[AMD_Robert] Concerning the AOTS image quality controversy

[poofyhairguy]

Exactly. See Warhammer Total War, pretty much same results. Evidence starting to strongly point to AMD built a better DX12 architecture.

When AOTS benchmarks leaked and a 290 is near a 980 ti:

"Beta software beta drivers. Wait for release."

When AOTS released and AMD slaughters everything but the 980 ti:

"RTS game almost no one will play. Will wait for a REAL Directx 12 game people will buy like Quantum Break."

When Quantum Break releases:

"The game is super buggy, we can't tell anything from such a broken game."

When Quantum Break is patched and a 390x still slaughters a 980:

"Directx 12 sucks. Oh well I don't care, I am going to replace my 980 with a 1080 anyway."

When the goalposts constantly shift NO game is a good indicator because people don't want to read the meter.

 

[.vodka]

Exactly. See Warhammer Total War, pretty much same results. Evidence starting to strongly point to AMD built a better DX12 architecture.

That, plus having GCN in both consoles to start cementing paradigms on how to get the most out of their product and possibly being the supplier of next gen consoles, if happening, further establishing GCN as the industry standard. Some of the latest console games are quite remarkable considering they're running on what amounts to a 7850 and crappy netbook CPUs. On top of that, we have their GPUopen initiative providing some of the consoles' flexibility on the PC, on open source code.

What else can you ask for if you were a developer and wanted to do some crazy stuff on the PC that is also scalable down to consoles? Want to bring some of that goodness to other platforms? Thanks to Vulkan, you can. No one would have even bothered with a move like Mantle to kickstart the chain of events even with the developers' crying out loud for an API like that.


nV can't touch this. Their proprietary walled garden of technologies especially Gameworks, at least on the consumer space, will probably stop being a problem in little time and will have to start playing the game by other, hopefully more fair-play-like means.

When AOTS benchmarks leaked and a 290 is near a 980 ti:

"Beta software beta drivers. Wait for release."

When AOTS released and AMD slaughters everything but the 980 ti:

"RTS game almost no one will play. Will wait for a REAL Directx 12 game people will buy like Quantum Break."

When Quantum Break releases:

"The game is super buggy, we can't tell anything from such a broken game."

When Quantum Break is patched and a 390x still slaughters a 980:

"Directx 12 sucks. Oh well I don't care, I am going to replace my 980 with a 1080 anyway."

When the goalposts constantly shift NO game is a good indicator because people don't want to read the meter.

The trend is there and the writing was on the wall for nV a long time ago. AMD sure likes to play the long game, even if it means to lose sometimes here and there.

AMD just needs to get a proper marketing team like nV's and get their products moving based on their strengths and creating that apple-like brainwashing halo around their brand that nV has succeeded in making, that's what mainly differentiates both companies lately (That AMD/ATI are alive after Hector Ruiz and the following draining of money and talent is a miracle...). nV managed to sell Fermi, a polished turd vs later Terascale cards in a highly competitive 50/50 market split. AMD couldn't have sold Maxwell as well as nV, that's a given.

To be fair, this has been a clever plotting by them, so far ahead in the future, seeing it come to fruition is exciting. I hope we may come back to a competitive GPU market where the consumer benefits with little gems here and there like the 8800GT, the GTX 460 1GB, the 4850, 5870, as of lately the 7970/290 and probably the RX 480 now, not one where nV scams people with cards that become obsolete in a year or two while asking an arm and a leg for them (I'm looking at you, Kepler. You too, Maxwell, in due time. You too, Founders Edition $100 tax for "craftsmanship".)

 

[IllogicalGlory]

If this result is representative of the RX 480's performance, then it's slower than a stock 290.

62.5/1.83 = 34.15

On 1080p crazy, my 290 gets over 35 FPS, and it's clocked at the standard 947/1250 clocks. If this is true, then the 480 is an absolute failure unless it uses like 80W and has unbelievable OCing headroom.

This is actually not quite true. My initial run was using 4X MSAA, not 8X. Still the difference is only 1 FPS.

XFX 290 Double Dissipation at 947/1250, 1080p Crazy + 8X MSAA (matching the settings on the Polaris demo).

Single GPU RX 480 ~ 34.1 FPS.

 

[swilli89]

When AOTS benchmarks leaked and a 290 is near a 980 ti:

"Beta software beta drivers. Wait for release."

When AOTS released and AMD slaughters everything but the 980 ti:

"RTS game almost no one will play. Will wait for a REAL Directx 12 game people will buy like Quantum Break."

When Quantum Break releases:

"The game is super buggy, we can't tell anything from such a broken game."

When Quantum Break is patched and a 390x still slaughters a 980:

"Directx 12 sucks. Oh well I don't care, I am going to replace my 980 with a 1080 anyway."

When the goalposts constantly shift NO game is a good indicator because people don't want to read the meter.

Yeah people really reaching to continue to discredit AMD's architecture decisions. I believe Hawaii will indeed go down as the #1 best value ever offered to gamers in the history of discrete graphics cards, no other card can ever claim to still be crushing games at high settings near 3 years after its release.

 

[Erenhardt]

7970 had a nice run. But yea, 290 beats it now.

 

[Dresdenboy]

This is actually not quite true. My initial run was using 4X MSAA, not 8X. Still the difference is only 1 FPS.

XFX 290 Double Dissipation at 947/1250, 1080p Crazy + 8X MSAA (matching the settings on the Polaris demo).

Single GPU RX 480 ~ 34.1 FPS.

Confirmed by Robert Hallock:

Abschließend verrät Hallock auch die Skalierung beim Einsatz der zwei Grafikkarten: Sie habe 183 Prozent betragen. Eine Radeon RX480 hat im Test damit 34,2 FPS erreicht – das entspricht 58 Prozent der Leistung der GeForce GTX 1080.

http://www.computerbase.de/2016-06/radeon-rx-480-vorgestellt/
1 RX480: 34.2 fps,
CF scaling factor: 183%

 

[Despoiler]

I don't know. Most people don't hold much weight to AoTS. I wouldn't come to any type of conclusion for either camp on this title.

It's just not a big title, but popularity doesn't mean something is good or bad. As a tech demo for what's possible right now it's amazing. Dan Baker is one of the handful of preeminent engine architects in the entire world. He knows what he is doing. He's said they can't do everything they want to until they move to a native DX12 engine. There is a lot more awesomeness to come.

Exactly. See Warhammer Total War, pretty much same results. Evidence starting to strongly point to AMD built a better DX12 architecture.

It's not even that they built a better one it's that they built one. Nvidia is stuck on an architecture designed for DX11 until at least Volta.

 

[sm625]

That entire reddit thread can be summarized in two words:

IPC Increases

 

[Bacon1]

Confirmed by Robert Hallock:

http://www.computerbase.de/2016-06/radeon-rx-480-vorgestellt/
1 RX480: 34.2 fps,
CF scaling factor: 183%

Source for that is in english here: https://www.reddit.com/r/Amd/comments/4m692q/concerning_the_aots_image_quality_controversy/

 

[Dresdenboy]

IllogicalGlory, AMD used an i7-5930K. Do you have an idea, how your OC'ed CPU compares to this one?

That entire reddit thread can be summarized in two words:

IPC Increases

That is the result of checking AT Forums about Bulldozer before starting the Zen project.

 

[geoxile]

Any word on why the shader count decreased so much for a ~230mm^2 die? Couldn't a full Hawaii/Grenada fit on that?

 

[moonbogg]

AOTS caries no weight with me. Straight up tech demo/sales tool. I really don't care which brand runs it better. I will never play the game.

 

[MrTeal]

Any word on why the shader count decreased so much for a ~230mm^2 die? Couldn't a full Hawaii/Grenada fit on that?

Well, nVidia went from 5.2B transistors on a 398mm² 28nm die (13.1M/mm²) for GM204 to 7.2B on 314 mm² (22.9M/mm²) on TMSC 16nm for GP104. That's 75.5% higher density. Hawaii was 14.16M/mm², Hawaii at 232mm² would be 26.7M/mm² or 88.7% more dense than 28nm. Apple's A9 was 9% denser on 14nm, so getting a full Hawaii worth of transistors into 232mm² might be touch and go. Even if you can, a greater percentage of the transistors might have been needed for non-CU things like the new codec, color compression, etc. Tonga for instance had a very similar transistor/mm² to Hawaii, but a lower SP/mm² ratio (5.7SP's/mm² for Tonga vs 6.43 for Hawaii)

 

[Zstream]

AOTS caries no weight with me. Straight up tech demo/sales tool. I really don't care which brand runs it better. I will never play the game.

It's a fun game

 

[geoxile]

Well, nVidia went from 5.2B transistors on a 398mm² 28nm die (13.1M/mm²) for GM204 to 7.2B on 314 mm² (22.9M/mm²) on TMSC 16nm for GP104. That's 75.5% higher density. Hawaii was 14.16M/mm², Hawaii at 232mm² would be 26.7M/mm² or 88.7% more dense than 28nm. Apple's A9 was 9% denser on 14nm, so getting a full Hawaii worth of transistors into 232mm² might be touch and go. Even if you can, a greater percentage of the transistors might have been needed for non-CU things like the new codec, color compression, etc. Tonga for instance had a very similar transistor/mm² to Hawaii, but a lower SP/mm² ratio (5.7SP's/mm² for Tonga vs 6.43 for Hawaii)

Apple's A9 was produced on TSMC's 16FF+ and Samsung's 14LPE, isn't the 14LPP AMD is using slightly denser than LPE? I guess it's possible that other stuff is taking up more space, but the change in shader count seems pretty drastic.

 

[dark zero]

Mmm. Maybe nerfing the double precission is biting nVIDIA back and hard. Time to verify it on mining.. one of the few real SW that uses the card to the max.

 

[Grooveriding]

Nvidia needs to be pushed to fix this rendering error on the 1080 then. Every review I saw of the 1080 included AOTS and made a big to do about the performance increase in that title over the 980ti/Titan X. No doubt it was part of the reviewer's guide for the card.

There may be a small to negligible difference with this fixed, or there may be a difference worth noting, making reviews of the card using results of this game incorrect.

 

[Armsdealer]

http://www.ashesofthesingularity.co...294-8cab-4399-8815-f956a670b68f/match-history

In what should be less cpu bottlenecked than any 1080p benchmark, you can see extreme 1440p Polaris cf scaling here of 150% ie 40 fps vs 60 fps.

I wouldn't get my panties all tied up thinking I'm only going to get 34 fps at 1080p crazy in their test setup with a single gpu. It will probably be more like 40 as seen here.

 

[Cookie Monster]

AOTS caries no weight with me. Straight up tech demo/sales tool. I really don't care which brand runs it better. I will never play the game.

I have to agree. Looks like one of those RTS games where you play it once or twice or never? Its like one big DX12 tech demo that has a bleak future because successful franchises like Totalwar, SC have their own engines that already do pretty damn well on top of the entire genre becoming a rarity these days. The former has a pretty good DX12 implementation already and SC2 for instance caters toward even the lowest end rigs (the gameplay speaks for itself without the need of all that fancy DX12 effects etc - most actually prefer playing it without all the bells and whistles because it gets in the way - Id know because I use to play starcraft competitively online for quite abit).

So I don't know why people are so obsessed with this glorified tech demo (Total war:Warhammer should be the game that gets more attention imo if people are interested in DX12 performance in a game that actually has awesome gameplay and content).

However, my opinion would change if they made a new red alert game based on this.

 

[itsmydamnation]

What people who haven't played ATOS dont realize is how good the AI is. AOTS isn't a tech demo its a funding model used to reduce the risk of making a complex game engine. in AOTS the enemy is hard(on the hard skill levels) because it is smart unlike most RTS games where the AI is hard because it cheats, this makes it replayablilty despite the lack of unit diversity/specialization etc really good even just in a single player skirmish.

The way it counters and adapts to your strategies is awesome. I dont get why a certain few ( actually i know exactly why) feel the need to attack something they have never played the game yet alone the way it was meant to be played.

 

[JujuFish]

This is actually not quite true. My initial run was using 4X MSAA, not 8X. Still the difference is only 1 FPS.

XFX 290 Double Dissipation at 947/1250, 1080p Crazy + 8X MSAA (matching the settings on the Polaris demo).

Single GPU RX 480 ~ 34.1 FPS.

This is highly irrelevant since the 480 test was done at 1440p VSR.

 

[IllogicalGlory]

This is highly irrelevant since the 480 test was done at 1440p VSR.

I wish. That was the Doom demo.

In Game Settings for both configs: Crazy Settings | 1080P | 8x MSAA | VSYNC OFF

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%.

Obviously, wait for reviews, but this demo should be showing the 480 at its peak. Maybe it's a compute performance bottleneck or something, but...

Still, maybe this thing's crazy efficient, but I have worries about performance compared to die size and potential unused throughput in Vega chips. Or maybe the clocks are really low and increasing them would kill the efficiency.

 

[dark zero]

I feel that drivers for 480 weren't that optimized....

 

[Armsdealer]

I wish. That was the Doom demo.

Obviously, wait for reviews, but this demo should be showing the 480 at its peak. Maybe it's a compute performance bottleneck or something, but...

Still, maybe this thing's crazy efficient, but I have worries about performance compared to die size and potential unused throughput in Vega chips. Or maybe the clocks are really low and increasing them would kill the efficiency.

... There is enormous uncertainty on the cf scaling.

We actually have real benchmarks from the AotS website with the same system with Polaris as single gpu and multi gpu. Why are you guys trying to fiddle with numbers trying to calculate the single gpu performance off of an uncertain cf scaling factor when we actually have real data as opposed to a reddit post?

In 1440p extreme c7 got 40 fps and 60 fps as cf. This is under the account radeondemo that I linked above. That should unwind your panties.

 

[IllogicalGlory]

... There is enormous uncertainty on the cf scaling.

We actually have real benchmarks from the AotS website with the same system with Polaris as single gpu and multi gpu. Why are you guys trying to fiddle with numbers trying to calculate the single gpu performance off of an uncertain cf scaling factor when we actually have real data as opposed to a reddit post?

In 1440p extreme c7 got 40 fps and 60 fps as cf. This is under the account radeondemo that I linked above. That should unwind your panties.

My panties have never been wound. If you check my posts in the Computex thread, I was hyping this card at first. And while it's a reddit post, it's also from someone who works at AMD.

And just so you know, Ashes doesn't respond to resolution changes the way most games do.

R9 290 Double Dissipation at 947/1250, 1440p Extreme:

The performance delta is practically the same. Specifically, the RX 480 is 1.8% faster compared to 0.4% in the 1080p test.