R9 nano released

[AidenShaw]

This card looks interesting as an AMD GPU solution (except the price and perhaps a little more VRAM would be nice for future proofing).

The Nano's main selling point is that it's less than 15.24cm long, and that's it's been crippled to drop the TDP by 100 watts.

Neither of which should be interesting to people with maxi-tower systems. Card length is not an issue, and for most towers you simply upgrade the ATX power supply if needed.

And even with the crippling to reduce the TDP, only one of them could be put in the MP6,1 form/powerfactor. (Gee Jony, if instead of just crippling it to need crutches, if we cripple it to wheel-chair levels it would work!)

 

[Fl0r!an]

I suppose it is too much to hope for a Mac Edition with a boot screen and no issues with OS X updates however...

If it gets ever supported, it'll have it's drivers built right into OS X. For AMD cards there's not much between 'not supported at all' and 'works OOB'.

 

[h9826790]

IMO, AMD is now quite focus on the console parts. This stock Nano is also a parts for living room PC console. Gamers now complaining the poor hardware in consoles. On the other hand, a full size PC is way too big for a normal families' living room (especially for those who live in the cities). So now a new market start, the living room console like gaming PC. The Nano seems to focus on this market, which just accidentally fit inside the cMP.

Even though Sapphire happy to release a Nano Mac Edition, I doubt will Apple allow it. They must get the licences from Apple to do so. And in Apple's point of view, they won't allow any 3rd party hardware for the cMP that will make the nMP looks bad. Also, there should be no hardware upgrade for the cMP once the 6,1 is released. So, I don't think it's gonna to happen. Our only hope is at some stage, Apple use that GPU in their Mac, and we can dump the image and flash the card by ourselves.

The good news is that the Nano is good for a half size cooler. That means a 3rd party can release a overclocked full size version which still fit the cMP's power envelope. That's the gap for 3rd party to play, some where between the Nano and FuryX. I am sure there is a gourp of people out there will be happy to get a dual 6pin 225W normal air cool "oversize Nano", which is more powerful than the Nano, but not as power hunger as the FuryX, also no need to deal with that hybrid cooling solution in a normal size PC.

 

[Draeconis]

Is there any evidence that 10.11 supports the R9 Nano?

I'd guess that the next-gen Mac Pro would have two, so seems likely it'd support it.

 

[pastrychef]

Nope it isn't. Asynchronous Compute is all about context switching between graphics and compute. In Nvidia cards, the commands are dispatched in order, so they cannot refresh the pipeline, because there is only one Asynchronous Compute Engine. Thats why the pipeline stalls when it goes from Graphics to Compute and graphics again. Because GCN cards have 8 ACE's or 4 ACE's and 2 HWS you can dispatch the orders accordingly and the pipeline can be more... utilized. The pipeline is not stalled because of the context switching in AMD GPUs. It is some form of Out-of-Order execution of the pipeline.

Nothing, no matter what will Nvidia do will change the fact that there is only one ACE, and is incapable of doing at the same time Graphics and Compute. Thats where you will get stalled. Well, planned obsolescence to be precise.

What is worse for Nvidia, they cannot do anything to optimize it, because the API talks to the GPU, and the GPU only has API driver, not specific drivers for games, which is the case of DirectX11 games - there is no room to optimize in Driver. Every optimization of the hardware performance is done by the developers, they have to add specific code to the application which is exact case of Ashes of Singularity benchmark and Nvidia GPUs.

Look at this: http://arstechnica.co.uk/gaming/201...-win-for-amd-and-disappointment-for-nvidia/1/ R9 290X ties GTx 980 Ti. But look at 4K results and Nvidia hardware and the results with 4 and 6 cores. To get more power you will need higher CPU core count in higher resolutions. That is exactly the emanation of the lack of context switching - the CPU has much more work to do.

Its pretty ironic. Low-Level APIs came to life to reduce the amount the CPU has to work. With Nvidia hardware its not the case, with the inability to switch contexts. Table has turned around. DX11 - that was where AMD GPUs were bottlenecked by CPU, and not utilizing the whole GPU in parallel way. Nvidia Maxwell power came from optimized drivers. Direct12 - Nvidia struggles in parallel environment, where GCN starts to spread its wings. It will be much worse, when developers will go more into compute complexity in their games and applications. Im really curious to see the effects, but from everything we know so far: the gap between AMD and Nvidia will start to get bigger. In plus for AMD, in minus for Nvidia.

I'm not even going to pretend I understand all the stuff written in that Arstechnica article. But the author seems to share your enthusiasm for AMD's ACE. However, I would like to know whether this benefits us OS X users. Even on the Windows side, you only benefit if you are using DX12...

 

[koyoot]

Not only its for DX12. Its also for RayTracing API FireRays that AMD released few weeks ago, and is based on Mantle driver. Asynchronous compute is also base of full virtualization on the GPU. It allows many applications in the same time to be executed on the GPU. Not one Application. Two different applications or even more applications run in parallel on the same, one GPU. That is the benefit from Asynchronous Compute.

 

[pastrychef]

Thanks for the explanation. But is it safe to assume that in OS X, we don't see any of these benefits?

 

[koyoot]

FireRays already is for OSX. Everything right now depends on what will Apple do.

 

[MacVidCards]

Not in10.11 final

 

[koyoot]

http://developer.amd.com/community/blog/2015/08/14/amd-firerays-library/
http://developer.amd.com/tools-and-sdks/graphics-development/firepro-sdk/firerays-sdk/

Section about Asynchronous Queries.

Everything right now depends on what will Apple do

Its about hardware, not software.

The problem with Tahiti and Pitcairn GPUs is that they have only 2 ACE's. Only Tonga from M295X from iMac 5K has 8 ACE's.

 

[goMac]

Section about Asynchronous Queries.

Asynchronous Queries is not the same thing as Async Compute. Asynchronous Queries could possible be faster due to Async Compute, but they're not the same concept.

Async compute should be usable by OpenGL or Metal with some driver changes, I'd assume. Apple already supports context switching for when more than one app uses the GPU at a time (or one app doing both compute and graphics, etc), they'd just need to hook into the new model if there are any changes to be made.

It's not really virtualization. A better comparison is Hyperthreading. Multiple applications can already share the same GPU at the same time, but it adds a decent performance penalty. The entire reason Apple put two GPUs on the new Mac Pro is so that there would be less sharing of hardware between OpenCL and OpenGL, and that would avoid that performance penalty.

 

[koyoot]

Asynchronous Queries, are using Asynchronous Compute Engines...

Im not sure, goMac, you understand that post. In one about FireRays SDK there is absolutely NOTHING about virtualization. I have no idea, where you get it.

Here is about Virtualization that uses Asynchronous Compute Engines. http://www.amd.com/en-us/press-releases/Pages/amd-unveils-worlds-2015aug31.aspx

 

[goMac]

Asynchronous Queries, are using Asynchronous Compute Engines...

As I mentioned, asynchronous queries are probably faster due to asynchronous compute, but asynchronous queries work without it as well. Asynchronous Queries don't require Asynchronous Compute, they're not directly analogous. It's like still being able to do multithreading on a single core box. Multithreading doesn't require multiple cores.

You could do async compute on previous GPUs without asynchronous compute. As a basic example, it's exactly what happens when you run two OpenGL apps at once.

Im not sure, goMac, you understand that post. In one about FireRays SDK there is absolutely NOTHING about virtualization. I have no idea, where you get it.

Here is about Virtualization that uses Asynchronous Compute Engines. http://www.amd.com/en-us/press-releases/Pages/amd-unveils-worlds-2015aug31.aspx

You said it was the base of virtualization, I misread that as you saying it was virtualization. You're right.

 

[koyoot]

goMac, about the Async Queries. That is obvious. It only tells the GPU what to do with the command, and how to execute it. It is possible to do it without AsyncComp, but it will be much worse idea, than with it.

But in this case - it uses AsynchCompute

 

[ssls6]

With the new price drop ($499) and if native support comes to El Capitan...then I'd be real tempted by this card.

 

[netkas]

I fear Fiji can get same fate as Cayman

 

[Mattww]

Having been reading articles online recently I'm really wanting something from the new range of energy efficient and faster "Polaris" AMD cards expected later in 2016 to become an option for the cMP and one last GPU upgrade.

However I really need boot screen support and for the drivers to be included as standard in each OS X update as with my earlier 4870, 5870 and 7950 so I'm not holding out too much hope really....

 

[netkas]

Bootscreen isn't hard, proper osx support is. I have here 290x with bootscreen, but osx support is so poor that it's not gonna see light in near feature.

 

[ssls6]

The R9 Nano is a 390x right? If not El Cap then maybe the OSX version after June.

 

[Stacc]

The R9 Nano is a 390x right? If not El Cap then maybe the OSX version after June.

Negative. The Nano is based on AMD's Fiji (aka Fury) chip. The 390x is Hawaii. There are partial drivers already for Fiji in El Capitan but it is not working properly. There is some chance (probably small) it ends up in the next mac pro revision which would bring driver support.

 

[ssls6]

Negative. The Nano is based on AMD's Fiji (aka Fury) chip. The 390x is Hawaii. There are partial drivers already for Fiji in El Capitan but it is not working properly. There is some chance (probably small) it ends up in the next mac pro revision which would bring driver support.

Thanks

 

[danboyy]

Thanks

Hey guys, was looking on the net and found your thread. Just FYI I am running a Sapphire R9 fury in my MacPro 5,1 and it does work plug'n play with El Capitan 10.11.3. I do have a slow refresh rate (or lag) for videos and some artifacts sometimes so it is not perfect... But still it works! It does fit really really tight in the case; I had to get the front case fan forward permanently, so the graphic card's handle is unlatched.

Naturally on the Windows side of things with the Crimson driver it works beautifully.

I can post pictures and Os info's if you ask.


Bye!

 

[netkas]

pics or didn't happen

opengl viewer and system profiler

most likely you only have framebuffer, no acceleration/opengl/video

 

[danboyy]

I can only watch videos in window mode for youtube, in fullscreen it's laggish.

Oh and I'm running this card off an auxiliary PSU in the optical bay to make sure I don't fry the motherboard.

 

[koyoot]

Is it overall meaningful information, or not?