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

[3DVagabond]

That's fine, but the results imply that p10 is gimped in dx12 vs 11. The aots (dx12) benchmarks are implying way way worse performance vs both amd AND nvidia cards. Than these leaks.

Previously we saw amd dx12 vs dx11 was superior vs nvidia, and that nvidia basically performed the same in both. P10 -if you believe the benches as well as the aots Computex demo - is materially inferior in dx12.

One of the following must be true:

-The leaks are wrong

-P10 has such extraordinary command processor benefits that its gap over gtx 980 is greater in dx11 vs dx12

-amd purposely downplayed p10 performance at Computex

-there were driver issues in all of the aots benches we've seen thusfar

Take your pick. I have no idea.

I understand you are trying to paint a picture with the various info you have. You can do that with enough verified info. To take various small bits of unverifiable, and likely in some cases fake info, is going to create an inaccurate conclusion.

I think we need to look at each individually and decide the possible validity. Even if they are "valid" we know we are dealing with ES's at different clocks with different drivers on different systems, etc...

 

[3DVagabond]

IMO it means that for $200 you get to pass the min requirements for performance as specified by Oculus and Valve.

That's too complicated for some to figure out, apparently. Even though that's exactly what AMD has been repeating for a while now.

 

[3DVagabond]

If all these leaks are correc and you're correct, then amd picked the worst game possible to demonstrate Polaris 10 at Computex. They should have picked a dx11 game.

So, do you think all of these leaks are correct? Because that's the only way there's even a discussion here.

 

[cusideabelincoln]

Seems too good to be true. AMD isn't going to sell a card for ~$230 that's only 20% off of a 1070, especially when that would make a full die Polaris 10 around equal performance for ~$100 less at the moment.

You can argue about AMD wanting to grow the VR market and all kinds of other stuff, but they're still a business and at that performance level it's leaving too much money on the table.

They've done something similar before. Actually it was probably a big business blunder than Polaris could be (I doubt it, because AMD really does need to gain market share), but it won them a lot of fans and forced Nvidia to respond. The HD 4870 released to be faster than the GTX 260 and was $200 cheaper.

 

[Vesku]

So you are saying he photoshopped whole benchmark in real time?

Why he is only forum warrior, then? With that skills he could be working in any movie industry including porn.

I suppose he photoshopped whole recording of video from benchmarking the GPU .

If it is a bios edit trick, same thing used by scammers to sell outdated GPUs on eBay, then no special video editing skills are necessary. Could be an OCed 390X or something.

 

[Armsdealer]

So, do you think all of these leaks are correct? Because that's the only way there's even a discussion here.

At first I didn't think so because these results are way off from the computex results, but the argument that amd didn't want to reveal their hand so early is perfectly reasonable.

Moreover, it is an actual video that 'misty rain' (a guy who has a desktop wallpaper of himself playing with a cat) has posted, not an image. I'm leaning towards believing that 3515 figure, which means an achievable overclock on rx480 puts it in 1070 territory, not 980ti territory as the most optimistic amongst us thought.

 

[Sweepr]

I knew that guy wasn't strange.

It's the same person who talked about a HBM version, 2816 SPs, 384-bit and 97% Radeon R9 390X performance.

 

[psolord]

It doesn't matter that there is a video? Because you say so?

Is it possible for R9 290 Lightining to get 1000 pts. more than on stock clocks in 3dMark Firestrike Ultra and get to Fury levels?

What it it's a Hawaii card, with edited BIOS, overclocked and with every detail setting disabled in the control panel?

Or maybe he is using a custom radeon pro profile with reduced settings or something.

 

[psolord]

Seems too good to be true. AMD isn't going to sell a card for ~$230 that's only 20% off of a 1070, especially when that would make a full die Polaris 10 around equal performance for ~$100 less at the moment.

You can argue about AMD wanting to grow the VR market and all kinds of other stuff, but they're still a business and at that performance level it's leaving too much money on the table.

They did that before. They can do it again.

Also don't forget that Nvidia has better cpu usage in DX11 and people with older systems could still prefer an Nvidia card anyway. I know DX12 is the future but there's still a huge backlog of DX11 games out there and more are coming out all the time.

 

[lukart]

Pricing is from what we know:
RX 480 4 GB - 199$
RX 480 8 GB - 229$
? - <300$

I think its more like
RX 480 4 GB - 199$
RX 480 4 GB OC - 219$
RX 480 8 GB - 239 or even 249$ for the oc models

 

[Mopetar]

They've done something similar before. Actually it was probably a big business blunder than Polaris could be (I doubt it, because AMD really does need to gain market share), but it won them a lot of fans and forced Nvidia to respond. The HD 4870 released to be faster than the GTX 260 and was $200 cheaper.

If those fans were worth anything they'd not be in quite the precarious situation that they are. If they wanted to, they could probably sell Polaris 10 chips for even less, but the simple truth is that AMD needs to return to profitability.

 

[cusideabelincoln]

They're in this position because they suck at marketing, and are hamstrung by other side of AMD that is failing. But that has nothing to do with anything. The point is there is precedent.

 

[3DVagabond]

I think its more like
RX 480 4 GB - 199$
RX 480 4 GB OC - 219$
RX 480 8 GB - 239 or even 249$ for the oc models

Who knows what the AIB's are going to charge for custom and/or O/C models. All AMD talks are reference models.

 

[sirmo]

If those fans were worth anything they'd not be in quite the precarious situation that they are. If they wanted to, they could probably sell Polaris 10 chips for even less, but the simple truth is that AMD needs to return to profitability.

Bulldozer has really hurt AMDs image as a whole. And I assume AMD has lost a lot of fans after that debacle.

I like AMD but I will buy whatever card is best for my needs. I did make a mistake buying the gtx780.. worst card I ever bought. And it's probably the last Nvidia card I buy as long as AMD is competitive.

AMD has also said they are returning to profitability in the second part of this year.

 

[Silverforce11]

28nm GCN products had bottlenecks that prevented full resource utilization in DX11. That was most notable with the Fiji cards, but it existed across the board to some extent. AotS in DX12 was one of the very few games that could push 28nm GCN cards near their theoretical limits.

If Polaris fixes the bottlenecks that plagued previous GCN generations, then it wouldn't be surprising to see the gap close between DX11 and DX12. That would mean even if the raw horsepower (and DX12 performance in AotS) was similar to Hawaii, it would do better in most real-world DX11 games and benchmarks.

This man gets it.

AMD's Command Processor is inefficient in DX11 because it was designed to work alongside ACEs which DX11 cannot access, ie. Graphics Queues via CP, Compute/Copy queues via ACEs.

It also lacks a deep queue buffer, so if the CPU is hammered, it stalls. NV's Gigathread has a deep buffer and it receives the queues from the software (driver) scheduler that's multi-threaded on the CPU, so it rarely gets stalled. In DX12, it loses it's advantage, and even suffers slightly via software scheduling overhead.

When GCN is driven under DX12, the CP can be fed with multi-core CPUs and so it's operating at a higher efficiency. Add ACEs to distribute compute queues and it offloads some of the work from the CP, as it was designed to work together.

This is why we see DX12 performance is reflective of the Tflops of GCN.

Polaris gets a lot of improvements to enhance it's performance in DX11. A new CP, Pre-Fetch, Cache system are all things that address the critical flaws by removing bottlenecks in DX11. I don't know whether this would improve it's DX12, but given how well current GCN runs in DX12, I have to wonder if there's room left to improve.

What it may mean is that if it's 5.5 TFlops, in DX12, it's going to be on-par with the 390X, but in DX11, it will be ahead. The biggest unknown though is how effective the Discard Accelerator will function. That may really skew the performance edge to Polaris.

 

[sirmo]

The biggest unknown though is how effective the Discard Accelerator will function. That may really skew the performance edge to Polaris.

Primitive discard accelerator shouldn't really make much of a difference on well optimized games. This is because game developers should be doing the culling to begin with.

Incidentally however where this might help a lot is to address Gameworks benchmark cheating by Nvidia.

 

[Armsdealer]

This man gets it.

AMD's Command Processor is inefficient in DX11 because it was designed to work alongside ACEs which DX11 cannot access, ie. Graphics Queues via CP, Compute/Copy queues via ACEs.

I think we do all appreciate this. What you should appreciate is that the dx11 performance that's now being leaked is not just similar to the dx12 performance we've seen, it's way higher. Ie the gap hasn't just closed, it's gone way in the other direction (ie dx11 is now amd's preferable api by a lot if amd wasn't downplaying rx480 performance at Computex).

The gtx 1080 gets a bit short of 5000 fs ultra points. Use whatever cf scaling factor you want and tell me how much faster two rx480s are vs a 1080 if the rx480 is scoring 3515. Get it?

To me the obvious answer was the new leaks are fake, but maybe it's just that amd just didn't want to reveal their hand at Computex because these new leak videos do seem pretty legit.

 

[casiofx]

Primitive discard accelerator shouldn't really make much of a difference on well optimized games. This is because game developers should be doing the culling to begin with.

Incidentally however where this might help a lot is to address Gameworks benchmark cheating by Nvidia.

That feature would work wonders in FPS games, lots of hidden geometries.

 

[Maverick177]

I knew that guy wasn't strange.

It's the same person who talked about a HBM version, 2816 SPs, 384-bit and 97% Radeon R9 390X performance.

That'd put the RX 480 on par with a regular Fury, sidegrade for me 100%.

 

[tential]

That's too complicated for some to figure out, apparently. Even though that's exactly what AMD has been repeating for a while now.

And many are stating that is useless. Who cares about being at the minimum spec for VR when you can't afford the headset? By the time you can, you're going to be below the specs for the actual games out. Not like there are many games out right now....Who wants to buy the minimum for tech they don't plan on using for at least a couple of years? Not me.

VR is too niche to matter to most $200 buyers.

 

[Stuka87]

That'd put the RX 480 on par with a regular Fury, sidegrade for me 100%.

Polaris is not intended for owners of Fury cards. You will want to wait for Vega. Unless you really want 8GB of vram now.

 

[zinfamous]

Unless amd downplayed rx480 perf in aots, the numbers don't add up. In aots, we're clearly seeing dx12 perf below 390x, and closer to 390, but slightly above 980, and now in dx11 we're seeing it way above 390x and way above 980.

Maybe it's just that there were driver issues at that point? Otherwise that means polaris dx12 vs 11 is worse than nvidia?

so now AMD is sabotaging their own leaks with their own turdy press release drivers.


...we are through the 5th wall, people.

 

[tential]

Why even bother analyzing AMD's AOTS results? They were so shady who cares. Just wait for reviews/more info. AMD is clearly just pulling our chain for as long as possible. I personally don't like it.

 

[zinfamous]

IMO it means that for $200 you get to pass the min requirements for performance as specified by Oculus and Valve.

That's too complicated for some to figure out, apparently. Even though that's exactly what AMD has been repeating for a while now.

You're both right.


Not to be outdone, nVidia allows you to pass those minimum specs at the generous cost of $370-400.

 

[Maverick177]

Polaris is not intended for owners of Fury cards. You will want to wait for Vega. Unless you really want 8GB of vram now.

10% above 390X is around Fury level sadly, so selling my Fury for around $250 ~ $300 and get the Polaris 10 is justified, I think.