Regarding the RX 480 AOTS benchmark at Computex

[Fulgrim]

Firstly, sorry if this isn't the right place for this but i thought it'd garner some attention better here.

@Linus Tech  @Slick

Now, on to the topic at hand. When watching the Computex livestream i did notice that the visual look of each benchmark differed slightly, and the presumption that people jumped to was that the RX 480 benchmark, had lower settings to be able to match/beat the framerate of a GTX1080, when running two RX 480's in crossfire. It seems that in fact the numbers are legit, and not cherry picked/altered to make the card seem better than what it is. AMD_Robert from Reddit cleared this up.

 

Skip to 20:38 for the AOTS section.

 

Quote

Hi. Now that I'm off of my 10-hour airplane ride to Oz, and I have reliable internet, I can share some insight.
System specs:
CPU: i7 5930K
RAM: 32GB DDR4-2400Mhz
Motherboard: Asrock X99M Killer
GPU config 1: 2x Radeon RX 480 @ PCIE 3.0 x16 for each GPU
GPU config 2: Founders Edition GTX 1080
OS: Win 10 64bit
AMD Driver: 16.30-160525n-230356E
NV Driver: 368.19
In Game Settings for both configs: Crazy Settings | 1080P | 8x MSAA | VSYNC OFF
Ashes Game Version: v1.12.19928
Benchmark results:
2x Radeon RX 480 - 62.5 fps | Single Batch GPU Util: 51% | Med Batch GPU Util: 71.9 | Heavy Batch GPU Util: 92.3% GTX 1080 – 58.7 fps | Single Batch GPU Util: 98.7%| Med Batch GPU Util: 97.9% | Heavy Batch GPU Util: 98.7%
The elephant in the room:
Ashes uses procedural generation based on a randomized seed at launch. The benchmark does look slightly different every time it is run. But that, many have noted, does not fully explain the quality difference people noticed.
At present the GTX 1080 is incorrectly executing the terrain shaders responsible for populating the environment with the appropriate amount of snow. The GTX 1080 is doing less work to render AOTS than it otherwise would if the shader were being run properly. Snow is somewhat flat and boring in color compared to shiny rocks, which gives the illusion that less is being rendered, but this is an incorrect interpretation of how the terrain shaders are functioning in this title.
The content being rendered by the RX 480--the one with greater snow coverage in the side-by-side (the left in these images )--is the correct execution of the terrain shaders.
So, even with fudgy image quality on the GTX 1080 that could improve their performance a few percent, dual RX 480 still came out ahead.
As a parting note, I will mention we ran this test 10x prior to going on-stage to confirm the performance delta was accurate. Moving up to 1440p at the same settings maintains the same performance delta within +/-1%.

Source

Below are 1440p benchmarks done of AOTS (not used in the livestream)
GTX 1080 AOTS benchmark

RX 480 Crossfire AOTS benchmark

Update: 

 

Thanks to @Notional for mentioning this tweet.

[Indus Monk]

2 minutes ago, Fulgrim said:

2x Radeon RX 480 - 62.5 fps

 

2 minutes ago, Fulgrim said:

GTX 1080 – 58.7 fps

$400 GPU setup beats a $600 GPU! Holy Balls!

[Tedol]

Stream compress the video, so noone can say if the setting was the same on both benchmarks 

Wait 27 day, and we will know, if it benchmark are true 

[Guuhan]

I give up.

[ivan134]

Async drivers for Paxwell coming soon TM? 

[zMeul]

4 minutes ago, Tedol said:

Stream compress the video, so noone can say if the setting was the same on both benchmarks 

the compression applies to the video - the left side is compressed the same amount as the right side

blaming the compression for the differences in the same video is utter nonsense 

[Nena Trinity]

2 minutes ago, ivan134 said:

Async drivers for Paxwell coming soon TM? 

I heard Nvidia claim it decrease performance so I doubt it...

[Tedol]

1 minute ago, zMeul said:

the compression applies to the video not the each individual sources - the left side is compressed the same amount as the right side

blaming the compression for the differences in the same video is utter nonsense 

ok

[Fulgrim]

4 minutes ago, ivan134 said:

Async drivers for Paxwell coming soon TM? 

Nvidia still implement Asynchronous Compute via software, which is inferior to AMD's hardware level Async Compute.

[TidaLWaveZ]

I'm still skeptical, I think the one on the right looks noticeably better.

[zMeul]

11 minutes ago, Fulgrim said:

the numbers are legit

AMD has yet to provide links to the actual runs showing the exact results as showed by AMD on stage

no one else seems to find them either

[ivan134]

2 minutes ago, Fulgrim said:

Nvidia still implement Asynchronous Compute via software, which is inferior to AMD's hardware level Async Compute.

That's the joke..... 

[suicidalfranco]

now how are people here going to find a new drama to focus on with this new batch of data

[Fulgrim]

1 hour ago, zMeul said:

AMD has yet to provide links to the actual runs showing the exact results as showed by AMD on stage

no one else seems to find them either

Not the ones used at computex, but here's some benchmarks

RX 480 Crossfire

http://www.ashesofthesingularity.com/metaverse#/personas/b0db0294-8cab-4399-8815-f956a670b68f/match-details/ac88258f-4541-408e-8234-f9e96febe303

GTX 1080

http://www.ashesofthesingularity.com/metaverse#/personas/b0db0294-8cab-4399-8815-f956a670b68f/match-details/a957db0f-59b3-4394-84cc-2ba0170ab699

 

[Notree]

Nvidia fan?  ...

[Trixanity]

29th will reveal everything, so no need to get one's panties in a bunch so early.

[GlassBomb]

13 minutes ago, AmbarChakrabarti said:

$400 GPU setup beats a $600 GPU! Holy Balls!

Right. Dual-GPU setup is often times cheaper, and higher performing than a single, more expensive card. But you get the drawbacks of high power consumption, and the normal jazz of issues that you have to deal with.

 

And with the that said, if you want the 8GB of vRAM, then you need another ~20 dollars per card as well.

 

Edit: Either way. I'm going to do what I always do. Wait until consumers get their hands on the actual consumer hardware.

[No Nrg]

I still think it's a bit ridiculous that people continue to use this game no one even plays as a benchmark in performance and call it definitive proof of one being better than the other.

 

Show me benchmarks from a whole list of games with one beating the other consistently across a majority of the list and maybe I won't call bullshit.

[LeStringMan]

2 minutes ago, ivan134 said:

That's the joke..... 

so what you're saying is Fermi is clearly superior to Pascal, since it actually has hardware async compute? Obviously GTX 580 master race.

[Shoob]

1 minute ago, Fulgrim said:

RX 480 Crossfire

http://www.ashesofthesingularity.com/metaverse#/personas/b0db0294-8cab-4399-8815-f956a670b68f/match-details/ac88258f-4541-408e-8234-f9e96febe303

GTX 1080

http://www.ashesofthesingularity.com/metaverse#/personas/b0db0294-8cab-4399-8815-f956a670b68f/match-details/a957db0f-59b3-4394-84cc-2ba0170ab699

 

Those are 1440p benchmarks. The results on the slide were from a 1080p benchmark. 

They are still pretty believable, but some people just love bandwagoning.

[Indus Monk]

2 minutes ago, GlassBomb said:

Right. Dual-GPU setup is often times cheaper, and higher performing than a single, more expensive card. But you get the drawbacks of high power consumption, and the normal jazz of issues that you have to deal with.

 

And with the that said, if you want the 8GB of vRAM, then you need another ~20 dollars per card as well.

 

Edit: Either way. I'm going to do what I always do. Wait until consumers get their hands on the actual consumer hardware.

i forgot that 2 GTX 950 > 1 GTX 970

 

But I completely agree with you! 

[TidaLWaveZ]

49 minutes ago, AmbarChakrabarti said:

 

$400 GPU setup beats a $600 GPU! Holy Balls!

 

36 minutes ago, GlassBomb said:

Right. Dual-GPU setup is often times cheaper, and higher performing than a single, more expensive card. But you get the drawbacks of high power consumption, and the normal jazz of issues that you have to deal with.

 

And with the that said, if you want the 8GB of vRAM, then you need another ~20 dollars per card as well.

EDIT:

 

It's a "< $500" solution so it's very likely the 8GB version.

[Trixanity]

As for picking AotS and utilizing 2 x 480: it's quite simple. AotS is usually a favorable outcome for AMD cards, hence the results look good. Picking 2 x 480 means that it beats the 1080, so the results look good. Picking a single card only would mean it lost and then people would say "it's slower" "but it's $500 less" "yes but it's slower" "it's much cheaper" "Fast. Me like fast. Fast is gooder" etc.

Point being: it's bad to show the competitor performing better in your marketing even though it makes perfect sense and even if your product is overall superior (generally speaking).

 

So here AMD wants to showcase that they can beat a 1080 with their new product by putting two of them together for an overall still much cheaper solution that probably clocks in at or below $500 for 2 x 480 8 GB with aftermarket coolers. All this just to avoid detractors saying AMD isn't beating Nvidia's latest and greatest. Obviously all marketing smoke and mirrors but necessary smoke and mirrors it is.

[ivan134]

7 minutes ago, No Nrg said:

I still think it's a bit ridiculous that people continue to use this game no one even plays as a benchmark in performance and call it definitive proof of one being better than the other.

 

Show me benchmarks from a whole list of games with one beating the other consistently across a majority of the list and maybe I won't call bullshit.

Dumb. How is number of people who play it relevant to testing the dx 12 support of cards? Would number of people playing it change the benchmark scores? If you think this is an outlier, you are welcome to go test Hitman, Quantum Break and Total War: Warhammer to see whether Nvidia sucks or not at asynchronous compute. People said Aots was one game and we should wait. We got 3 more and people are still saying wait. Don't worry we have Deus Ex, BF1, Mass Effect Andromeda, Battlezone and Watch Dogs 2 still coming up. How many more exactly do you need again?

[GlassBomb]

6 minutes ago, TidaLWaveZ said:

It's also a "< $500" solution so it could very likely be the 8GB version which for two will probably be around $460 or more.

I'm still generally not a fan of multi-GPU solutions. As I said before, I'm going to wait until average Joe can get their hands on cards.

 

And at this point in time, Dx12 benchmarks mean very little to me, since I'm still running Win7. And don't plan to switch for a good time.