Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

2016 nMP

Status
Not open for further replies.
koyoot said:
I have few things I can share about upcoming GPUs.

Polaris 10 and 11 are designed to work with HBM and GDDR5(X) GPUs.
There is absolutely no chance for big GPU from AMD with HBM2 in first 2 quarters. Q3 is the earliest where we can look for it.
It will also be Vega10. Vega is completely from the ground up designed to work ONLY with HBM.
About Polaris. It is mix of old and future. It is not groundbreaking in terms of performance of the GPUs, but efficiency.
I do not know at this point however if big Polaris will use HBM1. On he architectural side: it will be pretty interesting to see how it works. There may be few surprises. Especially with minimal frame rates.

Dual FuryX2 is still in plans because there will be no single chip that will be faster than that dual GPU. And it is made from 2 Fury Nano's.
Click to expand...

Why would AMD come out with 2 different architectures within a quarter of each other? Why not have all the chips use the same architecture? If they are coming that close together I am not convinced Vega is significantly different than Polaris. Vega is probably just the name of big Polaris.

tuxon86 said:
Should be "there is no actual single chip that will be faster than that dual GPU".

Unless you have a crystal ball of course :)
Click to expand...

At least for GPUs that are coming out this year, the prediction is pretty reliable. Remember, 14/16 nm is a new node, so we won't see the massive die sizes we saw in Fiji and GM200. Thats why the first round of GPUs will likely focus on compute performance and efficiency, then as the process matures we will see the big hot GPUs. Apple will likely choose the moderate sized efficient GPUs for the Mac Pro, like they did with Tahiti.

My guess is that the future Mac Pro GPU, whatever its code name is, will have Fury like performance or maybe slightly faster but in a 125 W to 150 W envelope.
 
The reason for making it different is simple: costs, and profits. I am not saying it will be absolutely different, but it will be different in the way it behaves. But the different name is absolutely appropriate here.

Secondly, you are absolutely correct with TDPs for GPUs. The question is, what will be the core counts. AMD said they are going to release next gen FirePro graphics in Q2.

One last bit at the end. 3072 GCN4 core GPU, with ~210 mm2 die size. What do you think? ;)
 
  • Like
Reactions: mrxak
koyoot said:
One last bit at the end. 3072 GCN4 core GPU, with ~210 mm2 die size. What do you think? ;)
Click to expand...

Sounds like a pretty reasonable guess. With power efficiency and per core improvements it should be a good deal faster than tahiti. I am hoping it ends up being a larger die at 4096 cores.
 
tuxon86 said:
Eh, in another threads many around here were saying that VR was just a fad that no Mac user were really interested in anyway... That is unless it has a bitten fruit logo on it then...
Click to expand...

Those same other threads are also filled with apocalyptic pronouncements that Apple is going to miss the boat on VR because we don't have the ability to install PCIe GPU cards. Then again, I remember back in 1998 being super unhappy that my new MP3 player didn't work with my Mac and being certain Apple was going to miss out on digital music...
 
  • Like
Reactions: mrxak and koyoot
I was really hoping for a good AMD GPU solution but it all seems to be getting messy again.
Guess we won't see the nMP before the end of the year after all.
By then, they might as well consider waiting for Skylake, but that would be 1+ years from now.
[doublepost=1454149516][/doublepost]VR at Apple is not news. Even now that they are again investing on it, there are no Apple machines up for it on their
portfolio at the moment.
This is a long term investment. Has to be.
Even the current solutions are somewhat clumsy.
Maybe in a couple of years...
 
I am pretty sure that Apple will enable the DSP in AMD's GPUs when VR is about to come. Making believable precise virtual 3D audio requires some power from CPU, that can be delegated to DSP. The next nMP could be the content creator tool for VR. Maybe this is why Apple has not updated Pro apps recently? They're waiting new features of nMP v2.

For consumer side, Apple could do VR same way as the did with Siri: make most of the rendering in the cloud so even an iPad could be a client machine.

In this vision, nMP has a future. Unless Apple's Swift for Linux is going to mean that future development is done on Linux...
 
Last edited:
Zarniwoop said:
For consumer side, Apple could do VR same way as the did with Siri: make most of the rendering in the cloud so even an iPad could be a client machine.
Click to expand...

I doubt that any cloud based VR rendering will deliver the low latency required. Unless you want a move-like-a-drunken-person-simulator as your only VR application. ;-)
 
Last edited:
  • Like
Reactions: mrxak, tuxon86 and askunk
One of the reasons why Kaby Lake iGPU will have 2x 128 MB of EDRAM is VR. Each memory pool can be used for each display/eye.
 
Augmented reality for masses is more likely to come first. But for a good VR experience it needs so much muscles rich people only can play with it. For the rest of us, part of the rendering work happens in a cloud. How about operator based cloud for low latency? Sure local GPU's render live 60fps per screen, but it will be united with cloud data seamlessly like a real time 3D video. Imagine a Facebook sized community interacting in VR. VR rendering could happen in three step method, locally in VR classes, in local computer and cloud. All data united seamlessly like todays augmented reality.
 
DearthnVader said:
Still saving my pennies for the MacPro7,1, hoping for Thunderbolt 3, USB-C, 18 Core Broadwell-EP, and official support for at least 128GB of Ram.

Of corse nVidia Quadro graphics with ECC VRam, but I know I won't get that.
Click to expand...

If there were any signs of nVidia GPUs being in the 7,1 , I probably would've have held out a bit longer. But, I don't see that happening unfortunately.
 
There is also another possibility. Apple will go for Broadwell EP CPUs, and... FIJI chips.

Not what I would do but...
 
LoftyHammond said:
If there were any signs of nVidia GPUs being in the 7,1 , I probably would've have held out a bit longer. But, I don't see that happening unfortunately.
Click to expand...
The monies burning a hole in my pocket, I keep hoping March, but I don't think we'll see a new MacPro until WWDC in June.

I can't get by without Cuda, so I at least need Thunderbolt 3's support for eGPU's.
 
Those of you who strongly prefer Nvidia GPUs, is it because of CUDA? If so, what applications only support CUDA and not OpenCL on OS X? From a developers perspective, it seems like only cross platform tools would fall in this category, as no OS X developer would require something that does not ship in any current macs.
 
  • Like
Reactions: mrxak
Stacc said:
Those of you who strongly prefer Nvidia GPUs, is it because of CUDA? If so, what applications only support CUDA and not OpenCL on OS X? From a developers perspective, it seems like only cross platform tools would fall in this category, as no OS X developer would require something that does not ship in any current macs.
Click to expand...
These developers have abandoned Apple OSX because of Apple's "kill CUDA" mentality.
 
Status
Not open for further replies.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back