AMD Polaris Thread: Radeon RX 480, RX 470 & RX 460 launching June 29th

[ShintaiDK]

You think you understand it better when no VR games even use paper spec features of Pascal? lol

2x potential with Pascal's VR hardware feature. Polaris is 10x potential.

I am not even making this up, because it does sound like a major case of "too good to be true", but here it's actually true. Go do a bit of reading on Foveated rendering.

Does it have a single pass rendering feature for example? No...

 

[ShintaiDK]

No, I am upgrading from a 290X 4GB to a 8GB 390X performance class with 2.5x better perf/w while making a profit to buy a cheap Steam game in the process.

I may setup a big Eth mining operation, the perf/w will make a lot of sense when it directly relates to profitability. AUS has similar kWh $ as Denmark, ridiculous 0.25 USD per kWh.

Its good you finally got your eyes up for perf/watt that others have enjoyed so long. However dont expect 2.5x

Power here is a lot more than 25 USD cents.

Good luck with the side grade

 

[Silverforce11]

Its good you finally got your eyes up for perf/watt that others have enjoyed so long. However dont expect 2.5x

Power here is a lot more than 25 USD cents.

I always appreciate perf/w. But just so you know, my custom 290X cost me $400 AUD. An equivalent performing 980 cost $779 AUD at the time, and even the 390X cost $550 AUD.

Perf/w matters, but not $379 difference.

I'll wait to see the final AUD price for the RX 480 8GB, if it's ~$350 AUD, I'll get 2 for CF. Both of them should have similar power consumption as a single 290X (or even less).

 

[Silverforce11]

A lot of gamers still have no idea about Foveated rendering, the next holy-grail of VR. This is why nobody raised the feature when the slides came out.

But if Polaris uarch has native hardware support for it, AMD is laying the foundation to dominate VR within the next few years.

Basic primer:

https://www.youtube.com/watch?v=6q3w0fiD0zg

https://www.youtube.com/watch?v=m4HSdz5lFpA

https://www.youtube.com/watch?v=mYdYD7FHZH0

Some folks were hyping up Pascal for VR, including me, when I found out it finally has fine-grained preemption support, but this is much bigger.

 

[flopper]

A lot of gamers still have no idea about Foveated rendering, the next holy-grail of VR. This is why nobody raised the feature when the slides came out.

But if Polaris uarch has native hardware support for it, AMD is laying the foundation to dominate VR within the next few years.

Basic primer:

https://www.youtube.com/watch?v=6q3w0fiD0zg

https://www.youtube.com/watch?v=m4HSdz5lFpA

https://www.youtube.com/watch?v=mYdYD7FHZH0

Some folks were hyping up Pascal for VR, including me, when I found out it finally has fine-grained preemption support, but this is much bigger.

Pascal is already old technology.
a downgrade for any amd owner as this shows how VR and amd works togheter with the 480


Threadcrapping and trolling are not allowed
Markfw900

 

[oussama-tn]

Yes... they did that for all launches that I recall. Either a OOS, or auto-notify.
Then, they flip the switch, and they are all in stock.

What are the differences between those cards ? except for warranties

 

[beginner99]

So short version, AMD fixed their tessellation. Guess that will be a non issue from now on, except for owners of older cards

No. AMD added hardware to fix games using CrapWorks.

 

[Dribble]

Nobody will buy polaris for VR, look at the steam VR thing it's barely faster then a 970. Stupidest marketing ever - try to appeal to the 0.1% of buyers with $800 VR systems with your $200 min spec card. (a) anyone blowing $800 on VR already has at least a 970 level of graphics (b) they will spend a lot more than $200 on the gpu upgrade. For those that say well buy polaris today and then get VR cheap in 2 years time - by the time VR gets that cheap everything will have moved on and polaris will be way beneath min spec.

VR is something for the top end vega chip not polaris.

 

[thedavexp]

Nobody will buy polaris for VR, look at the steam VR thing it's barely faster then a 970. Stupidest marketing ever - try to appeal to the 0.1% of buyers with $800 VR systems with your $200 min spec card. (a) anyone blowing $800 on VR already has at least a 970 level of graphics (b) they will spend a lot more than $200 on the gpu upgrade. For those that say well buy polaris today and then get VR cheap in 2 years time - by the time VR gets that cheap everything will have moved on and polaris will be way beneath min spec.

VR is something for the top end vega chip not polaris.

NDA is over soon so I would wait for benchmarks before jumping to conclusions on its performance

 

[Silverforce11]

Nobody will buy polaris for VR, look at the steam VR thing it's barely faster then a 970. Stupidest marketing ever - try to appeal to the 0.1% of buyers with $800 VR systems with your $200 min spec card. (a) anyone blowing $800 on VR already has at least a 970 level of graphics (b) they will spend a lot more than $200 on the gpu upgrade. For those that say well buy polaris today and then get VR cheap in 2 years time - by the time VR gets that cheap everything will have moved on and polaris will be way beneath min spec.

VR is something for the top end vega chip not polaris.

Have you even had a look at the VR games available? None of them are graphically demanding. The most is Project Cars and that thing makes people sick.

Most of the VR games out there runs at high FPS on a 970. The biggest issue is actually motion to photon latency to minimize nausea. Not raw performance.

 

[Dribble]

NDA is over soon so I would wait for benchmarks before jumping to conclusions on its performance

It gets 6.8. For comparison a 970 gets 6.5, and 1070, 1080, titan all max it out with a score of 11:
http://wccftech.com/radeon-rx-480-steam-vr-benchmark-most-affordable

 

[renderstate]

is it? I don't think so. Their hardware does tessellation faster. Now amd wants to fight this with a cheat and skip tessellation which has no visual impact.

There is nothing cheap with what AMD is doing, quite the opposite. It's smart to get rid of polygons that don't contribute anything to the image as early as possible in the pipeline to avoid bubbles/stalls.

 

[StinkyPinky]

Geez, not even one site has "accidently" uploaded their review early and then "pulled" it a few minutes later after they realise their "mistake". What's going on?

 

[Thala]

is it? I don't think so. Their hardware does tessellation faster. Now amd wants to fight this with a cheat and skip tessellation which has no visual impact

You seem to not understand. The discard feature is not directly related to tesselation. It checks any primitive for visibility.
With your argument you could also declare z-buffer checks as cheating, as it discards pixels, which are not visible.

It's smart to get rid of polygons that don't contribute anything to the image as early as possible in the pipeline to avoid bubbles/stalls.

It is not so much about bubbles/stalls but more about avoiding unnecessary work in the fragment shaders.

 

[renderstate]

You think you understand it better when no VR games even use paper spec features of Pascal? lol



2x potential with Pascal's VR hardware feature. Polaris is 10x potential.



I am not even making this up, because it does sound like a major case of "too good to be true", but here it's actually true. Go do a bit of reading on Foveated rendering.

You are definitely making it up since you showed no numbers and no understanding of how VR rendering works. Stop trolling.

 

[renderstate]

No. AMD added hardware to fix games using CrapWorks.

Not true. AMD simply fixed an issue with their architecture that console developers work around in SW by doing culling in compute. It's simply common sense and I am really surprised AMD HW did not support this before the 480.

 

[sontin]

You are definitely making it up since you showed no numbers and no understanding of how VR rendering works. Stop trolling.

He has no clue what he is talking about. Since the Pascal launch he sounds so bitter... :\

Valve is using Foveated Rendering in combination with nVidia's Multi -Resolution Shading - slides 18++: http://alex.vlachos.com/graphics/Alex_Vlachos_Advanced_VR_Rendering_Performance_GDC2016.pdf

Polaris is maybe as good as Maxwell but Pascal is way ahead when it comes to VR.

 

[Erenhardt]

You seem to not understand. The discard feature is not directly related to tesselation. It checks any primitive for visibility.
With your argument you could also declare z-buffer checks as cheating, as it discards pixels, which are not visible.

Well, if you are running on 400m sprint and cut corners along the way, the end result is the same - you are on the finish line, but you didn't to the work you were supposed to do.

Anyway, that is very subjective and oftopic

Polaris is maybe as good as Maxwell but Pascal is way ahead when it comes to VR.

LOL.

 

[Thala]

You are definitely making it up since you showed no numbers and no understanding of how VR rendering works. Stop trolling.

Both features (multi-viewport single-pass and foveated rendering) are currently not employed in any game i am aware of.
And then again, if looking closer at both features:
1) foveated rendering also brings benefits to NVidia cards, its just not hw accelerated. So the real factor AMD vs NVidia is much smaller than 10x
2) I cannot imagine any realistic situation where multi-viewport single-pass brings anywhere close to 2x gain. The saving can only be in the geometry stage. Anything after (and including) the viewport transform are different for both eyes.

So i would be very careful when comparing both features with advertised numbers.

 

[ShintaiDK]

GCN 1.2

GCN 1.3

Slides dont add up.

 

[Silverforce11]

He has no clue what he is talking about. Since the Pascal launch he sounds so bitter... :\

Lol what, bitter? Me, the guy who claims Pascal will finally bring fine-grained preemption to NV GPUs long before it was published in their whitepapder... after exposing their lies about that feature in Maxwell.. Me, the guy who said Pascal is finally good for VR. It's finally able to do what GCN 1 does. Congrats.

I also called it early, no real Async Compute in Pascal either. But it didn't stop NV lying about that feature support, again.

Now time for NV to catch up, maybe in Volta when they add hardware Foveated rendering like Polaris has, where the real perf gains are to be had AND required for accurate display output for human eye/brain perception to make VR indistinguishable from reality.

Now, if you don't have anything to talk about for Polaris, get out and stop trashing it.

 

[Leadbox]

GCN 1.2
http://images.anandtech.com/doci/8460/ColorCompress_575px.png

GCN 1.3
[img]http://cdn.videocardz.com/1/2016/06/Radeon-RX-480-Presentation-VideoCardz_com-15-900x506.jpg

Slides dont add up.[/QUOTE]
What are you trying to add?

 

[Erenhardt]

What are you trying to add?

Apples to oranges.

One is bandwidth throughput the other is firestrike per theoretical max bandiwdth.

 

[renderstate]

Both features (multi-viewport single-pass and foveated rendering) are currently not employed in any game i am aware of.
And then again, if looking closer at both features:
1) foveated rendering also brings benefits to NVidia cards, its just not hw accelerated. So the real factor AMD vs NVidia is much smaller than 10x
2) I cannot imagine any realistic situation where multi-viewport single-pass brings anywhere close to 2x gain. The saving can only be in the geometry stage. Anything after (and including) the viewport transform are different for both eyes.

So i would be very careful when comparing both features with advertised numbers.

You don't understand and you are drinking AMD kool aid slides. The "foveated rendering" they mention has been pioneered by NVIDIA in their VR SDK and it runs in a single pass on Maxwell without using the slow geometry shader path. The 480 doesn't support this in a single fast pass according their slide. So no, they are not accelerating foveated rendering in HW, while NVIDIA has been doing it since Maxwell.

 

[antihelten]

You are definitely making it up since you showed no numbers and no understanding of how VR rendering works. Stop trolling.

The 10x number that Silverforce mentioned is fairly well established and actually on the low side.

Microsofts research in foveated rendering achieved a 5-6x speedup on a normal monitor with a FOV of 47 degrees. On a display with a higher FOV (i.e. a HMD) such as say 100 degrees (roughly what the Rift has), the speedup is expected to be around 30-40x.

The only real issue here is figuring out how to put a pair of high speed cameras inside a HMD to track eye-movement. This is of course a fairly sizable issue from en engineering perspective, but not an insurmountable one. Several companies are working on it, with the most well known probable being Fove

Microsoft research paper

You don't understand and you are drinking AMD kool aid slides. The "foveated rendering" they mention has been pioneered by NVIDIA in their VR SDK and it runs in a single pass on Maxwell without using the slow geometry shader path. The 480 doesn't support this in a single fast pass according their slide. So no, they are not accelerating foveated rendering in HW, while NVIDIA has been doing it since Maxwell.

What Nvidia does (Multi-res shading and lens matched shading) is known as fixed foveated rendering, and it is a pale imitation of proper foveated rendering.

Furthermore based on the latest Polaris slides AMD would appear to have is capable of exactly the same thing.

It's also worth noting that these techniques, whilst nice to have, are hardly revolutionary, with speedups probably being in the 1.1-1.2x range (based on Valve's research in fixed foveated rendering).

Single pass stereo rendering for the vertex stage has nothing to do with foveated rendering by the way and is a feature that can be applied either separately or in conjunction with the above stuff.