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

[amenx]

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.

By the time VR matters/matures, with enough supporting titles and improved, better priced VR hardware, we'll probably already be on Volta, Navi. I doubt many people will factor in VR performance when looking at lower-mid range Polaris.

 

[Gikaseixas]

You don't understand and you are drinking AMD kool aid slides

And you could drop the condescending tone and insults by removing what i just quoted.

On topic:
I thought i would wake up and read a bunch of reviews. No 00h01 reviews for this launch

 

[Silverforce11]

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

I didn't want to post the 30-40X because it sounds too crazy to be true (it is true), so I settled on 10x.

Despite that, some shills just attacked me, accusing me of being clueless despite facts being on my side (clearly they didn't do their homework before being fast on the insults).

Foveated rendering is the holy grail of VR. It's what all the major players are working towards. Being able to hardware accelerate it is a step towards that goal of fully functional 30-40x speed improvements as well as visually correct VR output.

 

[antihelten]

Foveated rendering is the holy grail of VR. It's what all the major players are working towards. Being able to hardware accelerate it is a step towards that goal of fully functional 30-40x speed improvements as well as visually correct VR output.

It is arguably the holy grail of VR (or at least one of them), but it will probably be a few more years before we see it in a consumer HMD, and by that time we will probably be on 10nm GPUs, so whether or not a given vendor has hardware support for it right now probably won't matter all that much.

 

[Silverforce11]

https://www.youtube.com/watch?v=5CJBaZ9V2Eo

^ First review online!

Very strange, in older games it's lagging behind the 290X/390, but some more recent ones it's ahead. Especially GameWorks titles. Hmm.

 

[ShintaiDK]

https://www.youtube.com/watch?v=5CJBaZ9V2Eo

^ First review online!

Slower than 290X and 390 unless tessellation is used.

Barely any OC at all. Less than 2%.

 

[beginner99]

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.

Hint: NV doesn't support it at all, also not in Pascal. Difference in last gen was that NV just had faster tessellation and abused that in CrapWorks to cripple AMD. But NV cards suffered as well, especially older ones, just not as much as AMDs. Thats actually why it is crap. It helps no one. Even NVs own users suffer from it.

And software culling is absolutely standard. It is also done in Frostbite engine for years because it obviously added value for both brands of cards and on consoles.

http://www.slideshare.net/DICEStudio/culling-the-battlefield-data-oriented-design-in-practice

http://www.frostbite.com/2016/03/optimizing-the-graphics-pipeline-with-compute/

 

[Flapdrol1337]

Slower than 290X and 390.

except in fallout 4 where it has a big lead for some reason.

 

[Head1985]

slower than 290x and 390 unless tessellation is used.

Barely any oc at all. Less than 2%.

omg 32rops....Yep another <redacted> up design after fury...


cussing is not allowed in the technical forums
Markfw900

 

[selni]

except in fallout 4 where it has a big lead for some reason.

Yeah some big jumps but mostly pretty close at 1080p and lower at 1440p. Ouch, maybe Kyle wasn't just ranting?

Should still sell a bunch given nvidia's pricing though, it's not a bad deal at all.

 

[krumme]

Really looking forward to all the 200 usd card review comparisons eg. 380, 960 and 480 4GB.

 

[selni]

Really looking forward to all the 200 usd card review comparisons eg. 380, 960 and 480 4GB.

It's going to murder everything out there in that price range by the looks.

 

[krumme]

Anyones guess where the card ends up on TPU perf/cost evaluation on 1080?
https://www.techpowerup.com/reviews/MSI/GTX_1070_Gaming_X/26.html

Whats your guess?

 

[Silverforce11]

except in fallout 4 where it has a big lead for some reason.

It runs faster than 290X and 390 in all the GameWorks title, just look at Just Cause 3, ACU etc. Looks like that primitive discard accelerator is allowing it to avoid gimping and run faster.

Weird to see it under perform in older games and in more recent ones, it's doing most likely above 390X.

 

[thedavexp]

It runs faster than 290X and 390 in all the GameWorks title, just look at Just Cause 3, ACU etc. Looks like that primitive discard accelerator is allowing it to avoid gimping and run faster.

Weird to see it under perform in older games and in more recent ones, it's doing most likely above 390X.

Odd that they included a really OCed 970 rather than stock like the other cards

 

[sze5003]

Well I forgot to work from home today but I did check newegg and it looks like they still have auto notify on. Will wait for reviews and feedback here before I decide what I'm going to do.

 

[antihelten]

Slower than 290X and 390 unless tessellation is used.

Barely any OC at all. Less than 2%.

Depends upon which resolution you look at.

1440P:
3% slower than 390
1% slower than 290X

1080P:
1% faster than 390
3% faster than 290X

This would put the 480 at 62-64% on TPU's 1440p chart and 63-65% on their 1080P chart.

 

[Thala]

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

Oh dear. Now you are starting to mix things up. Foveated rendering and multi-viewport single-pass are orthogonal features. They are both very well researched techniques, which would benefit rendering for VR.
If you would have read carefully what i wrote, i already acknowledged, that foveated rendering can be implemented with any traditional GPU and you will see huge benefits.

So what you are claiming is, that RX480 does not have any hardware support for foveated rendering and thus no advantage at all over other architectures, despite it being mentioned in the slides?

 

[tg2708]

Saw a tech of tomorrow review and I have to say the card is a bit underwhelming and is also undeserving of the hype it has received. Very good price so lets hope with newer drivers the lead widens when compared to older gen cards.

 

[Silverforce11]

omg 32rops....Yep another f0ck up design after fury...

I wanna see from reputable sites like AT, Computerbase, Guru3d & TPU before making that call.

But yes, if it's performance drops off as resolution rises, it's strong indicator of ROP bottleneck.

 

[Head1985]

I wanna see from reputable sites like AT, Computerbase, Guru3d & TPU before making that call.

But yes, if it's performance drops off as resolution rises, it's strong indicator of ROP bottleneck.

Well it should be at 390x/GTX980 performance.Like i said before if Its Under 390x then its 100% because only 32Rops.

I cant help but 2048SP with 64Rops will be alot faster than 2304SP with 32rops..

Its like another FIJI-4096SP for nothing

 

[Gikaseixas]

Saw a tech of tomorrow review and I have to say the card is a bit underwhelming and is also undeserving of the hype it has received. Very good price so lets hope with newer drivers the lead widens when compared to older gen cards.

Well i don't blame AMD for beating by large margins last gen same class cards such as GTX 960 and 380X.

Now, all those who raised expectations just from rumors and early leaks were just setting themselves for disappointment.

Some people wanted it to do what Radeon 4800 did to 260GTX and 280GTX and by the specs, one could tell it wouldn't happen, specially when AMD positioned it to replace the 380X, so getting close or beating 390X in some titles is a heck of an achievement.

 

[96Firebird]

$270 at Newegg for the XFX reference model that is OC'd to 1328MHz...

 

[MTDEW]

Maybe it's because I'm not looking to upgrade yet since I have a 390x.
But I really don't see the problem, isn't it performing right where AMD said it would?

And...

AIB partners are reporting to me this morning that with extremely good custom air coolers the 480 GPUs are seeing from 1480MHz to 1600MHz clocks, but tell me it is a "lottery draw" on GPUs.

And considering the source you KNOW this is most likely a true statement also, so waiting for AIB versions may be worth it too.

 

[3DVagabond]

omg 32rops....Yep another <redacted> up design after fury...

So, if it had 64 ROPs would it still be 232mm² and $200? Is there a better combination at that size/price?