Will there be a new Mac Pro in 2015?

[koyoot]

http://wccftech.com/intel-amd-nvidia-future-industry-hsa/
Remember that lately there has been few additions to HSA foundation. Apple is pushing it as hard as they can, both on ARM front and x86.

This is the future for Apple. And Mac Pro is the emanation of this idea. The most interesting here part is the last bit, about OpenCL 2.0 in this context.

 

[filmak]

http://wccftech.com/intel-amd-nvidia-future-industry-hsa/
Remember that lately there has been few additions to HSA foundation. Apple is pushing it as hard as they can, both on ARM front and x86.

This is the future for Apple. And Mac Pro is the emanation of this idea. The most interesting here part is the last bit, about OpenCL 2.0 in this context.

Thanks koyoot. Very informative and useful article.

 

[flat five]

Wouldn't be worth the trouble if 7,1 GPUs won't be available separately though (and they won't).

why won't they be?
someone with d300s (i.e.- mvc) could go out and buy d700s.. and put them in their mac pro.. today.

 

[koyoot]

Flat, its about new, different model GPUs for Mac Pro, not service parts.

 

[flat five]

Flat, its about new, different model GPUs for Mac Pro, not service parts.

not what florian is saying.
at least as i read it.
he's saying there's a chance of putting a 7.1 gpu in a 6.1 except we wouldn't be able to source 7.1 gpus as a separate purchase so the point is moot..
that's the bit i'm saying is not true (like- i'm not guessing.. the 6.1 gpus are available as separate purchases.. same will be true of 7.1)
(if... if you're willing to spend thousands of dollars on new gpus )

 

[Fl0r!an]

(if... if you're willing to spend thousands of dollars on new gpus )

Yeah exactly that's what I ment. Of course you can pull some Dxxx from a MacPro or try to get them as service part from Apple (not easy for current MacPro as far as I know), but both choices will be ridiculously expensive.
Last time I saw D700s on eBay, they were $1000 each. For a GPU that doesn't do anything more than a $200 HD 7970 can do in OS X.
Not talking about putting R&D expenses for developing a MP6,1 EFI with increased GPU compatibility on top of that...

 

[flat five]

Yeah exactly that's what I ment. Of course you can pull some Dxxx from a MacPro or try to get them as service part from Apple (not easy for current MacPro as far as I know),

nah.. if you're a apple authorized dealer or service center, you can buy & sell parts for mac computers.

these people are taking it to the extreme and selling every single individual part in a nmp:
http://www.macpartsonline.com/mac-pro-parts/mac-pro-late-2013-a1481-parts.html

 

[Fl0r!an]

I don't think Apple gives them out freely in large (= bigger than one per serial number) quantities, but I'm no AASP, that's just what I heard. I think most MacPro parts only come from MacPros that have been ripped apart.

 

[flat five]

I don't think Apple gives them out freely in large (= bigger than one per serial number) quantities, but I'm no AASP, that's just what I heard. I think most MacPro parts only come from MacPros that have been ripped apart.

heh.. why would they hoard those?
if someone wants to spend $2000+ on gpus, why wouldn't apple sell them?
that's more than most people spend on complete apple products.. pretty sure they'd be happy to sell some parts at those prices.

 

[koyoot]

Addition to my last post with link about HSA. Fiji and Tonga are a bit different architecture than the GPUs that were released before them.

Lets start from the beginning.
http://assets.hardwarezone.com/img/2013/10/radeon-r9-290-core-1.jpg Radeon R9 290X block diagram.
http://media.redgamingtech.com/rgt-website/2015/08/AMD-Radeon-R9-Nano-Fiji-GPU-Block-Hot-Chips.jpg

The differences in the upper part of the block are apparent. Now, read this post about it: http://forums.anandtech.com/showpost.php?p=37669975&postcount=14
Here is the thread about it: http://forums.anandtech.com/showthread.php?t=2444961
Here is the cause for the thread: http://ir.amd.com/phoenix.zhtml?c=74093&p=RssLanding&cat=news&id=2083146
AMD opened new world of possibilities of the use of GPU.

A bit of personal post. Everybody that know me, and read my posts on polish Apple forum: myapple.pl can say that last year, I was gigantic fan of Maxwell architecture. I was staggered about the benchmarks, and what people said about Maxwell architecture. I was hoping that we can see in the "Thrash can" GM204 GPUs. I was going that we can see "thrash can-like" Mac Mini, with quad core E3 Xeon CPU and GM107 or something like this. But then, Apple decided to completely ditch the Nvidia from the lineup, with iMac 5K. I wondered why, and started digging. Ive seen that real world OpenCL performance of Nvidia GPUs is rubbish, and R9 280X is 3 times faster in Final Cut Pro X than GTX 970 while using OpenCL. Then I started analyzing the architectures of AMD and Nvidia hardware, and differences became apparent.

Everything what benefits Nvidia is due to software. Drivers, CUDA, even virtualization is made by software. AMD is other way around - what benefits them is the hardware. Asynchronous compute is the pinnacle here. The problem for AMD was that before low-level API world the software was not able to extract all of their capabilities from hardware. GPUs would stall in some part, not being fully utilized. Times has changed in this case.

Apple is pushing for HSA foundation instead of CUDA. Apple is pushing OpenCL, as the go-to solution. The benefits for it are clear, as you can see in the WWCF article. And you also may know why Mac Pro has 450W PSU, and Dual, undervolted, and declocked GPUs that are wide enough. You now know why it is AMD instead of Nvidia: Asynchronous Compute/HWS. It is time to end the arguments here.

And focus on technical stuff .

 

[koyoot]

http://ext3h.makegames.de/DX12_Compute.html

Tremendous analysis of differences between GCN and Maxwell.

 

[zephonic]

But then, Apple decided to completely ditch the Nvidia from the lineup, with iMac 5K. I wondered why, and started digging. Ive seen that real world OpenCL performance of Nvidia GPUs is rubbish, and R9 280X is 3 times faster in Final Cut Pro X than GTX 970 while using OpenCL. Then I started analyzing the architectures of AMD and Nvidia hardware, and differences became apparent.

Everything what benefits Nvidia is due to software. Drivers, CUDA, even virtualization is made by software. AMD is other way around - what benefits them is the hardware. Asynchronous compute is the pinnacle here. The problem for AMD was that before low-level API world the software was not able to extract all of their capabilities from hardware. GPUs would stall in some part, not being fully utilized. Times has changed in this case.

Apple is pushing for HSA foundation instead of CUDA. Apple is pushing OpenCL, as the go-to solution. The benefits for it are clear, as you can see in the WWCF article. And you also may know why Mac Pro has 450W PSU, and Dual, undervolted, and declocked GPUs that are wide enough. You now know why it is AMD instead of Nvidia: Asynchronous Compute/HWS.

Good post.

About the nMP's PSU, it no longer has to power possible internal expansions, so they know exactly how much power is needed for the machine.
The cMP had to provide enough power for God knows what end-users or after-market suppliers were going to stick in there, hence the 1000W PSU.

 

[Kpjoslee]

Apple is pushing Metal n I think OpenCL has somewhat questionable future in Apple platform. For Apple using AMD gpus across the board... I think it has to do more with financials rather than grand technical vision lol.

 

[koyoot]

Why would you not use a hardware that is better for your needs and less expensive to buy, at the same time?

Only for brand reasons?

P.S. Do you even understand how does low-level API like Metal work for OpenCL?

 

[Kpjoslee]

Why would you not use a hardware that is better for your needs and less expensive to buy, at the same time?

Only for brand reasons?

P.S. Do you even understand how does low-level API like Metal work for OpenCL?

I am talking about which company would have given better deals for Apple to use lol. Nothing to do with "brand reasons"
From what I understand, Metal doesn't work under OpenCL or vice versa. They are separate APIs, what are you trying to say?

 

[koyoot]

Ok, then you don't understand at all API and OpenCL.

Metal is only a framework for OpenComputingLanguage. Thanks to Metal application knows how to use OpenCL to execute the job it has to do.

And no, OpenCL is NOT an API.

https://www.khronos.org/opencl/ Here you have direct explanation of OpenCL, and the use of it. Hint: Heterogenous architecture is: HSA.


What is more interesting is that Final Cut Pro X without any API knows how to execute commands while using Asynchronous Compute. That is direct reason why it is so much faster in OpenCL work on AMD than on Nvidia hardware.

 

[linuxcooldude]

Apple refusing to include modern GPU drivers in their OS doesn't reflect badly on MacVidCards or Nvidia, it reflects badly on Apple. Nvidia had Beta drivers for El Cap BEFORE release. They then released FINAL EL Cap drivers less then 24 hours after release.

Again, if you want to whine about Apple not including modern GPU drivers, Apple is the one and only cause of that issue. Stick with the cards that Apple dictates, or find courage and try the newer ones that are available, up to you. Cowering in the corner and dodging questions go well together.

Apple using limited select hardware should be of no surprise to anyone here. Thats their business model. But don't blame Apple when problems arise in using unofficial/unsupported graphic cards. The end user takes sole responsibility on what he puts in his machine. Other then normal, realistic, support issues. If using unlimited graphic cards or other parts is someones preference, they are likely using the wrong platform to do so.

 

[Melodist]

Apple is pushing Metal n I think OpenCL has somewhat questionable future in Apple platform. For Apple using AMD gpus across the board... I think it has to do more with financials rather than grand technical vision lol.

I think AMD is a good partner for Apple to work with since they came up with that integrated memory controller on the FX CPU line which will be an old / new standard for both AMD and Nvidia graphics cards, called High Bandwidth Memory 2 in its next iteration.

But Apple, please offer an average graphics solution with more CPU for us audio folks. Otherwise, I'll really have to reconsider my options before I get saddle-sore.

 

[linuxcooldude]

Apple is pushing Metal n I think OpenCL has somewhat questionable future in Apple platform. For Apple using AMD gpus across the board... I think it has to do more with financials rather than grand technical vision lol.

It could be multiple reasons. Pricing and working in a more open platform with openCL verses a more proprietary Nvidia. OpenCL is now getting more push from software companies now that the nMP uses AMD. I think that is good as it will even benefit the PC workstation market too.

 

[Melodist]

heh.. why would they hoard those?
if someone wants to spend $2000+ on gpus, why wouldn't apple sell them?
that's more than most people spend on complete apple products.. pretty sure they'd be happy to sell some parts at those prices.

um, if someone can afford a Mac Pro for beyond 3k, then they probably don't mind spending another 2 grand on better graphics cards if that's their thing.

But as the nMP is an overpriced design box for now, there is barely appeal on either side.

 

[MacVidCards]

heh.. why would they hoard those?
if someone wants to spend $2000+ on gpus, why wouldn't apple sell them?
that's more than most people spend on complete apple products.. pretty sure they'd be happy to sell some parts at those prices.

Here you go again with "Who are you going to believe, me or your lying eyes"

Apple doesn't offer up Dx00s as parts for consumers to buy. They just don't. If they did, we would see them all over Amazon and EBay with various vendors trying to undercut each other.

Instead we see quite clearly that just a few people have access, and as a result a $100 part sells for $1,000. The grand opening up of nMP upgrades via Apple Store and Best Buy didn't happen, instead it is the same as ever. Apple will sell you a D700, but only if yours dies and a service center in their network removed dead one and sends it back.

When Apple has sold specific GPU upgrades they were on their site. A couple of people selling ones that "fell off a truck" isn't going to lead to wide availability.

I really doubt that 7,1 GPUs will be pin compatible, lots of changes coming.

 

[Kpjoslee]

Ok, then you don't understand at all API and OpenCL.

Metal is only a framework for OpenComputingLanguage. Thanks to Metal application knows how to use OpenCL to execute the job it has to do.

And no, OpenCL is NOT an API.

https://www.khronos.org/opencl/ Here you have direct explanation of OpenCL, and the use of it. Hint: Heterogenous architecture is: HSA.


What is more interesting is that Final Cut Pro X without any API knows how to execute commands while using Asynchronous Compute. That is direct reason why it is so much faster in OpenCL work on AMD than on Nvidia hardware.

Got ya. OpenCL is a framework, not an API. Metal itself is an API though. Calling Metal only a framework for OpenCL sounds like it oversimplifies things too much. Reason I question OpenCL's future on Apple platform is that Apple is free to change to whatever it would adapt better to their hardware rather than wait for Khronos Group sets the standard.

I don't care about the whole Nvidia v. AMD, especially when it comes to Apple side lol. If Apple decides to go Nvidia next round of Macs, we deal with it and vice versa when Apple goes with AMD. We don't have much of a choice, do we?

 

[996085]

I really doubt that 7,1 GPUs will be pin compatible, lots of changes coming.

I'm curious as to what leads you to this conclusion?

 

[goMac]

Ok, then you don't understand at all API and OpenCL.

Metal is only a framework for OpenComputingLanguage. Thanks to Metal application knows how to use OpenCL to execute the job it has to do.

And no, OpenCL is NOT an API.

https://www.khronos.org/opencl/ Here you have direct explanation of OpenCL, and the use of it. Hint: Heterogenous architecture is: HSA.


What is more interesting is that Final Cut Pro X without any API knows how to execute commands while using Asynchronous Compute. That is direct reason why it is so much faster in OpenCL work on AMD than on Nvidia hardware.

Metal's compute and OpenCL are not related. They are two different technologies. There is a separate OpenCL API for running OpenCL. (A framework IS an API.)

Like I said earlier, you don't need some special AMD API for async compute. OpenCL already supports async compute. OpenGL on Mac OS X can do async graphics as well.

What AMD is adding is like Hyperthreading: It's better hardware support for something you can already do in software.

 

[Melodist]

I'm curious as to what leads you to this conclusion?

I too think that's to be expected after an update gap of two years, mind you for right after launching a complete new and drastic redesign from head to toe.