory module which greatly speeds up memory access and width without stepping down process size. Has Intel decided on a memory integerface for their x-point memory? A processor+chipset that basically presents a series of something lower latency and higher speed then current PCIe is needed (even optical) and doable. A dram module like that is also possible. However, conventionally they don't have a memory process suitable for microprocessors yet. But maybe somebody will make a 3D memory process suitable for both using substrate capacitance. Then things will change again.
But watch this sort of socket, interesting things are apparently coming.
Wayne Morellini
Well-known member
- Joined
- Jan 3, 2008
- Messages
- 6,157
- Reaction score
- 0
- Points
- 0
I was just about to buy the latest Shure microphone .. looked at the specs, added it to my basket .. then realised that it did NOT use the latest USB-C style connector ... and changed my mind.
Apples move to this connector is going to light a fire under accessory and gadget makers, and help accelerate the industry onto the USB-C standard that is going to be around for a very long time.
If Apple DID include 1 x OLD style non-reversable USB connector ... it would have NOT had the same effect on 3rd party companies.
As it would have been straightforward for Apple to include legacy connectors, this must be an example of Apple willing taking a 'hit' on sales to push the industry forward.
AJ
You guys do know if the device has a USB socket on it you can get a usbx to USBC cable. Other devices you can get an longer extension cable convertor that takes the cludegy covertor point away. Solves some issues. Just buy new memory sticks and an old USB hub to use old sticks at home.
Wayne Morellini
Well-known member
- Joined
- Jan 3, 2008
- Messages
- 6,157
- Reaction score
- 0
- Points
- 0
The thing does need like 10 ports to be sure. Aren't there TB chips to give more ports?
Antony Newman
Well-known member
Intel : KL : Purley
(Not an expert in this area) but I believe that Intel have plenty of spare pins for:
+) 400 GB/sec local high speed memory (even more important if there are 72 cores with 4 threads a piece)
+) 100 Gb fabric (as Intel want to contain the server network to their own 'slightly faster' IP)
+) FGPA accelerators (compression / encryption)
It could be good socket for Apple - especially if Intel ALLOWED a compatible low end server chip ... but I just can't imagine Intel wanting to allow the HPC to have too big a jump in memory bandwidth.
Apple : MacOS : Hurricane compatibiltity (again) : realised that the secure enclave on the MBP 2016 may have an ARM (with a secure enclave) .. but that it is NOT a Hurricane core ... which means that Apple are experimenting with MacOS ARM compatibility - OR that they are threatening Intel with a Switch.
Cables : Shure do offer lightning compatibility (which is a predominantly USB2 connector) in their device.
Hmm : Maybe Apple should drop Lightning and switch to USB-C on their phones too?
AJ
(Not an expert in this area) but I believe that Intel have plenty of spare pins for:
+) 400 GB/sec local high speed memory (even more important if there are 72 cores with 4 threads a piece)
+) 100 Gb fabric (as Intel want to contain the server network to their own 'slightly faster' IP)
+) FGPA accelerators (compression / encryption)
It could be good socket for Apple - especially if Intel ALLOWED a compatible low end server chip ... but I just can't imagine Intel wanting to allow the HPC to have too big a jump in memory bandwidth.
Apple : MacOS : Hurricane compatibiltity (again) : realised that the secure enclave on the MBP 2016 may have an ARM (with a secure enclave) .. but that it is NOT a Hurricane core ... which means that Apple are experimenting with MacOS ARM compatibility - OR that they are threatening Intel with a Switch.
Cables : Shure do offer lightning compatibility (which is a predominantly USB2 connector) in their device.
Hmm : Maybe Apple should drop Lightning and switch to USB-C on their phones too?
AJ