Next-Gen Radeon "Polaris" Nomenclature Changed?

[G33k2Fr34k]

Wccftech, but they've just corrected it to "oh, we don't know the fps".
http://wccftech.com/amd-radeon-rx-480-graphics-card-doom-1440p-benchmark/

If that's true, then the 480 is close to the R9 Fury, which means that the 480X is close to the Fury X. That's pretty good given the 110W and 130W rumored TDPs.

 

[Caring1]

144Hz was monitor, come on guys...

Demo was capped at 60 fps.
Which puts RX 480 slightly below 980/Fury (non-Ti, non-X).

That makes sense to me and matches some of the FS benches i've seen placing it above the GTX970.

 

[dem343]

"Doom" (2016) at 2560x1440 @ 144 Hz, with the OpenGL API.


Is this a good thing, bc I have no idea about OGL bench numbers

remember VULKAN is OpenGL 5.x
it means the gameplay was running under VULKAN instead of dx

 

[G33k2Fr34k]

That makes sense to me and matches some of the FS benches i've seen placing it above the GTX970.

I think it's slightly above the 980, but either way, I hope AMD prices it at $200, which translates to $250 CAD. That would make it my next GFX.

 

[FordGT90Concept]

Let's do some math!

RX 480
2560 SPs
unknown clock

R9 390
2560 SPs
1000 MHz

R9 390 got 57.4 fps so lets scale it:

1.0 GHz = 57.4 fps
1.1 GHz = 63.1 fps
1.2 GHz = 68.9 fps
1.3 GHz = 74.6 fps
1.4 GHz = 80.4 fps
1.5 GHz = 86.1 fps
1.6 GHz = 91.8 fps
1.7 GHz = 97.6 fps (beats GTX 1080)

I doubt it will be clocked higher than 1.5 GHz (1.1-1.2 GHz is the most realistic). Unless AMD made some massive improvements to OpenGL rendering, the RX 480 isn't likely to top GTX 1080.

 

[librin.so.1]

R9 390 got 57.4 fps so lets scale it:

"scale" it? Did you mean... lerp it?

 

[Caring1]

Speculating based on numbers I've seen on FireStrike, Polaris seems to be between the GTX970 and GTX1080, which is quite a big gap.
Although there is an unknown card on par with the overclocked GTX1080, but with lower clocks lurking.

 

[medi01]

If that's true, then the 480 is close to the R9 Fury, which means that the 480X is close to the Fury X.

Agreed, but biggest if here is whether it's 480 and not 480x.

Although there is an unknown card on par with the overclocked GTX1080, but with lower clocks lurking.

That was cross fire of the same card. (looked a bit misleading with cryptic card name). Note the CF at the end.

 

[FordGT90Concept]

Yeah, Crossfire. If those cards were running at reference clocks, RX 480 is looking to land somewhere between Fury and 390X.

http://wccftech.com/amd-radeon-r9-480-polaris-10-july/

That said, a dual Polaris card is entirely plausible--just not coming soon.

Remember, Polaris is a <150w card. GTX 1070 is too... Going from the data in this thread, it looks like GTX 1070 should be faster than Polaris but only by 4% (predominantly because of DX11).

 

[prtskg]

Consider a ridiculous situation. X is a typo. Instead of 480X, they wrote X480.

 

[IceScreamer]

Is it in any way possible that the X could stand for Crossfire? I'm not speculating, I'm genuinely wondering if that is possible. But not regular Crossfire we had so far, but some form of driverless connection between two on board GPU's, two smaller dies working as one. Can anyone more knowledgeable explain if/why this is/isn't possible?

 

[FordGT90Concept]

Good idea and it would explain the "CF" benchmark. RX 480 (assuming it means two) would be an ideal GPU for VR games which AMD is heavily promoting. It could also run away from GTX 1080 in DX12/Vulkan games.

We knew AMD was heading in this direction but I didn't expect it this soon. It makes sense though because AMD wants to promote dual GPU graphics solutions so that the DX12/Vulkan games & market is ready for Navi.

 

[IceScreamer]

Good idea and it would explain the "CF" benchmark. RX 480 (assuming it means two) would be an ideal GPU for VR games which AMD is heavily promoting. It could also run away from GTX 1080 in DX12/Vulkan games.

Yea it crossed my mind when I saw the "CF" tag, but as you pointed out it's kinda too soon for a dual GPU already.

 

[FordGT90Concept]

It would explain the launch delay. They could be manufacturing the dual Polaris 10 cards as a response to the GTX 1080. No reason why that couldn't happen inside of a few months.

 

[IceScreamer]

It would explain the launch delay. They could be manufacturing the dual Polaris 10 cards as a response to the GTX 1080. No reason why that couldn't happen inside of a few months.

Hmmm, lets say it turns out to be true, wouldn't that be kind of a bearer of bad news. Bringing out a dual GPU (albeit with smaller dies, if this theory holds up of course) at launch?

 

[Ferrum Master]

Factually wrong.

Wrong. As long it uses GLSL it will use opengl as a core in application part. I haven't seen a native Vulkan engine. All Vulkan enabled games we have are vulkan ports with actually less efficiency as it actually should be. It is actually as always a code mess in reality than the all adverts tell you. ID won't recode all of their engine just in a flash, economically they won't even bother to do it.

 

[FordGT90Concept]

Hmmm, lets say it turns out to be true, wouldn't that be kind of a bearer of bad news. Bringing out a dual GPU (albeit with smaller dies, if this theory holds up of course) at launch?

DX12/Vulkan thrives on exactly that. AMD wants to move in the direction of multiple, smaller dies because of the modularity (manufacturing custom APUs) and savings on production costs (less of the wafer is lost to defects). The problem with DX11 and OpenGL is that the software side couldn't handle multiple GPUs. The driver had to do it all and it did a horrible job at it. DX12/Vulkan hands multiple GPU management over to the developers. GPUs are on the path to go through the same transition CPUs did over the last decade.

A dual Polaris card would be the cheapest (~$500-600 MSRP probably), best card for VR so I think there's a market it for it now albeit a small one.

 

[Caring1]

That was cross fire of the same card. (looked a bit misleading with cryptic card name). Note the CF at the end.

Not the cards I have seen, clearly marked 1.

 

[IceScreamer]

DX12/Vulkan thrives on exactly that. AMD wants to move in the direction of multiple, smaller dies because of the modularity (manufacturing custom APUs) and savings on production costs (less of the wafer is lost to defects). The problem with DX11 and OpenGL is that the software side couldn't handle multiple GPUs. The driver had to do it all and it did a horrible job at it. DX12/Vulkan hands multiple GPU management over to the developers. GPUs are on the path to go through the same transition CPUs did over the last decade.

A dual Polaris card would be the cheapest (~$500-600 MSRP probably), best card for VR so I think there's a market it for it now albeit a small one.

Yea, I remember the talk about multiple GPU's becoming the future, at least on AMD's part. So where does that put Vega chips exactly? If this whole 480 X-fired deal was to be true then the Vega chips should be plenty faster than the P10 chips crossfired, or the somewhat poor scaling (in those "leaked" 3D11 benches) is intentional so that this CF deal slots right under Vega. I don't know really, this speculation is getting out of hand, and I'm annoying myself and probably others.

 

[medi01]

Not the cards I have seen, clearly marked 1.

I'm talking about this chart, note how first bar is CF of 67DF:C7:

http://videocardz.com/60253/amd-radeon-r9-480-3dmark11-benchmarks

I think what C4 is what LIKELY WAS (they might rename it though) Polaris 10 - 480 and C7 is/was 480x.

I doubt C4 could be Polaris 11 (given it consumes MUCH less power and that they've officially demoed Polaris 11 vs 960 with capped 60fps)

 

[Caring1]

I'm talking about this chart....

You quoted me, and I wasn't talking about that chart, but I understand if you cant keep up with the conversation.

 

[librin.so.1]

Wrong. As long it uses GLSL it will use opengl as a core in application part.

First of all, GLSL is a shading language and is itself a separate spec from OpenGL.
And sure, if one enables the VK_NV_glsl_shader extension, one can load GLSL shaders into Vulkan. But that doesn't mean it needs to go through OpenGL. Because it doesn't.
Other than that, if one decides not to enable the VK_NV_glsl_shader extension, the only shaders Vulkan can load is SPIR-V, while SPIR-V can be compiled from anything. Heck, You could even use Visual F***ING Basic for shaders if You'd bother writing a VB -> SPIR-V compiler/translator.

I haven't seen a native Vulkan engine. All Vulkan enabled games we have are vulkan ports with actually less efficiency as it actually should be. It is actually as always a code mess in reality than the all adverts tell you. ID won't recode all of their engine just in a flash, economically they won't even bother to do it.

How does that make Vulkan depend on OpenGL? If a game uses a Vulkan renderer that is a wrapper on top of their OpenGL renderer, it doesn't mean Vulkan itself depends on OpenGL. What it does mean is "developers are lazy", no more, no less. Otherwise, by this same logic one could say "OpenGL uses Direct3D", since for many games, they don't bother writing a proper OpenGL rendered and instead just wrap their Direct3D renderer in OpenGL. *rolls eyes*

 

[FordGT90Concept]

Yea, I remember the talk about multiple GPU's becoming the future, at least on AMD's part. So where does that put Vega chips exactly? If this whole 480 X-fired deal was to be true then the Vega chips should be plenty faster than the P10 chips crossfired, or the somewhat poor scaling (in those "leaked" 3D11 benches) is intentional so that this CF deal slots right under Vega. I don't know really, this speculation is getting out of hand, and I'm annoying myself and probably others.

Vega is almost identical to Fury X in hardware (4096 shaders, HBM2 instead of HBM). There's no reason why they can't do a dual Vega too with the price adjusted to match. They can easily market the dual Polaris card as the budget VR card and dual Vega card as the performance VR card. Dual Polaris should be better at VR than a single Vega. They would likely be priced about the same with the latter being marketed towards 4K or Eyefinity gaming rather than VR.

I don't know how well 3D Mark handles multiple cards. If it is relying on Crossfire (which the "CF" would imply), 3D Mark likely isn't optimized for D3D12/Vulkan multi-GPU support. Considering previous testing was done with Ashes of Singularity, my guess is no. That 3D Mark benchmark, therefore, is likely close to a worst-case-scenario (ye olde, driver-implemented Crossfire) for a dual Polaris. It does not reflect the future of the multi-GPU technology.

2 x Fury X = 58% improvement over single using D3D12 multi-GPU in Ashes of Singularity
2 x Polaris 10 C7 = 43% improvement over single using Crossfire in 3D Mark

15% may not sound like much but if it was linear and translated to a D3D12 multi-GPU version of 3D Mark, 2 x Polaris 10 C7 would have gotten 28,535 which lands just above GTX 1080.

That said, 3D Mark only has a D3D11 benchmark which NVIDIA has a home-field advantage. Too bad we don't have any numbers for D3D12 Polaris.

This logic is flawed for two reasons:
1) 3D Mark only does as high as D3D11. GeForce has a tremendous advantage in D3D11 so the graph is intrinsically biased towards NVIDIA from the start.
2) Multi-GPU (non-driver implementation) requires D3D12 or Vulkan. Ashes of the Singularity was used for multi-GPU benchmarking. The numbers quoted above are correct.

As far as I know, there are no D3D12 Crossfire benchmarks available to guage how multi-GPU does compared to Crossfire. Further, assuming any of this has any bearing comparing D3D11 to D3D12 between vendors is simply foolish.

3D Mark 11 is a worst-case scenario for Polaris (because GCN has never been great at DX11) and multi-GPU (because it relies on ye olde driver-implemented Crossfire/SLI). Put the two together in a D3D12, dual Polaris 10 situation and I think the card would run away from GTX 1080.

Let's focus on Ashes of the Singularity:
DX11 versus DX12 = 7% boost for GCN
http://wccftech.com/nvidia-gtx-1080-async-compute-detailed/

D3D12 Multi-GPU versus single = 58%
http://www.anandtech.com/show/9740/directx-12-geforce-plus-radeon-mgpu-preview

Combined D3D11 single-GPU -> D3D12 multi-GPU = 65%
18,060 points (3DMark 11 single-GPU) * 1.65 = 29,799 points (theoretical 3DMark 12 multi-GPU)

And remember, this is throwing out the GCN async compute advantage because GeForce is terrible at it. Add in async (because what true D3D12 test wouldn't use it?) and dual Polaris 10 easily tops 30k (31,786 to be exact). If we adjust GTX 1080s scores to D3D12 + async, it's score would fall to 26,983 points.

I think what C4 is what LIKELY WAS (they might rename it though) Polaris 10 - 480 and C7 is/was 480x.

I doubt C4 could be Polaris 11 (given it consumes MUCH less power and that they've officially demoed Polaris 11 vs 960 with capped 60fps)

I'm thinking:
RX 480 = C7 CF
R9 480 = C7
R7 470 = C4

All three are Polaris 10 because Polaris 11 is 67FF.

 

[medi01]

I wasn't talking about that chart,

Care to link chart you were talking about?

I'm thinking:
RX 480 = C7 CF
R9 480 = C7
R7 470 = C4

All three are Polaris 10 because Polaris 11 is 67FF.

Sounds reasonable (and well in line with Raja's talk about how important mulit-gpu is and how AMD is going to educate developers).

And yikes. Risky strategy to say the least.

 

[Prima.Vera]

You guys should play the Stock Market. The amount of speculation on this thread is ubber ridiculous! )