Concerned by Nvidia's current state of DX12 and Vulkan performance

[Toss3]

I am not. I am comparing GTX 980 Ti with Fury X and both are on 28nm and RX 480 with GTX 1060.

The OP's point was that Nvidia is showing negative scaling when using DX12, and that could be a problem going forward. Not that it is going to matter if nvidia is still faster despite the negative scaling, but it would suck to be stuck on dx11 for the foreseeable future.

 

[Doom112]

The OP's point was that Nvidia is showing negative scaling when using DX12, and that could be a problem going forward. Not that it is going to matter if nvidia is still faster despite the negative scaling, but it would suck to be stuck on dx11 for the foreseeable future.

Even AMD is showing negative scaling in some DX12 games. The point is that their is no clear future and even their is no clear future if developer will use DX12 or ditch it like they did with DX10.

 

[chroniclard]

If I am getting 60+ FPS in my game of choice does it matter that its not especially well optimised or handles DX12 well? Nope. Brute force or future looking/optimised card, doesnt matter.

This is a non existant problem.

 

[Toss3]

Even AMD is showing negative scaling in some DX12 games. The point is that their is no clear future and even their is no clear future if developer will use DX12 or ditch it like they did with DX10.

If you look at the minimum fps though the RX 480 usually sees a pretty good increase. Still that could be eliminated with a decent enough CPU. Really don't like what we've been seeing from DX12 thus far; developers don't seem to want to put in the time to make it run better than DX11, so maybe the whole close to the metal thing isn't going to work out? Not sure why the ports are running so crappy - wasn't the whole argument for DX12 that it would make porting easier? Same hardware in the Xbox One, so why is it running like ****?

 

[tyler_jrb]

No not really. I don't think nvidia are really pushing for it yet as the need for DX12 perf is low. Less than a handful of good DX12 titles and as always there are some that favour AMD and some nvidia.

Have to wait and see with Volta.

 

[Panos]

NV DX12 support is not as good as the DX11.
And that means any Pascal card has same issues, even the GTX1080Ti if ever comes out.

Here is my results on the TW Warhammer.

1920x1080
Score 77.2, GPU 1080 @2190/2762, CPU 6700k @4.8, Panos DX11 Link 372.90 Drivers
Score 66.2, GPU 1080 @2190/2762, CPU 6700k @4.8, Panos DX12 Link 372.90 Drivers

2560x1440
Score 50.5, GPU 1080 @2190/2762, CPU 6700k @4.8, Panos DX11 Link 372.90 Drivers
Score 41.8, GPU 1080 @2190/2762, CPU 6700k @4.8, Panos DX12 Link 372.90 Drivers

And that is representative of all DX12 games with my 1080.
DX11 is better......

Or Kaap's results at 2560x1440
Score 66.8, GPU TitanP @2126/2728, CPU 6950X @4.4, Kaapstad DX11 Link 372.70 Drivers
Score 56.4, GPU TitanP @2126/2728, CPU 6950X @4.4, Kaapstad DX12 Link 372.70 Drivers


While have a look at AMD cards, barely any loss of fps, for sure not 20% the NV cards have.

Score 39.2, GPU Fury X @1190/500, CPU 5960X @4.4, Kaapstad DX11 Link 16.9.2 Drivers
Score 38.8, GPU Fury X @1190/500, CPU 5960X @4.4, Kaapstad DX12 Link 16.9.2 Drivers
Score 37.7, GPU Nano @1100/550, CPU 4690k @4.7, Radox-0 DX11 Link 16.9.2 Drivers
Score 36.4, GPU Nano @1100/550, CPU 4690k @4.7, Radox-0 DX12 Link 16.9.2 Drivers
Score 29.0, GPU RX480 @1380/2200, CPU 6950X @4.0, Kaapstad DX11 Link 16.9.2 Drivers
Score 28.0, GPU RX480 @1380/2200, CPU 6950X @4.0, Kaapstad DX12 Link 16.9.2 Drivers

(Please bear in mind that the settings for all above is AA 8x and ultra)


And we ain't going to see Volta until 2018 now, as NV announced. Just a rebaged 1080/1070 for 2017, leaving AMD open field for Vega.

 

[Caracus2k]

Gotta love the people saying 'wait for vega'.... It's increasing looking likely that we won't see a top end vega part from amd until well into 2017 by which time NVidia will either already have released or be soon to release their replacements' to the 1070 and 1080. So performance 'somewhere' between the 1080 and 1070 (assuming the notoriously overhyping and under delivering amd marketing machine doesn't let us down again) doesn't really seem all that exciting for those of us on the forum who tend to stay within the top 2/3 performing cards on the market most of the time.

Amd still provides better longevity with the likes of 7970 vs 680gtx but its a small silver lining to an otherwise dark gloomy cloud

 

[VincentHanna]

Isn't the problem more the fact that Nvidia DX11 drivers were so good compared to AMD, that DX12 is putting them on a much more even keel (since driver optimisation under DX12 isn't anywhere near as important)

Obviously lack of proper ASYNC compute on current Nvidia hardware is also an issue, as now even the less powerful cards from AMD all handle Async compute correctly.

There's been very few DX12 only titles built from the ground up yet though (Forza Horizon 3 is the only one that comes to mind) and though that game does need some patches, performance on Nvidia hardware is pretty good compared with AMD.

I'm fairly confident that Pascal and Maxwell just arent fully fledged DX12 cards and we'll have to wait for Volta to increase the gap once more.

 

[TheMorningStar]

I must admit I find it odd that AMD didn't offer a 490 based on a 480 with a doubling of specs, that would have been a beast of a card and you would think it would need little development time. I'm guessing they are just putting it all into Vege, Zen shouldn't be holding up gfx development as they split the departments unless resources are still fluid in the workforce.

 

[SiDeards73]

AMD already stated they have 2 teams working on CPU tech alone, while the current Zen team are finishing and polishing their 2nd team is already working on the next CPU architecture.

I think they probably do the same on GPU but the team is probably smaller? Would explain the long wait time between Polaris and Vega ?

 

[DarrenM343]

Tend to agree about the "here and now" comment. I reckon NVidia will be on top of any issues when it really matters, when "proper" DX12 games hit the market. It might mean the best DX12 performance from Nvidia will come with the next generation of cards however, when DX12 games built from the ground up are the norm.
NV's hardware approach is the same. They don't usually provide gimmicky/marketing features but provide the most optimal solution for it's intended use, so for example, no HBM until it really makes a good difference and the tech is advanced enough, and no big bus on cards likely to be used at lower resolution, unless it really makes a difference etc, and enough memory too for the cards intended use.

 

[Rroff]

I must admit I find it odd that AMD didn't offer a 490 based on a 480 with a doubling of specs, that would have been a beast of a card and you would think it would need little development time. I'm guessing they are just putting it all into Vege, Zen shouldn't be holding up gfx development as they split the departments unless resources are still fluid in the workforce.

A really long shot - but supposedly there was some problem with TSMC 16nm FF above a certain core size at one point where power/thermal gains just went out the window and the 480 is conveniently right below what that would be on 14nm FF and also not showing great thermal/power properties for the node size - so IMO despite some claims to the opposite at one point by certain people Samsung/GF 14nm was/is actually 6 months behind TSMC and they hadn't actually fully sorted that hurdle yet - which means at that point they simply couldn't do it - but I'd assume they probably can ramp upto something bigger about now.

 

[llamadal]

Why be concerned over something that's hardly utilised at this point? I don't give a damn about something that might work better in xxx timeframe when I want to be playing now.

Only a small handful use dx12 and even fewer do it well.

I think Darren sums it up pretty well.

 

[NeoWave]

I must admit I find it odd that AMD didn't offer a 490 based on a 480

Possibly because of the power draw? I get the feeling the RX480 is a stop gap to fill a hole that appeared in their road map to Vega - they needed something to sell so did the best they could with what that had. That said, the RX480 is a decent card for the price imho. I just wouldn't expect much more than default performance from it.

As for Nvidia and DX12/Vulcan, I'm sort of (lol) sure Tom Petersen admitted they were behind with it compared to AMD but that overall it was masked by Team Green having higher performance cards. However, they did expect to have the issues solved so I'd probably be tempted to wait for the Pascal 2.0 refresh and Vega release if possible - if nothing else, it may save a few quid of card prices.

 

[opethdisciple]

Tend to agree about the "here and now" comment. I reckon NVidia will be on top of any issues when it really matters, when "proper" DX12 games hit the market. It might mean the best DX12 performance from Nvidia will come with the next generation of cards however, when DX12 games built from the ground up are the norm.
NV's hardware approach is the same. They don't usually provide gimmicky/marketing features but provide the most optimal solution for it's intended use, so for example, no HBM until it really makes a good difference and the tech is advanced enough, and no big bus on cards likely to be used at lower resolution, unless it really makes a difference etc, and enough memory too for the cards intended use.

Ah yes. I forgot about HBM and the larger buses. Again AMD first to market. Ahead of trends. (good or bad).

I'm not bashing Nvidia with this thread. It's just from my personal perspective DX12 and Vulkan have been a no show so far with my 970. And it's not gone unnoticed how good AMD cards seem to do in titles using next gen API's.

Is BF1 DX12 from day one? Thats going to be an interesting metric. DICE seem to be good with implementing next gen API's in to their games, so if anyone can get it right it might be DICE.

 

[DarrenM343]

Ah yes. I forgot about HBM and the larger buses. Again AMD first to market. Ahead of trends. (good or bad).

I'm not bashing Nvidia with this thread. It's just from my personal perspective DX12 and Vulkan have been a no show so far with my 970. And it's not gone unnoticed how good AMD cards seem to do in titles using next gen API's.

Is BF1 DX12 from day one? Thats going to be an interesting metric. DICE seem to be good with implementing next gen API's in to their games, so if anyone can get it right it might be DICE.

Thing you have to remember is that given what Nvidia can do, they've 'probably' chosen this stuff is just lower priority right now. I doubt it's a case of they just can't master it.

I don't think it's a case of AMD behing ahead of trends personally, I've seen it as "good marketing" by AMD in the past, ie, add stuff you don't really need but it helps sell their products.FuryX probably sold better because of the HBM for example than if it didn't have it. People love new tech even if at the time it's a bit premature to be that worthwhile having. It's not as if in some cases it was new tech either, like the bus size for example. Nvidia sometimes take a step back in bus size, because their new cards simply don't need it so why provide it? Gtx 480 was 384bit for example if I remember correctly, 680 256bit. People moaned on this forums at the time but 680 was of course better at everything than the 480 and simply didn't need it. Think the 7970 of the same era (as 680) had 512bit memory bus and might have needed it because the architecture was of course different from NVidia's, it didn't mean it was better though because of it.

 

[TheMorningStar]

Nah, 79xxx series had 3gb so they were 384bit or whatever, I believe they faced a 256bit card and the 79 won out with mature drivers.

 

[Disco_P]

It does concern me. I buy high end cards and keep them for as long as possible. GTX titan lasted 3.5 years before I wanted to upgrade. I hope I can get 3 years out of my 1080 but if it seriously under performs in future DX12 only games I might have to upgrade sooner and it'll lose all resale value.

 

[Nexus18]

Been saying this for ages now, nvidia concentrate on the here and now where as AMD look to the future, imo this can only be a good thing for AMD in the long run. Already seeing people getting disgruntled at nvidia over not supporting adaptive sync and the lack of gains from dx 12 + vulkan.

Happens with everything in business:

https://hbr.org/2014/07/ceos-get-to-know-your-rivals/

Look for weaknesses that present opportunities. “It’s important to understand your rival’s weak spots and strategy,” says Will Ethridge, Pearson Education’s former CEO. “I knew one rival whose CEO was focused on driving profitability and bringing up margins. I knew that meant he would be relatively focused on short-term results while mostly ignoring product development. I also realized he would go after our top sales people to meet those short-term sales goals. So we worked extra hard to protect our key sales people. At the same time, we continued to invest in long-term product development and overseas markets, knowing it was unlikely he could follow us in the short term.”

Another area where this is applicable other than just APIs is the free/adaptive vs g sync situation. Just have to look at how much adaptive/free sync has taken of despite being over a year late compared to gsync, the magnitude of choice out there for adaptive/free sync is overwhelming now and the gap is only going to get bigger and bigger.

 

[Rroff]

^^ In this case though nVidia doesn't ignore product development. AMD really need to capitalise on those long term strategies and they simply don't - they get a small bonus in the tail end of a product cycle and already behind come the next and they simply aren't catching up - a long term strategy has to pay off at some point.

There are some cases where AMD being ahead of the game has pushed tech development but they've rarely been able to significantly gain from it themselves infact more of times than the other way around AMD jumping early has ended up playing into nVidia's hands i.e. look at tessellation.