AppleInsider › Forums › Mac Hardware › Future Apple Hardware › AMD Understand the value of OpenCL
New Posts  All Forums:Forum Nav:

AMD Understand the value of OpenCL

post #1 of 26
10/17/12 at 5:38pm
Thread Starter 

By way of Khronos.org: http://www.tomshardware.com/reviews/geforce-gtx-650-ti-benchmark-gk106,3318-14.html

 

 

 

 

 

 

 

 

 

 

 

 

Note: This last one is an outlier but then without checking the source code on how heavy the test relies on specific OpenGL extensions it's hard to judge.

 

 

 

Seeing as OpenCL is getting a huge push in LLVM/Clang it is rather clear that Apple will invest heavier in AMD support than Nvidia who has a long way to go [they want CUDA] and let us hope the Mac Mini and iMac get some hefty bumps on the GPGPU front.

Reply
Reply
post #2 of 26
10/18/12 at 2:11am
I'm not sure Apple cares. I've been promoting AMD for a long time for this very reason, but yet Apple went with NVidia in the latest MBPs. It will be very interesting to see what comes next week. Frankly I really think Apple lost all focus when they delivered the Last Mac Pro update and didn't even bother with a new GPU card. If any machine needed better OpenCL performance it was the Mac Pro.

I have to wonder if their future path involves Intels Xeon Phi as OpenCL accelerators.
Reply
Reply
post #3 of 26
10/18/12 at 11:42am
Quote:
Originally Posted by mdriftmeyer 
Seeing as OpenCL is getting a huge push in LLVM/Clang it is rather clear that Apple will invest heavier in AMD support than Nvidia who has a long way to go [they want CUDA] and let us hope the Mac Mini and iMac get some hefty bumps on the GPGPU front.

On the iMac and Mini side, we'd have to compare mobile GPUs and not just raw performance but power draw too but the 7970M seems to be slightly faster than the desktop 680 for OpenCL and the 680M could be half that:

http://forum.notebookreview.com/gaming-software-graphics-cards/672298-opencl-benchmark-thread.html
http://www.luxrender.net/luxmark/search/search?page=1&benchmark_type=Room&benchmark_mode=ANY&os_type=ANY&dev_count=1&device_unique_name=Apple_SEPARETOR_GeForce%20GTX%20680_SEPARETOR_32

The Mac Pro won't get an update until late 2013 - they might put out upgrade kits but I doubt it.

I wonder if this can be resolved with drivers. The NVidia GPUs outperform the AMD GPUs in almost everything else so maybe Apple's own drivers will perform better with OpenCL.
Reply
Reply
post #4 of 26
10/19/12 at 9:08pm
Quote:
Originally Posted by Marvin View Post

On the iMac and Mini side, we'd have to compare mobile GPUs and not just raw performance but power draw too but the 7970M seems to be slightly faster than the desktop 680 for OpenCL and the 680M could be half that:
I see the most interesting position for AMD right now is with their APUs. Their good OpenCL support make the line very suitable for midrange hardware like the Mini and maybe even the 13" rMBP. Not that I think Apple will use AMD processors but the much better GPU performance relative to Intel makes for some wishful thinking.
Quote:
http://forum.notebookreview.com/gaming-software-graphics-cards/672298-opencl-benchmark-thread.html
http://www.luxrender.net/luxmark/search/search?page=1&benchmark_type=Room&benchmark_mode=ANY&os_type=ANY&dev_count=1&device_unique_name=Apple_SEPARETOR_GeForce%20GTX%20680_SEPARETOR_32
The Mac Pro won't get an update until late 2013 - they might put out upgrade kits but I doubt it.
Where do you get this idea? My impression is that the new Pro will come real early in 2013. I wouldn't be surprised if they come fairly quick after Xeon Phi.
Quote:
I wonder if this can be resolved with drivers. The NVidia GPUs outperform the AMD GPUs in almost everything else so maybe Apple's own drivers will perform better with OpenCL.
When drivers on Linux perform better than Mac drivers we have a problem.
Reply
Reply
post #5 of 26
10/19/12 at 10:45pm

Look at the supremely fast X-Gene ARMv8 64-bit processor. If Apple can do the A6, they can certainly match the X-Gene.

Add plenty of Rogue GPU cores for speedy OpenCL and OpenGL. Custom memory controllers delivering extreme memory bandwidth.

 

Wouldn't you want to build your own CPUs instead of pay Intel and Nvidia their exorbitant prices? You could deliver unheard of performance and the chips would be super cheap.

 

They just need Rosetta again for Intel --> ARMv8 translation. QuickTransit (Rosetta) is now owned by IBM, who would be very happy to take on the project and license it to Apple.

Although Apple would have enough in house expertise to do their own real-time translator now. Their LLVM expertise would be perfect.

Reply
Reply
post #6 of 26
10/20/12 at 5:33am
Quote:
Originally Posted by wizard69 
My impression is that the new Pro will come real early in 2013. I wouldn't be surprised if they come fairly quick after Xeon Phi.

Tim Cook: "Although we didn’t have a chance to talk about a new Mac Pro at today’s event, don’t worry as we’re working on something really great for later next year"

If he'd meant early next year, he would have said:

Tim Cook: "Although we didn’t have a chance to talk about a new Mac Pro at today’s event, don’t worry as we’re working on something really great for early next year"

They won't use Sandy Bridge Xeons at that stage when Ivy Bridge E is coming in the 2nd half of the year:

http://www.techspot.com/news/47849-ivy-bridge-e-delayed-until-second-half-of-2013.html

That would be a major architecture change within 6 months.

I think it would be better for Apple to start designing around a single CPU. It says here Haswell-EP will be at least 10-cores:

http://www.chiphell.com/thread-497401-1-1.html

Ivy Bridge EP will apparently have 10-12-cores too:

http://vr-zone.com/articles/exclusive-update--ivy-bridge-e-ep-die-to-have-12-cores-not-10-/17314.html

There will be so few people buying a 20-core/40-thread machine that it wouldn't be worth making a DP machine. If they can use a variant of Intel's co-processor or their own ARM co-processor, even better but it might not be needed.

AMD will have the Sea Islands GPUs by then with unified memory.

http://www.tweaktown.com/news/25944/amd_radeon_hd_8970_8950_to_launch_next_year_packing_5_1_billion_transistors_each/index.html

If they ditch the optical bay and the 2nd CPU and get either a 6970 or 6990 in there in one PCI slot with no other expansion ports, they can probably run the machine on a 500-600W PSU.

10/12-core single Ivy-Bridge EP CPU 150W
Radeon 8970 ~200W
Up to 64GB DDR4 RAM
256GB SSD
20Gbps Falcon Ridge TB controller with 4-6 ports

Smaller form factor of course and zero-config TB chaining (up to 100m with optical cables). Similar prices to the single CPU models that exist now.
Reply
Reply
post #7 of 26
10/20/12 at 6:29am
Quote:
Originally Posted by Marvin View Post


Tim Cook: "Although we didn’t have a chance to talk about a new Mac Pro at today’s event, don’t worry as we’re working on something really great for later next year"
 

 

I'm not saying you're wrong, but that solitary quote is pretty thin.  Sometimes people speak/write inarticulately.  That could just as easily mean "something really great for later.  Next year." with "later" meaning "next year" either early or late.

A good brain ain't diddly if you don't have the facts
Reply
A good brain ain't diddly if you don't have the facts
Reply
post #8 of 26
10/20/12 at 6:50am
Quote:
Originally Posted by Marvin View Post


Tim Cook: "Although we didn’t have a chance to talk about a new Mac Pro at today’s event, don’t worry as we’re working on something really great for later next year"
If he'd meant early next year, he would have said:
Tim Cook: "Although we didn’t have a chance to talk about a new Mac Pro at today’s event, don’t worry as we’re working on something really great for early next year"
They won't use Sandy Bridge Xeons at that stage when Ivy Bridge E is coming in the 2nd half of the year:
http://www.techspot.com/news/47849-ivy-bridge-e-delayed-until-second-half-of-2013.html
That would be a major architecture change within 6 months.
I think it would be better for Apple to start designing around a single CPU. It says here Haswell-EP will be at least 10-cores:
http://www.chiphell.com/thread-497401-1-1.html
Ivy Bridge EP will apparently have 10-12-cores too:
http://vr-zone.com/articles/exclusive-update--ivy-bridge-e-ep-die-to-have-12-cores-not-10-/17314.html
There will be so few people buying a 20-core/40-thread machine that it wouldn't be worth making a DP machine. If they can use a variant of Intel's co-processor or their own ARM co-processor, even better but it might not be needed.
AMD will have the Sea Islands GPUs by then with unified memory.
http://www.tweaktown.com/news/25944/amd_radeon_hd_8970_8950_to_launch_next_year_packing_5_1_billion_transistors_each/index.html
If they ditch the optical bay and the 2nd CPU and get either a 6970 or 6990 in there in one PCI slot with no other expansion ports, they can probably run the machine on a 500-600W PSU.
10/12-core single Ivy-Bridge EP CPU 150W
Radeon 8970 ~200W
Up to 64GB DDR4 RAM
256GB SSD
20Gbps Falcon Ridge TB controller with 4-6 ports
Smaller form factor of course and zero-config TB chaining (up to 100m with optical cables). Similar prices to the single CPU models that exist now.

 

Sexy beast machine. :)

 

Lemon Bon Bon.

You know, for a company that specializes in the video-graphics market, you'd think that they would offer top-of-the-line GPUs...[/
Reply
You know, for a company that specializes in the video-graphics market, you'd think that they would offer top-of-the-line GPUs...[/
Reply
post #9 of 26
10/20/12 at 6:59am

All roads converging on Haswell being a very tasty year for Macs 'Pro' and iMac.  (Hopefully retina as well for iMac in the 3k by 2k res'?)

 

...and those Sea Island GPUs...

 

...oh...bay...bee.

 

 

I'd take that smaller form factor for the Pro too.  But they could seriously do with a tower in the 1300-1500 uk £ range.  The current Quad £2k price is an absolute joke then you add a 27 inch 'badge' monitor for another grand for medicore 'so last, last, last, last' year performance...with a crap GPU...and laughable ram...

 

Good post though, Marv'.  Finger on the pulse.*  Always a great read. :)

 

 

Lemon Bon Bon.

 

 

Quote:
Smaller form factor of course and zero-config TB chaining (up to 100m with optical cables). Similar prices to the single CPU models that exist now.
You know, for a company that specializes in the video-graphics market, you'd think that they would offer top-of-the-line GPUs...[/
Reply
You know, for a company that specializes in the video-graphics market, you'd think that they would offer top-of-the-line GPUs...[/
Reply
post #10 of 26
10/20/12 at 7:14am
Quote:
Originally Posted by retroneo View Post

Look at the supremely fast X-Gene ARMv8 64-bit processor. If Apple can do the A6, they can certainly match the X-Gene.

Add plenty of Rogue GPU cores for speedy OpenCL and OpenGL. Custom memory controllers delivering extreme memory bandwidth.

 

Wouldn't you want to build your own CPUs instead of pay Intel and Nvidia their exorbitant prices? You could deliver unheard of performance and the chips would be super cheap.

 

They just need Rosetta again for Intel --> ARMv8 translation. QuickTransit (Rosetta) is now owned by IBM, who would be very happy to take on the project and license it to Apple.

Although Apple would have enough in house expertise to do their own real-time translator now. Their LLVM expertise would be perfect.

Well.  We can't rule anything out with Apple.  They incinerated the PPC Alliance / PPC Chips when...it stalled and outlived it's usefulness and couldn't deliver performance per watt performance.

 

If you look at the A6 performance in a phone...it's astonishing.  It's more powerful than some Macs that shipped not so long ago...  (Thinks about PPCs...)

 

If Arm can pack a punch with a 64 bit processor...Apple adapts it...and packs in some Rogue Cores...  Who knows what could happen.  Apple like control.  And if they can save on vendor prices and design the whole widget...they've shown with the iPhone/iPad what their ultimate ambitions are.

 

I think the Mac will be irrevocably drawn along by iOS and the iDevices.  We're in that transition..?

 

In short.  I think you're post is pretty much on the money.  Not a 'too far' distant possible future.

 

Lemon Bon Bon.

You know, for a company that specializes in the video-graphics market, you'd think that they would offer top-of-the-line GPUs...[/
Reply
You know, for a company that specializes in the video-graphics market, you'd think that they would offer top-of-the-line GPUs...[/
Reply
post #11 of 26
10/20/12 at 7:25am
Quote:
Originally Posted by retroneo View Post

Look at the supremely fast X-Gene ARMv8 64-bit processor. If Apple can do the A6, they can certainly match the X-Gene.
Add plenty of Rogue GPU cores for speedy OpenCL and OpenGL. Custom memory controllers delivering extreme memory bandwidth.
I would hope that Apple would go the same way AMD is going and merge the GPU with the CPU. That is in the sense that they are equals with respect to memory access.
Quote:
Wouldn't you want to build your own CPUs instead of pay Intel and Nvidia their exorbitant prices? You could deliver unheard of performance and the chips would be super cheap.
Unheard of performance? Seriously do you think it is that easy to beat Intel at the performance game. Apple is doing extremely well with A6 it I'm left with the impression that Intel has yet to get serious about ATOM.
Quote:
They just need Rosetta again for Intel --> ARMv8 translation. QuickTransit (Rosetta) is now owned by IBM, who would be very happy to take on the project and license it to Apple.
Although Apple would have enough in house expertise to do their own real-time translator now. Their LLVM expertise would be perfect.

The real answer is native apps, iPhone or iOS clearly demonstrates this in respect to Java. The other issue is that an Intel processor is not easy to emulate at all, performance will suck no matter what you do software wise with an emulator.

I really have nothing against ARM devices considering the fact that I own an iPad and iPhone. The problem with laptops is that i86 and virtual machine technology lets me support all of the legacy crap that is out there. You simply can't do that on ARM with the level of confidence one has with i86. Now many seem to think you can just say screw legacy support but if that is your job you really don't have that choice.
Reply
Reply
post #12 of 26
10/20/12 at 8:05am
Quote:
Originally Posted by Marvin View Post

Tim Cook: "Although we didn’t have a chance to talk about a new Mac Pro at today’s event, don’t worry as we’re working on something really great for later next year"
I didn't interpret it that way at all. Rather something like "later, next year".
Quote:
If he'd meant early next year, he would have said:
Tim Cook: "Although we didn’t have a chance to talk about a new Mac Pro at today’s event, don’t worry as we’re working on something really great for early next year"
Well no, he could have meant several things by that statement. It is the translation of speech into text that is the problem here.
Quote:
They won't use Sandy Bridge Xeons at that stage when Ivy Bridge E is coming in the 2nd half of the year:
http://www.techspot.com/news/47849-ivy-bridge-e-delayed-until-second-half-of-2013.html
Yeah well I have no confidence in Intels projected delivery dates considering what happened to Sandy Bridge E. Besides I suspect that they will not be using that hardware at all. Intel is apparently building up a family of devices around the Xeon Phi branding thus I expect a move to that platform. Frankly preparations for a major architecture shift could explain the joke of the last Mac Pro "upgrade".
Quote:
That would be a major architecture change within 6 months.
I think it would be better for Apple to start designing around a single CPU. It says here Haswell-EP will be at least 10-cores:
http://www.chiphell.com/thread-497401-1-1.html
Ivy Bridge EP will apparently have 10-12-cores too:
http://vr-zone.com/articles/exclusive-update--ivy-bridge-e-ep-die-to-have-12-cores-not-10-/17314.html
It will be interesting to see what intel actually delivers into a degenerate market. I suspect that Intel, like AMD, will be shuffling plans and doing whatever to adapt to the rapidly changing market conditions. So I'm of the mind that I will believe it when I see it.
Quote:
There will be so few people buying a 20-core/40-thread machine that it wouldn't be worth making a DP machine. If they can use a variant of Intel's co-processor or their own ARM co-processor, even better but it might not be needed.
Well it depends upon whom Apple is targeting. There is a segment of users that will take whatever power you can throw at them. The other problem with high core count chips is that reaching maximum performance isn't always possible as congestion in the various data paths throttle cores.
Quote:
AMD will have the Sea Islands GPUs by then with unified memory.
http://www.tweaktown.com/news/25944/amd_radeon_hd_8970_8950_to_launch_next_year_packing_5_1_billion_transistors_each/index.html
Lets hope hat AMD is around long enough to realize these chips.
Quote:
If they ditch the optical bay and the 2nd CPU and get either a 6970 or 6990 in there in one PCI slot with no other expansion ports, they can probably run the machine on a 500-600W PSU.
10/12-core single Ivy-Bridge EP CPU 150W
Radeon 8970 ~200W
Up to 64GB DDR4 RAM
256GB SSD
20Gbps Falcon Ridge TB controller with 4-6 ports
Smaller form factor of course and zero-config TB chaining (up to 100m with optical cables). Similar prices to the single CPU models that exist now.

That sounds really good though admittedly a bit out of my performance zone needs. This is where my frustration with Apple comes in, I really want a 4-6 core machine, running at desktop speeds, for under $1500. Of course the minute you bring up XMac people think tower and as such competition with the Mac Pro. I on the other hand don't see a tower as the form I'm looking for. Rather a fat Mini, that is a Mini that is a bit taller to handle a power supply and cooling system for a 50-75 watt processor and a reasonable GPU.

The thing is as the Mac Pro morphs into a machine with far more processor cores (something we both agree on) the gap between the Pro and the other desktop, the Mini, just gets wider and wider. Apples lack of a headless midrange desktop just defies explanation.

In any event this then comes back to the discussion of OpenCL and AMD chips. If Apple continues to castrates the Minis CPU performance the least they could do is to upgrade to an AMD APU that gives us much better GPU performance. Minis CPU performance. Is so pathetic that it could be matched by an AMD APU for the most part and their advanced OpenCL support would actually run many apps faster than the Intel solution.

That is a solution for the Mini. For the Mac Pro I really don't know what is up. My speculation surrounds Intels Xeon Phi due to its pending release late this year. How Apple will integrate Phi into its OpenCL and Grand Central Dispatch infrastructure is unknown at this time. They may still try to leverage GPU compute or they may give up on it. Hard to tell. The problem becomes an issue of power, put a fast GPU, fast CPU and a Phi coprocessor in the box and you start to talk about lots of watts. Even with the shrunk chips of today such a box would be very power hungry. Intel was talking 300 watts at one time for the PHi coprocessor card. A Phi main processor with a built in supercomputer networking interface won't be cool running either. In a nut shell I see the Mac Pro remaining an expensive high performance machine. However next year could deliver far more bang for the buck then people can imagine.
Reply
Reply
post #13 of 26
10/20/12 at 8:20am
Quote:
Originally Posted by Lemon Bon Bon. View Post

All roads converging on Haswell being a very tasty year for Macs 'Pro' and iMac.  (Hopefully retina as well for iMac in the 3k by 2k res'?)

...and those Sea Island GPUs...

...oh...bay...bee.


I'd take that smaller form factor for the Pro too.  But they could seriously do with a tower in the 1300-1500 uk £ range.
This is really the great mystery, where is this missing midrange Mac? By the way a tower isn't required at all. In fact it would be a bit foolish to jam modern technology into a tower format.
Quote:
 The current Quad £2k price is an absolute joke then you add a 27 inch 'badge' monitor for another grand for medicore 'so last, last, last, last' year performance...with a crap GPU...and laughable ram...
Pathetic would be the word. I can almost understand the CPU upgrade if they have an overhaul planned for next year. But why stay with a 3 year old video card that plugs in a slots. It just smacks of incompetence.
Quote:
Good post though, Marv'.  Finger on the pulse.*  Always a great read. 1smile.gif


Lemon Bon Bon.


Hey who knows maybe Apple will deliver hardware next week that relaxes my anger with respect to the desktop stagnation. Debut an iMac that doesn't require a magician to service and I might even consider it as a midrange machine. Or simply debut a Mini with really good performance.

Speaking of the Mini, All Apple really needs to do there is to boost the machine to allow for an Intel processor in the 55 watt range, something a bit better than mobile hardware. And provide space for one of the blade SSDs out of the MBP (while maintaining the HD slots). This would not be the entry device but maybe the so called server model. In other words Apple give us something for our money at the top end. That something being processor performance.
Reply
Reply
post #14 of 26
10/20/12 at 8:42am
Quote:
Originally Posted by Flounder 
I'm not saying you're wrong, but that solitary quote is pretty thin.  Sometimes people speak/write inarticulately.  That could just as easily mean "something really great for later.  Next year." with "later" meaning "next year" either early or late.

It could but if you assume they will launch early next year, they'd have to go with Sandy Bridge EP. That doesn't make sense when they could have done it this year and Ivy Bridge EP will be out just a few months later. What reason would they have for a minor update in June after 2 years and then do a SB update 7 months later? In the context of the available hardware, 'later next year' fits when it means later on in the year.

While 'later' can be misinterpreted, Ivy Bridge EP's launch date can't be and I strongly believe that either the next MP will be based on Ivy bridge or they will put a Xeon in the iMac.
Quote:
Originally Posted by Lemon Bon Bon 
All roads converging on Haswell being a very tasty year for Macs 'Pro' and iMac. (Hopefully retina as well for iMac in the 3k by 2k res'?)

It wouldn't be surprising to me if the Retina iMac was in fact the 'iMac Pro' with a Xeon. 4K displays are ridiculously expensive. If Apple comes along with one at $2499 with say a 95W 6/8-core Ivy Bridge Xeon chip and an 8970M, that's going to work for a lot of people.

Here's what Steve Jobs said about the desktop in 1995:

"The desktop computer industry is dead. Innovation has virtually ceased. Microsoft dominates with very little innovation. That's over. Apple lost. The desktop market has entered the dark ages, and it's going to be in the dark ages for the next 10 years, or certainly for the rest of this decade.
It's like when IBM drove a lot of innovation out of the computer industry before the microprocessor came along. Eventually, Microsoft will crumble because of complacency, and maybe some new things will grow. But until that happens, until there's some fundamental technology shift, it's just over.

The most exciting things happening today are objects and the Web. The Web is exciting for two reasons. One, it's ubiquitous. There will be Web dial tone everywhere. And anything that's ubiquitous gets interesting. Two, I don't think Microsoft will figure out a way to own it. There's going to be a lot more innovation, and that will create a place where there isn't this dark cloud of dominance."

This was of course before the iMac but the same statement can be reapplied. Where's the innovation in the Mac Pro for the target audience? I wouldn't say Apple is complacent with the MP but they are indifferent towards it because it hasn't achieved great success and it can't. They can't own the workstation tower market any more and if they did, not enough people would care.

The exciting things today are happening with mobile devices because they are ubiquitous - this was the shift that has unseated Microsoft's dominance. This is great for hardware manufacturers, OS developers, software developers, publishers and consumers.

Every product Apple makes has an area of emphasis. When the demand for that area of emphasis is so small, why make the product? The demand for power is still there but there's a lot less demand for the highest power. People are content playing games at medium quality, 720p H.264 encoding at double real-time is acceptable, even resource-intensive productive apps run comfortably. The only demand left is for deadline-restricted rendering and computation. Once they get the right CPU/GPU setup and software adapts to it, this heterogeous computing will move people further towards the lower machines.

Apple believes in AIO. The Mini and MP are the only machines they make that aren't. They can't make an iMac cheap enough to take away the Mini's area of emphasis as both server and low cost Mac but they can make a fast enough iMac to take away the Mac Pro's area of emphasis. Yes there's expansion but I believe the place for expansion is on the outside.
Reply
Reply
post #15 of 26
10/20/12 at 12:38pm
Quote:
Originally Posted by Marvin View Post


It could but if you assume they will launch early next year, they'd have to go with Sandy Bridge EP. That doesn't make sense when they could have done it this year and Ivy Bridge EP will be out just a few months later. What reason would they have for a minor update in June after 2 years and then do a SB update 7 months later? In the context of the available hardware, 'later next year' fits when it means later on in the year.
Quote:
Honestly I think you are trying to hard here. First, I don't think Apple really cares when Intel releases any of its XEON chips. Just look at the current Mac pro and the rather late updates it has gotten. Beyond that they might just give up on the current XEON chips for the XEON Phi which is to debut late Nov. or December if rumors are to be believed.
While 'later' can be misinterpreted, Ivy Bridge EP's launch date can't be and I strongly believe that either the next MP will be based on Ivy bridge or they will put a Xeon in the iMac.
Quote:
Well considering what has happened with Sandy Bridge E you can't be sure that any date will hold for the Ivy Bridge based XEONs.
It wouldn't be surprising to me if the Retina iMac was in fact the 'iMac Pro' with a Xeon. 4K displays are ridiculously expensive. If Apple comes along with one at $2499 with say a 95W 6/8-core Ivy Bridge Xeon chip and an 8970M, that's going to work for a lot of people.
Quote:
Given just a few design changes that could work for me, but then again I've never really been drawn to the Mac Pro. Simply put it has never been properly priced for the type of workstation the Pro is.
Here's what Steve Jobs said about the desktop in 1995:
"The desktop computer industry is dead. Innovation has virtually ceased. Microsoft dominates with very little innovation. That's over. Apple lost. The desktop market has entered the dark ages, and it's going to be in the dark ages for the next 10 years, or certainly for the rest of this decade.
Quote:
Very interesting and timely as Apple is really killing the desktop computer right now. However they are not completely dead and never will be. The other thing is it has just about been the ten years projected and frankly technology has changed enough that Apple could do real innovation on the desktop. They do it for the laptop market so why not spend a little cash to innovate on the desktop? This is what is frustrating to me, they really haven't even tried to introduce something truly new in years. The Mini was about the last new machine they have introduced. The only reason the iMac sells so well is that customers have little choice if they want to stay on the Mac platform with a reasonably competent machine.
It's like when IBM drove a lot of innovation out of the computer industry before the microprocessor came along. Eventually, Microsoft will crumble because of complacency, and maybe some new things will grow. But until that happens, until there's some fundamental technology shift, it's just over.
Quote:
To look at it another way MS is crumbling and the time is right for Apple to step up to the plate with hardware that shows a little innovation. There is no doubt in my mind that Steve was right about the PC industry and the lack of innovation but that does't excuse the lack of innovation form Apple. Apple really has no excuse, they don't have Mircosoft holding them back, at this point even Intel can't hold them back.
The most exciting things happening today are objects and the Web. The Web is exciting for two reasons. One, it's ubiquitous. There will be Web dial tone everywhere. And anything that's ubiquitous gets interesting. Two, I don't think Microsoft will figure out a way to own it. There's going to be a lot more innovation, and that will create a place where there isn't this dark cloud of dominance."
Quote:
Honestly I think the web at times is overblown. Yes it is important, just as any form of communications is important, but it isn't the end all some make it out to be. The biggest problem with the web is that it is turning into a huge pay to play sand box which many people will reject. This is especially the case as rates continue to increase and service levels decrease, it is just an expense many can live without.
This was of course before the iMac but the same statement can be reapplied. Where's the innovation in the Mac Pro for the target audience? Quote:
This is a good question which I think is where much of the frustration about the Mac Pro comes from. The lack of innovation or just plain attention to the product is appalling.
I wouldn't say Apple is complacent with the MP but they are indifferent towards it because it hasn't achieved great success and it can't. They can't own the workstation tower market any more and if they did, not enough people would care.
Quote:
Actually I think the time is rip for some organization like Apple to own the workstation market. If not own certainly lead it into the future.
The exciting things today are happening with mobile devices because they are ubiquitous - this was the shift that has unseated Microsoft's dominance. This is great for hardware manufacturers, OS developers, software developers, publishers and consumers.
Quote:
Exciting yes. Growing Yes. Massive yes. Everything you say is true but that doesn't mean desktop hardware will cease to exist. Nor does it mean that innovation has to stop there. If anything it means that innovation has to be considered to keep the market fresh. Frankly it is almost like Apple here is having a KODAK moment. I don't mean that in the good way of taking pictures either. In a nut shell they are so wrapped up in their current success they are ignoring the bigger picture.
Every product Apple makes has an area of emphasis. When the demand for that area of emphasis is so small, why make the product? The demand for power is still there but there's a lot less demand for the highest power. People are content playing games at medium quality, 720p H.264 encoding at double real-time is acceptable, even resource-intensive productive apps run comfortably. The only demand left is for deadline-restricted rendering and computation. Once they get the right CPU/GPU setup and software adapts to it, this heterogeous computing will move people further towards the lower machines.
Quote:
Actually I have to disagree on many points here. It comes down to a couple of things. One is that time is money and for professional use there are many apps where cutting production time in half justifies new hardware. The other is that as computational power increases computational complexity increases, thus you never really have all the computational power you would like. High performance workstations, at reasonable prices, just mean that apps get moved to lower ranking employees desks. This is currently what is happening in design automation where 3D and simulation are the new "CAD".
Apple believes in AIO. The Mini and MP are the only machines they make that aren't. They can't make an iMac cheap enough to take away the Mini's area of emphasis as both server and low cost Mac but they can make a fast enough iMac to take away the Mac Pro's area of emphasis. Quote:
Nope! They can never make the iMac as fast as the Mac Pro for any given technology node. Frankly they can't even make an iMac that is competitive with lower cost desktop hardware from other manufactures.
Yes there's expansion but I believe the place for expansion is on the outside.

 

Yeah well I'm not one to budge on the idea that some things best reside inside the box. Thunderbolt will change that some, but it is also a severe limitation over internal expansion.
Reply
Reply
post #16 of 26
10/21/12 at 12:22am
Quote:
Originally Posted by wizard69 View Post
When drivers on Linux perform better than Mac drivers we have a problem.

 

Not really.  Most high performance computing is done on Linux systems, so of course drivers are going to be geared towards this fact. 

 

Apple makes consumer products, Linux is the antithesis of a consumer product (although Ubuntu is close to being usable for the average computer user).

Reply
Reply
post #17 of 26
10/21/12 at 1:00am
Thread Starter 
Quote:
Originally Posted by wizard69 View Post


I see the most interesting position for AMD right now is with their APUs. Their good OpenCL support make the line very suitable for midrange hardware like the Mini and maybe even the 13" rMBP. Not that I think Apple will use AMD processors but the much better GPU performance relative to Intel makes for some wishful thinking.
Where do you get this idea? My impression is that the new Pro will come real early in 2013. I wouldn't be surprised if they come fairly quick after Xeon Phi.
When drivers on Linux perform better than Mac drivers we have a problem.

 

Where do you get this idea from?


Firstly, if we're discussing OpenGL lets just throw the Linux into the trash heap. The Desktop Environment is now approaching OpenGL ES 2.0 profile status. Meanwhile, OS X 10.8 is on the full OpenGL 3.2.x minimum to run the Desktop Environment.

 

AMD is working heavily with Apple and the LLVM/Clang Community to make LLVM/Clang the compiler platform solution for their OpenCL/OpenGL tools, along with Apple. A lot of performance gains are very nearing public consumption.

 

If you are discussing the OpenCL profiles available for OS X, AMD, Nvidia and Intel the two mature profiles for CPU/GPGPU acrossed the board are from AMD and Apple.

 

Nvidia is behind and so is Intel.

 

Intel is fully committed to OpenCL and are also in on the new OpenCL SPIR work for LLVM IR that will be available for LLVM/Clang 3.2 which is 4.x+ in XCode.

 

Release of the LLVM/Clang 3.2 is slated for Mid December.

 

The Mac Pro is a 1KW Power System. There will be nothing < 1KW moving forward when attempting to upgrade their designs to allow for larger GPGPU options from third party vendors. Dual 7970 AMD GPGPUs have a minimum 850 W Recommended Power Supply requirement and that's not in a system that allows the backplane Apple's does, it's many other features alone will have one wanting a 1KW minimum.

 

More to the point,

 

Apple in the past few weeks received some key patents regarding System Agnostic processing whether offloading onto a CPU, a DSP, a GPGPU using GCD on a Workstation, Laptop, Desktop, Embedded Solution, you name it. The patent was filed over 5 years ago.

 

http://www.freepatentsonline.com/8286196.html

 

It distinctively includes support for SLI and Crossfire in their filing.

 

Download the PDF. It is very illuminating.

 

I don't see Apple scaling down their Workstation beast into a smaller form factor, unless they planned to offer a Mac Pro and a Mac Pro Ultra where the Pro Ultra is the traditional case design and the Pro is a Mid-tower Case design.

 

In the end, I see Apple attempting to be taken more serious in the Workstation market and taking their design patents and patents into providing a more energy efficient with a higher energy total footprint option for expansion supporting PCI-E 3.x default.

 

I'm sure noise will be reduced in these new designs, better efficiencies achieved with higher performing ICB components, etc. And I imagine they will want an ECC 128/256GB possibility in their designs internally. Now whether we see that as BTO for consumption is a different matter.

 

I think the delays are more on what Apple is bringing to the table and not so much what Apple has to do to wait for others to bring to the table.

 

The AMD lines of GPGPUs are just superior for OpenCL solutions and the lead is going to widen moving forward. Apple is committed to OpenCL/OpenGL throughout their OS.

Reply
Reply
post #18 of 26
10/21/12 at 8:50am
Quote:
Originally Posted by mdriftmeyer 
Meanwhile, OS X 10.8 is on the full OpenGL 3.2.x minimum to run the Desktop Environment.

They can get OpenGL 4 in Linux though:

http://www.youtube.com/watch?v=ZojsR4zwjt8

There's not a hope for that in OS X until Apple updates the drivers. Maybe they should open source the drivers if they haven't so that AMD/NVidia/other can put out custom versions.
Quote:
Originally Posted by mdriftmeyer 
The Mac Pro is a 1KW Power System. There will be nothing < 1KW moving forward when attempting to upgrade their designs to allow for larger GPGPU options from third party vendors. Dual 7970 AMD GPGPUs have a minimum 850 W Recommended Power Supply requirement and that's not in a system that allows the backplane Apple's does, it's many other features alone will have one wanting a 1KW minimum.

The Mac Pro only allows 300W on the PCI slots so you can only have 1 powerful GPU.
Quote:
Originally Posted by mdriftmeyer 
MI don't see Apple scaling down their Workstation beast into a smaller form factor, unless they planned to offer a Mac Pro and a Mac Pro Ultra where the Pro Ultra is the traditional case design and the Pro is a Mid-tower Case design.

In the end, I see Apple attempting to be taken more serious in the Workstation market and taking their design patents and patents into providing a more energy efficient with a higher energy total footprint option for expansion supporting PCI-E 3.x default.

I'm sure noise will be reduced in these new designs, better efficiencies achieved with higher performing ICB components, etc. And I imagine they will want an ECC 128/256GB possibility in their designs internally. Now whether we see that as BTO for consumption is a different matter.

I think removing the optical bay automatically allows a smaller form factor. If they keep the dual processors and slots, it won't shrink by much but I don't see how they can easily support slots and Thunderbolt. They have to put the dedicated GPU framebuffer into an IGP but they can't as Xeons don't have them. This means either abandoning TB, which is not a great idea because it won't support the TB Cinema Displays or ditch the slots except for internal use e.g they'd stick a single GPU inside the machine with no outward ports and save space.

They'd need 16 RAM slots for 256GB RAM so I don't see that happening. I don't think they need to go beyond 6 slots for up to 96GB RAM and this can be used with one CPU. They are already quite cramped:

http://www.apple.com/uk/macpro/design.html#memory

One day, storage and RAM should be combined anyway.

I doubt they're trying to be taken more seriously in the workstation market. When the Mini went without an update for 1.5 years, people were dancing on its empty grave. The Mac Pro will have gone without a proper update for pretty much 3 years next June and the conclusion is that they're coming up with the biggest, baddest and best workstation anyone will have ever seen.

The iMac can hold a 95W chip. A 10-core 95W IB/Haswell Xeon would be just fine.

Wouldn't the following machine make for a great workstation?:

iMac Pro
10-core 2.5GHz Xeon 95W
64GB RAM
256GB SSD with BTO up to 768 + 2TB HDD
Retina 27" display (4K QHD)
2GB Radeon 8970M with unified memory
2x 20Gbps Thunderbolt, 4x 5Gbps USB 3
$2499

They will even be able to get their hands on a 70W 10-core in Q3 2013:

http://pcper.com/news/Processors/Intel-Planning-10-core-Xeon-E5-2600-V2-Ivy-Bridge-EP-CPU

Like I say, it's either an IB MP in Q3 2013 or it's an iMac Pro. I personally think at that point in time, an iMac Pro would make more sense for Apple.
Reply
Reply
post #19 of 26
10/21/12 at 12:36pm
Quote:
Originally Posted by mdriftmeyer View Post

Where do you get this idea from?
Things change, Intel and AMD, to a lesser extent, have been throwing a lot of effort into their Linux drivers.
Quote:

Firstly, if we're discussing OpenGL lets just throw the Linux into the trash heap. The Desktop Environment is now approaching OpenGL ES 2.0 profile status. Meanwhile, OS X 10.8 is on the full OpenGL 3.2.x minimum to run the Desktop Environment.

AMD is working heavily with Apple and the LLVM/Clang Community to make LLVM/Clang the compiler platform solution for their OpenCL/OpenGL tools, along with Apple. A lot of performance gains are very nearing public consumption.
Err if that was the case would Apple be pursuing NVidia? Now AMD is all in with OpenCL support, that I freely admit, but is that effort going to support Mac OS? I really think AMD sees a bigger win in Linux and Windows support.

As a side note LLVM/CLang has become huge. Many companies are working with Apple on its development even if they have no direct interest in support Apple hardware.
Quote:
If you are discussing the OpenCL profiles available for OS X, AMD, Nvidia and Intel the two mature profiles for CPU/GPGPU acrossed the board are from AMD and Apple.
AMD has committed to OpenCL in a big way, I don't dispute that. However if Apple doesn't ship hardware with AMD inside it doesn't mean much.

I think the point that many mis, that maybe I didn't make clear, was that because of AMDs commitment to OpenCL their APUs would actually be good solutions for Apples lower end machines. Things like the Mini and even a retina 13"MBP. A significantly faster integrated GPU that also supports OpenCL well, would result in very balanced performance in these machines. In a nut shell AMD has more to offer Apples lower end machines than Intel does.
Quote:
Nvidia is behind and so is Intel.

Intel is fully committed to OpenCL and are also in on the new OpenCL SPIR work for LLVM IR that will be available for LLVM/Clang 3.2 which is 4.x+ in XCode.

Release of the LLVM/Clang 3.2 is slated for Mid December.

The Mac Pro is a 1KW Power System. There will be nothing < 1KW moving forward when attempting to upgrade their designs to allow for larger GPGPU options from third party vendors. Dual 7970 AMD GPGPUs have a minimum 850 W Recommended Power Supply requirement and that's not in a system that allows the backplane Apple's does, it's many other features alone will have one wanting a 1KW minimum.
Maybe I've confused you here, I have no problem with Apple offering a workstation class machine. It is Apples lack of a midrange machine that bothers me.

In the case of a new design Mac Pro I was trying to point out that power usage isn't likely to decline much at all. If Apple plugs in a Xeon Phi co processor that can easily be 300 watts right there. In otherwords I don't see Mac Pro power requirements dropping if they deliver what I think is coming. As a side bar the 300+ watts for a Phi coprocessor is for basically the prototype processors which could be greatly reduced by a process shrink. In the end I just don't see the power demands of a Mac Pro shrinking all that much even if the box does.
Quote:
More to the point,

Apple in the past few weeks received some key patents regarding System Agnostic processing whether offloading onto a CPU, a DSP, a GPGPU using GCD on a Workstation, Laptop, Desktop, Embedded Solution, you name it. The patent was filed over 5 years ago.

http://www.freepatentsonline.com/8286196.html

It distinctively includes support for SLI and Crossfire in their filing.

Download the PDF. It is very illuminating.

I don't see Apple scaling down their Workstation beast into a smaller form factor, unless they planned to offer a Mac Pro and a Mac Pro Ultra where the Pro Ultra is the traditional case design and the Pro is a Mid-tower Case design.
Massive towers are dead. Imagine a machine that instead of drive bays had a couple of slots for high performance SSD storage. No more optical or magnetic drive supported in the box. Imagine a motherboard with the GPU integrated right on the motherboard next to the CPU. Imagine these cards oriented such that cooling is vastly improved as the drop the traditional expansion card format that does more to trap heat than anything.

The whole desktop market is just crying out for real innovation from somebody. Currently Apple appears to be the only company that could even attempt to redefine what a workstation is or does. One possibility here is a much smaller more modular box that allows for cost control while delivering high performance computing.
Quote:
In the end, I see Apple attempting to be taken more serious in the Workstation market and taking their design patents and patents into providing a more energy efficient with a higher energy total footprint option for expansion supporting PCI-E 3.x default.
Honestly I suspect Apple will downplay expansion. They may give us one or two PCI-E slots for conventional cards but the rest of the PCI hardware will go to support internally installed modules, those being the GPU, SSD and the Phi coprocessor. Unless of course Intel has a processor solution up its sleeves with a lot more high speed capable PCZi -Express lanes.
Quote:
I'm sure noise will be reduced in these new designs, better efficiencies achieved with higher performing ICB components, etc. And I imagine they will want an ECC 128/256GB possibility in their designs internally. Now whether we see that as BTO for consumption is a different matter.
One thing is obvious, we both think a new generation of machine is coming. We seem to be far apart on what that new machine is. Even things like RAM are ripe for replacement with new technology. Intel has been working on a 3D RAM solution for sometime now that could launch with these machines. You could potentially pop the top on next years Mac Pro and not recognize most of the hardware.
Quote:
I think the delays are more on what Apple is bringing to the table and not so much what Apple has to do to wait for others to bring to the table.
Frankly I'm 180 on this. I believe Apple has more than a few technologies they are waiting on.
Quote:
The AMD lines of GPGPUs are just superior for OpenCL solutions and the lead is going to widen moving forward. Apple is committed to OpenCL/OpenGL throughout their OS.

Exactly my point about AMD. But again it doesn't seem like Apple agrees if they are not using the GPUs. Further I see the biggest benefit from AMD in the hardware from Apple not using discrete GPUs. The extra GPU power of an APU in the base Mini for example.

At least at this point in time where intel integrated GPUs still suck big time.
Reply
Reply
post #20 of 26
10/21/12 at 1:01pm
Quote:
Originally Posted by Marvin View Post

They can get OpenGL 4 in Linux though:
http://www.youtube.com/watch?v=ZojsR4zwjt8
There's not a hope for that in OS X until Apple updates the drivers. Maybe they should open source the drivers if they haven't so that AMD/NVidia/other can put out custom versions.
The Mac Pro only allows 300W on the PCI slots so you can only have 1 powerful GPU.
I think removing the optical bay automatically allows a smaller form factor. If they keep the dual processors and slots, it won't shrink by much but I don't see how they can easily support slots and Thunderbolt. They have to put the dedicated GPU framebuffer into an IGP but they can't as Xeons don't have them. This means either abandoning TB, which is not a great idea because it won't support the TB Cinema Displays or ditch the slots except for internal use e.g they'd stick a single GPU inside the machine with no outward ports and save space.
This is an interesting problem. That is how do you allocate your available PCI Express bandwidth.

Because of Thunderbolt I see the GPU being integrated on the motherboard or on a daughter card with a non standard interface. I see TB support as extremely important to Apple going forward and extra lanes may be needed to support the TB chip and GPU implementation. I could even see them offering TB ports without video support. Either way I'm expecting four TB ports supporting video. So between the video card and the TB ports that is a lot of lanes right there.

No add a coprocessor, traditional I/O and suddenly you have very few lanes left over for expansion cards. PCI Express 3 throws a wrench in things due to the faster transfers, however this would be a long term design and thus I can't see Apple shorting either the video nor the coprocessor so you loose at least 32 lanes right there, plus maybe four for TB. Apple could very well need a chipset with more than 40 lanes to properly support a high performance machine.
Quote:

They'd need 16 RAM slots for 256GB RAM so I don't see that happening. I don't think they need to go beyond 6 slots for up to 96GB RAM and this can be used with one CPU. They are already quite cramped:
http://www.apple.com/uk/macpro/design.html#memory
One day, storage and RAM should be combined anyway.
I don't ever see that happening.
Quote:
I doubt they're trying to be taken more seriously in the workstation market. When the Mini went without an update for 1.5 years, people were dancing on its empty grave. The Mac Pro will have gone without a proper update for pretty much 3 years next June and the conclusion is that they're coming up with the biggest, baddest and best workstation anyone will have ever seen.
Well not exactly. I wouldn't call it a conclusion but rather a hope or dream that they will put a tiny bit of effort into the entire desktop line. Really it would be tiny because they have a basis in the laptops to work from. Well for the low end tiny. A totally refactored Mac Pro is a challenge. My point with the Mac Pro is that Apple is the only workstation player right now that can actually innovate in this area. That only if they want too.
Quote:
The iMac can hold a 95W chip. A 10-core 95W IB/Haswell Xeon would be just fine.
Wouldn't the following machine make for a great workstation?:
iMac Pro
10-core 2.5GHz Xeon 95W
64GB RAM
256GB SSD with BTO up to 768 + 2TB HDD
Retina 27" display (4K QHD)
2GB Radeon 8970M with unified memory
2x 20Gbps Thunderbolt, 4x 5Gbps USB 3
$2499
It is still an iMac with an iMac built in LCD. The SSD is way too small for workstation usage.
Quote:
They will even be able to get their hands on a 70W 10-core in Q3 2013:
http://pcper.com/news/Processors/Intel-Planning-10-core-Xeon-E5-2600-V2-Ivy-Bridge-EP-CPU
Like I say, it's either an IB MP in Q3 2013 or it's an iMac Pro. I personally think at that point in time, an iMac Pro would make more sense for Apple.
At that point they will have left the workstation market completely. These won't be suitable replacements for people currently using Mac Pros.
Reply
Reply
post #21 of 26
10/21/12 at 1:41pm
Thread Starter 

What part of System-Wide versus Application Specific do you not get? I don't mean to come off caustic, but this misperception continues with OpenGL.

 

OpenGL for the Desktop Environment in KDE 4.8.x - 4.9.2 and GNOME 3.6.x is OpenGL ES 2.0 partially. KWin is not even ready for OpenGL 2.1 fully. OS X 10.4 was OpenGL 2.1 compliant throughout.

 

OpenGL Drivers and their 3.x/4.x extensions for Applications are available for OS X just like they are for Linux and Windows.

 

We are talking about the actual Desktop UI being fully accelerated with OpenGL 3.2.x and soon 4.x. Xorg XWindows is

 

 

 

 

 

 

 

NOTE: The GLX Client Version 1.4 versus the GLX OpenGL Information version of 3.3.

 

The Desktop Environment on this particular Debian Sid Linux Box with an older Nvidia 8600GTS maxes out at the OpenGL 3.3 driver. I can show you the AMD 7750 presently supporting the OpenGL 4.2 as well using the fglrx driver.

 

None of the Desktop Environment [GUI] takes advantage of even a full OpenGL 2.1 profile. The extensions are leveraged by games and engineering applications but not the environment running these applications.

 

Huge difference. OS X across the entire environment is OpenGL 3.2.x accelerated and Display PDF.

 

Linux doesn't compare.

 

OpenCL is throughout OS X with GCD. Linux has nothing like it and OpenCL is currently being used sparingly in Blender, GIMP via Gegl, Handbrake and other small scale applications, outside of large scale commercial solutions.

 

Apple has the ball, not the other way around.

 

Martin's blog on the status of KWin and using Mesa's upcoming OpenGL 3.x compliance shows you the present status and gap between the two platforms.

 

http://blog.martin-graesslin.com/blog/2012/09/who-needs-glx-kwin-does-not/

 

Mesa 9 just got released: http://www.mesa3d.org/relnotes-9.0.html

 

Major caveats: It's OpenGL 3.1 support is stuck on Intel.

 

Quote:

 

Mesa 9.0 implements the OpenGL 3.1 API, but the version reported by glGetString(GL_VERSION) or glGetIntegerv(GL_MAJOR_VERSION) / glGetIntegerv(GL_MINOR_VERSION) depends on the particular driver being used. Some drivers don't support all the features required in OpenGL 3.1. OpenGL 3.1 is only available if requested at context creation because GL_ARB_compatibility is not supported.

 

New features

Note: some of the new features are only available with certain drivers.

  • Added new Gallium3D - nv30 driver
  • Added new Gallium3D - radeonsi driver
  • Added OpenCL state tracker Clover
  • Completed VDPAU state tracker (video decoding support is currently limited to MPEG1 and MPEG2)
  • GL_ARB_base_instance
  • GL_ARB_blend_func_extended
  • GL_ARB_debug_output
  • GL_ARB_invalidate_subdata - Currently a "no-op" implementation. This extension is always enabled in all drivers.
  • GL_ARB_shader_bit_encoding
  • GL_ARB_texture_buffer_object
  • GL_ARB_timer_query
  • GL_ARB_transform_feedback3
  • GL_ARB_transform_feedback_instanced
  • GL_ARB_uniform_buffer_object
  • GL_EXT_unpack_subimage for ES 2.0
  • GL_EXT_read_format_bgra for ES 1.1 and 2.0
  • GL_EXT_texture_rg for ES 2.x
  • GL_NV_read_buffer for ES 2.0
  • GLX_ARB_create_context_robustness
  • EGL_KHR_create_context
  • EGL_KHR_surfaceless_context - This replaces the EGL_KHR_surfaceless_{gles1,gles2,opengl} extensions that were never approved by Khronos.
  • EGL_EXT_create_context_robustness

 

 

This is what you're frustrated at: https://developer.apple.com/graphicsimaging/opengl/capabilities/

 

The Core Profile is at OpenGL 3.2. That means every pixel of the OS and its applications can fully leverage that Profile.

 

Apple is presently working on making sure the performance, optimization and leverage of OpenGL, OpenCL are fully available for LLVM/Clang moving forward.

 

Who is working with them to get that OpenGL status current to 4.x, OpenCL to 1.2 and soon the 2.x branch? Intel, AMD, ARM and even Nvidia.

 

Most if not all Engineering Applications have yet to fully use OpenGL 3.1 never mind 4.3 that is now the published revision and yet to be fully implemented at the hardware level by Nvidia or AMD, never mind Intel or ImgTec.

 

At the rate Mesa is working they should have OpenGL 3.3 support by 2014. Right about the time OpenGL 5 is nearing OpenGL 5.1 release spec.

 

 

 

 

 

Quote:
Originally Posted by wizard69 View Post


Things change, Intel and AMD, to a lesser extent, have been throwing a lot of effort into their Linux drivers.
Err if that was the case would Apple be pursuing NVidia? Now AMD is all in with OpenCL support, that I freely admit, but is that effort going to support Mac OS? I really think AMD sees a bigger win in Linux and Windows support.
As a side note LLVM/CLang has become huge. Many companies are working with Apple on its development even if they have no direct interest in support Apple hardware.
AMD has committed to OpenCL in a big way, I don't dispute that. However if Apple doesn't ship hardware with AMD inside it doesn't mean much.
I think the point that many mis, that maybe I didn't make clear, was that because of AMDs commitment to OpenCL their APUs would actually be good solutions for Apples lower end machines. Things like the Mini and even a retina 13"MBP. A significantly faster integrated GPU that also supports OpenCL well, would result in very balanced performance in these machines. In a nut shell AMD has more to offer Apples lower end machines than Intel does.
Maybe I've confused you here, I have no problem with Apple offering a workstation class machine. It is Apples lack of a midrange machine that bothers me.
In the case of a new design Mac Pro I was trying to point out that power usage isn't likely to decline much at all. If Apple plugs in a Xeon Phi co processor that can easily be 300 watts right there. In otherwords I don't see Mac Pro power requirements dropping if they deliver what I think is coming. As a side bar the 300+ watts for a Phi coprocessor is for basically the prototype processors which could be greatly reduced by a process shrink. In the end I just don't see the power demands of a Mac Pro shrinking all that much even if the box does.
Massive towers are dead. Imagine a machine that instead of drive bays had a couple of slots for high performance SSD storage. No more optical or magnetic drive supported in the box. Imagine a motherboard with the GPU integrated right on the motherboard next to the CPU. Imagine these cards oriented such that cooling is vastly improved as the drop the traditional expansion card format that does more to trap heat than anything.
The whole desktop market is just crying out for real innovation from somebody. Currently Apple appears to be the only company that could even attempt to redefine what a workstation is or does. One possibility here is a much smaller more modular box that allows for cost control while delivering high performance computing.
Honestly I suspect Apple will downplay expansion. They may give us one or two PCI-E slots for conventional cards but the rest of the PCI hardware will go to support internally installed modules, those being the GPU, SSD and the Phi coprocessor. Unless of course Intel has a processor solution up its sleeves with a lot more high speed capable PCZi -Express lanes.
One thing is obvious, we both think a new generation of machine is coming. We seem to be far apart on what that new machine is. Even things like RAM are ripe for replacement with new technology. Intel has been working on a 3D RAM solution for sometime now that could launch with these machines. You could potentially pop the top on next years Mac Pro and not recognize most of the hardware.
Frankly I'm 180 on this. I believe Apple has more than a few technologies they are waiting on.
Exactly my point about AMD. But again it doesn't seem like Apple agrees if they are not using the GPUs. Further I see the biggest benefit from AMD in the hardware from Apple not using discrete GPUs. The extra GPU power of an APU in the base Mini for example.
At least at this point in time where intel integrated GPUs still suck big time.
Reply
Reply
post #22 of 26
10/21/12 at 6:08pm
Quote:
Originally Posted by wizard69 View Post


I would hope that Apple would go the same way AMD is going and merge the GPU with the CPU. That is in the sense that they are equals with respect to memory access.
Unheard of performance? Seriously do you think it is that easy to beat Intel at the performance game. Apple is doing extremely well with A6 it I'm left with the impression that Intel has yet to get serious about ATOM.

 

Sure it's not easy to beat Intel, but the first ARMv8 64 bit processor shows much promise. And the A6 shows that Apple has a processor development team that rivals the best.

 

Check out the X-Gene's performance and imagine a SoC that's solely designed for the purpose of running OS X in a Mac.

 

http://images.anandtech.com/galleries/1532/Screen Shot 2011-11-14 at 1.41.15 PM.png

Reply
Reply
post #23 of 26
10/21/12 at 11:29pm
Quote:
Originally Posted by retroneo View Post

 

Sure it's not easy to beat Intel, but the first ARMv8 64 bit processor shows much promise. And the A6 shows that Apple has a processor development team that rivals the best.

 

Check out the X-Gene's performance and imagine a SoC that's solely designed for the purpose of running OS X in a Mac.

 

http://images.anandtech.com/galleries/1532/Screen Shot 2011-11-14 at 1.41.15 PM.png

A power-optimized design comparing energy efficient to a current-generation processor that is tricked out to run a single-thread as fast as possible. Yes, that's a fair comparison.

 

There's a reason why Intel hasn't been ramping up the core count, (we're still at the same 4C/8T since Nehalem circa 2008!). General purpose software typically doesn't do well with threads.

The X-Gene has 32C/32T. It's clearly not even in the same class of design. This graph really should have been comparing to Intel's MIC.

 

Also: Update from Anandtech:

 

Quote:
APM has since pulled the slide it shared with us originally making the comparison to Intel's Sandy Bridge architecture. The implication being that its performance estimates may have been a bit too aggressive, only time will tell...
Reply
Reply
post #24 of 26
10/22/12 at 12:01pm
Quote:
Originally Posted by AzN1337c0d3r View Post

A power-optimized design comparing energy efficient to a current-generation processor that is tricked out to run a single-thread as fast as possible. Yes, that's a fair comparison.

 

There's a reason why Intel hasn't been ramping up the core count, (we're still at the same 4C/8T since Nehalem circa 2008!). General purpose software typically doesn't do well with threads.

The X-Gene has 32C/32T. It's clearly not even in the same class of design. This graph really should have been comparing to Intel's MIC.

 

Also: Update from Anandtech:

 

 

I think the slide is an interesting snippet of things to come from Arm and the direction Apple is going with mobile computing.

 

Surely the race is on for Apple's business...by the time iOS matures enough to replace Macs all together...  Will Intel be able to beat ARM to that inevitable(?) destination?

 

It's going to be an interesting 2-3 years.  The iPad/iPhone's power is exploding near vertically.  You can already do more with them than you could PPC Macs?

 

Lemon Bon Bon.

You know, for a company that specializes in the video-graphics market, you'd think that they would offer top-of-the-line GPUs...[/
Reply
You know, for a company that specializes in the video-graphics market, you'd think that they would offer top-of-the-line GPUs...[/
Reply
post #25 of 26
10/23/12 at 12:02am
Quote:
Originally Posted by Lemon Bon Bon. View Post

I think the slide is an interesting snippet of things to come from Arm and the direction Apple is going with mobile computing.
Maybe maybe not. Personally I'd like to see Apple go 64 bit in iOS tablets as soon as possible.
Quote:
Surely the race is on for Apple's business...by the time iOS matures enough to replace Macs all together...  Will Intel be able to beat ARM to that inevitable(?) destination?
I think Intel lost already. ATOM was a sign that they just didn't get it.
Quote:
It's going to be an interesting 2-3 years.  The iPad/iPhone's power is exploding near vertically.  You can already do more with them than you could PPC Macs?
I continue to be amazed at just how "good" my iPad is as a computing device. This is only an iPad 3, double the speed and add more storage and this little device will be hard to resist for many users.
Quote:
Lemon Bon Bon.

Speaking of storage I do wish that Apple would pull its head out of its ass and do something significant about flash storage space. IOS devices need far more affordable space. While a couple TB would be nice I can accept that the technology isn't there yet. Instead provide us with three real And significant options, say 32, 128, and 256 GB models. Maybe even 512GB at the top end. Make iPads fly.
Reply
Reply
post #26 of 26
10/23/12 at 5:42pm
Quote:
Originally Posted by wizard69 View Post

I think Intel lost already. ATOM was a sign that they just didn't get it.

I dont think so. Everywhere I go, the netbook is the 2nd most common electronic device I see (after smartphones).

A lot of consumers are just looking for the cheapest device that can run x86-64. Atom fills that role nicely.

Reply
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Future Apple Hardware
AppleInsider › Forums › Mac Hardware › Future Apple Hardware › AMD Understand the value of OpenCL