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

AMD Confirms RX 480 At $199 USD, Other APU & Polaris Announcements

koyoot said:
Both have exactly 36 GFLOPs/watt, and similar performance in DX12, and Vulkan, per watt. Only DX11 with high overhead on AMD GPUs is where AMD's efficiency tanks. And drives people's mindshare...
Click to expand...

GTX 1080: 8.9 TFLOPS/166 W = 54 GFLOPS/watt
RX 480: 5.8 TFLOPS/163 W = 36 GFLOPS/watt

Power consumption is from "typical gaming" scenario at techpowerup. Clearly AMD has some ground to make up in efficiency.

Hitman (Directx 12) numbers (AMD's far and away most favorable gaming benchmark)

GTX 1080: 65 FPS/166 W = 0.39 FPS/W
RX 480: 41 FPS/163 W = 0.25 FPS/W

And here AMD loses in efficiency at their best gaming benchmark.

When Apple produced its A series chip at 14/16 nm across TSMC and global foundries it had better efficiency at TSMC. I don't think its crazy to think that some of AMD's efficiency problems may be due to the fact that Global Foundries' process is not as good.
 
It depends what Vega 10 GPU will be. 320mm2 die and 3072 GCN cores with HBM1, clocked at 1.45 GHz and under 140W of TDP?

Look at the WX series. 1792 GCN cores, with 4 TFLOPs - under 75W of TDP. That means it must have over 1130 MHz core clock. RX 480 has 1266 MHz, and 150W of TDP. It is extremely pushed over its boarders.
Stacc said:
GTX 1080: 8.9 TFLOPS/166 W = 54 GFLOPS/watt
RX 480: 5.8 TFLOPS/163 W = 36 GFLOPS/watt

Power consumption is from "typical gaming" scenario at techpowerup. Clearly AMD has some ground to make up in efficiency.

Hitman (Directx 12) numbers (AMD's far and away most favorable gaming benchmark)

GTX 1080: 65 FPS/166 W = 0.39 FPS/W
RX 480: 41 FPS/163 W = 0.25 FPS/W

And here AMD loses in efficiency at their best gaming benchmark.

When Apple produced its A series chip at 14/16 nm across TSMC and global foundries it had better efficiency at TSMC. I don't think its crazy to think that some of AMD's efficiency problems may be due to the fact that Global Foundries' process is not as good.
Click to expand...
Come back to this calculations when AMD will release the direct competitor for GTX 1080, and GTX 1070.

If you want to compare this, compare RX480 with GTX 1060. How does it look then? Both are on exactly the same level.

And what is funny, you know how I calculated?
1280 CC's x2 x 1.706 MHz/ 120W of power consumption.
2304 GCN x 2 x 1.266 / 160W of power consumption. What are the results? 36 GFLOPs/watt. For both GPUs.
Now calculate the efficiency based on these numbers: http://www.golem.de/news/geforce-gt...ingen-direct3d-12-und-vulkan-1607-122214.html
 
Last edited:
Stacc said:
GTX 1080: 8.9 TFLOPS/166 W = 54 GFLOPS/watt
RX 480: 5.8 TFLOPS/163 W = 36 GFLOPS/watt

Power consumption is from "typical gaming" scenario at techpowerup. Clearly AMD has some ground to make up in efficiency.

Hitman (Directx 12) numbers (AMD's far and away most favorable gaming benchmark)

GTX 1080: 65 FPS/166 W = 0.39 FPS/W
RX 480: 41 FPS/163 W = 0.25 FPS/W

And here AMD loses in efficiency at their best gaming benchmark.

When Apple produced its A series chip at 14/16 nm across TSMC and global foundries it had better efficiency at TSMC. I don't think its crazy to think that some of AMD's efficiency problems may be due to the fact that Global Foundries' process is not as good.
Click to expand...
We're comparing now Airbus 380 to Boeing 737-800. It's not a fair comparison.
 
Direct3D12-Vulkan-Test-02.png

Direct3D12-Vulkan-Test-17.png

Direct3D12-Vulkan-Test-04.png

Direct3D12-Vulkan-Test-06.png

Direct3D12-Vulkan-Test-08.png

Direct3D12-Vulkan-Test-10.png

General conclusion on ALL of the gaming forums is that it is dead tie between these two GPUs. Both in terms of raw performance and efficiency.

Everywhere, but not on this forum ;).
 
Zarniwoop said:
We're comparing now Airbus 380 to Boeing 737-800. It's not a fair comparison.
Click to expand...

However, both of their fuel consumptions are around 25kg/ton/hour. As long as both machines are from the same generation, they should have roughly the same efficiency (per unit).
 
koyoot said:
Direct3D12-Vulkan-Test-02.png

Direct3D12-Vulkan-Test-17.png

Direct3D12-Vulkan-Test-04.png

Direct3D12-Vulkan-Test-06.png

Direct3D12-Vulkan-Test-08.png

Direct3D12-Vulkan-Test-10.png

General conclusion on ALL of the gaming forums is that it is dead tie between these two GPUs. Both in terms of raw performance and efficiency.

Everywhere, but not on this forum ;).
Click to expand...

I'm not sure where these benchmarks come from, but at Guru3D it seems to me that the GTX 1060 and the RX 480 are essentially even in performance (in directx12/vulkan) but the 1060 consumes 30 W less.

index.php

Win Nvidia

index.php


win AMD

index.php


Tie

index.php


Tie

index.php


Win Nvidia

index.php


Win Nvidia

In my mind the directx12/vulkan performance is a toss up but Nvidia has superior directx11 performance and better power consumption.
 
So last night (CET) AMD introduced new software. And this is the future. Hardware is reaching it's limits (CPU's already there), but there's a lot of room for software optimization. And if you look at the numbers AMD brought up, they'll get crazy figures out of RX480 for 3D rendering. The best part is, the open source software was also introduced for OS X.

Second thing that was introduced is Radeon Pro SSG. It was seemingly built on Fiji chip. I'd see this is going to be part of Mac Pro in the future. GPU to slot #1 (Polaris 10 Pro or XT) and Fiji to second with SSG. (or tomorrows GPU tech). That would be a video editing tool for 8k.

These two thing alone could be the reason why Apple chose AMD. Together with OpenCL + high optimization + Metal + SSG, Mac Pro will have a strong future.
 
Last edited:
Those benchmarks are from Golem.de. And they compared the affect DX12 and Vulkan has on both vendors in terms of performance.
Stacc said:
In my mind the directx12/vulkan performance is a toss up but Nvidia has superior directx11 performance and better power consumption.
Click to expand...
That Doom benchmark was made in OpenGL. You can check in the golem.de benchmarks differences in Vulkan and OpenGL, and how it affects the performance on both vendors.

Conclusion is this: Nvidia GPUs are better in legacy APIs, than AMD because of driver overhead. When you lift the driver overhead AMD starts to pull away from Nvidia, at the expense of higher power consumption. General efficiency is similar if you compare both GPUs in DX12 and Vulkan, even when we exclude the games that prefer particular architecture one way or another.
 
Zarniwoop said:
Who says that Apple has to use two SAME GPU's to the next nMP? Let's say that GPU is DX series Polaris 10 Pro and XT. But for GPGPU they use Fury Nano. That way they'd get best of both worlds: latest GPU tech with DP 1.4 and fastest GPGPU AMD has at the moment for under 175W TDP class.
Click to expand...
Good point, and that would also make it easy to have the compute GPGPU optional.
 
Zarniwoop said:
Second thing that was introduced is Radeon Pro SSG. It was seemingly built on Fiji chip. I'd see this is going to be part of Mac Pro in the future. GPU to slot #1 (Polaris 10 Pro or XT) and Fiji to second with SSG. (or tomorrows GPU tech). That would be a video editing tool for 8k.
Click to expand...

This is some slick tech by AMD. Given that the SSD in the mac pro is already mounted on the GPU, I wonder if Apple could take advantage of this. Use any left over free space on the SSD as extra swap space for the GPU.

koyoot said:
Conclusion is this: Nvidia GPUs are better in legacy APIs, than AMD because of driver overhead. When you lift the driver overhead AMD starts to pull away from Nvidia, at the expense of higher power consumption. General efficiency is similar if you compare both GPUs in DX12 and Vulkan, even when we exclude the games that prefer particular architecture one way or another.
Click to expand...

Given that we are talking about Apple we have to remember that there are thermal constraints. Every GPU Apple ships in a mac is basically limited to 125 W. Right now I think Nvidia can deliver better performance with Pascal (both GP104 and GP106) in that power envelope than AMD with Polaris almost regardless of the task.

AidenShaw said:
Good point, and that would also make it easy to have the compute GPGPU optional.
Click to expand...

I wouldn't mind seeing this type of solution but you could expand the possibilities to something like a straight up compute GPU like the GP100 (although it would not fit in the thermal envelope) or something like Intel's Xeon Phi.
 
Stacc said:
Given that we are talking about Apple we have to remember that there are thermal constraints. Every GPU Apple ships in a mac is basically limited to 125 W. Right now I think Nvidia can deliver better performance with Pascal (both GP104 and GP106) in that power envelope than AMD with Polaris almost regardless of the task.
Click to expand...
That is your opinion only. Apple and AMD might disagree with you on that.
 
Last edited:
  • Like
Reactions: filmak
Raja at the Capsaicin keynote have said that next GPU from AMD is coming "soon".

Let the speculation begin. IMO. It might be HBM1 chip with 3072 GCN cores, made on TSMC 16 nm process. Why do I believe that is the case? Because that is the manufacturer that AMD used to build first HBM GPU. GloFo does not have any experience with it, also Amor which connects for AMD the interposer with the dies and with memory is working only with... TSMC.

For Nvidia, interposer/HBM chips manufactured are completely by TSMC, and that includes putting the interposer, memory and the GPU dies together.
 
filmak, the man even named his daughter after the color :)
He must really like it, and it's efficient following AMD's target...
I was gonna say only blue was missing from the RGB trio but in fact Intel is the blue team, so maybe Intel will buy AMD soon.

I guess AMD slipped real bad when going with GloFo. TSMC has had a good track record and all points to having the lead, even if it's on 16nm vs 14nm of GloFo. This is quite evident on Polaris and Pascal I guess. Of course, the arch plays a crucial role here, but still...
 
Tim Cook sees a huge potential in AR (and VR) in the long run. For iDevices of course but I say the development platform (nMP if we're lucky) should be solid, right?
 
Stacc said:
This is some slick tech by AMD. Given that the SSD in the mac pro is already mounted on the GPU, I wonder if Apple could take advantage of this. Use any left over free space on the SSD as extra swap space for the GPU.
Click to expand...
I have been thinking a bit on this. In theory, it would be possible if the SSD would be mounted through coherent fabric, that connects the GPUs, and SSD.

Now the information about RAID soft is Sierra, about possibilities of dual SSD in MP, are taking "strange" shape... It might be interesting to wait and see what will happen.

One more thing. If you have something like this, all you need is GPU that has very high compute performance, and fast memory like HBM as a cache. It does not need to have high amount of that cache, tho.
 
ManuelGomes said:
http://appleinsider.com/articles/16...raphics-intensive-games-to-metal-on-macos-ios
Click to expand...

The problem is that Metal is a subset (at best) of Vulkan. There are concepts that you can express in Vulkan that will simply never be able to be implemented in Metal, which means this MetalVK thing will never really work that well. If it were the other way around (implementing the Metal API on top of Vulkan) then you'd be in business.

For example, what if your Vulkan app uses geometry shaders? Metal does not support them at all, even in Sierra. How does MetalVK deal with that?
 
ActionableMango said:
I feel like I've been hearing that for about 8 years. Its always just around the corner.
Click to expand...
Before it was pure impossible. Way too big overhead in openGL already, and to translate DX to oGL... no way. Now, Metal will play the GPU and it just needs a driver. There's so small overhead, that it is almost like your writing directly to GPU. So, it depends on the virtual machine makers, how good driver they can make.
 
It may never be as fast as Windows, however at least right now there is a chance that it will not be slower than 95% of Windows 10 performance.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back