Vulkan and DX12 on new GPU'S

[Noughtboy]

Since when was 160w a lot of power. The RX480 uses more power than the gtx1060 yes but it does not use a lot of power when looking at cards like the gtx480 and R9 290x. I agree though stay away from the stock cooler and get a Devil or another decent cooler.

I will refer you to my post above (#9) asking about the "power gap". One of my issues is my computer is on for 10 hours everyday even without gaming and the RX480 seems to chew through triple the power for multi monitor setups which negates any short term price difference in purchase price to the tune of around £12 p.a.

Now it isn't the quantum per se it is the fact of inefficiency. I could be less annoyed if it because of additional hardware overhead giving longevity as asked about in the post #9 above. £12 per annum is cheaper than a new GPU after 2 or three years.

 

[fuzzylogic]

Don't buy a reference card. Especially since your interested in keeping your card for a long time. Do yourself a serious favour and get a decent aftermarket card.

AMD produce terrible coolers. Their cards use a lot of power and run hot. They desperately need good cooling solutions to do them justice both in terms of performance and overall user experience.

I know. I'm just worried about impending price rises. The Red Devil and Gigabyte cards are the absolute limit of what I wanted to pay. I really wanted the standard 8GB Nitro card but they're out of stock everywhere.

 

[Scougar]

Only viable AMD if developers use proper Async, or Nvidia actually start using it.

Right now... it's in Nvidia's favor to not implement such tech because AMD put dedicated hardware in there that takes up die space meaning they can't compete when it's not in use. I honestly hope they don't do this, but hey... such is life.

 

[Doom112]

Only viable AMD if developers use proper Async, or Nvidia actually start using it.

Right now... it's in Nvidia's favor to not implement such tech because AMD put dedicated hardware in there that takes up die space meaning they can't compete when it's not in use. I honestly hope they don't do this, but hey... such is life.

Their are thing in which, Nvidia has upper hand for example tiled resources, efficiency, better clock speed.

 

[D.P.]

Nvidia's current cards do not have new api features such as async compute being done at a hardware level, Future Nvidia cards likely will meaning the push for most games to use these features in DX12 & Vulcan will increase, so in 2 or 3 years time you may find Nvidia's current cards such as the 1060 and the older maxwell cards not doing as well compared to AMD's current cards such as the 480 which do have hardware support for those new api features..

Wrong, DX12 multi-engine queues are computed in hardware in both Maxwell and Pascal.

 

[D.P.]

Legacy product for Nvidia means they aren't going to receive any optimisation with new drivers.

Is different than Legacy drivers to support very old hardware.

So much BSon this forum.

Legacy Simply means The GPUs are no long manufactured, it has nothing to do with driver support in the slightest.

 

[SiDeards73]

I will refer you to my post above (#9) asking about the "power gap". One of my issues is my computer is on for 10 hours everyday even without gaming and the RX480 seems to chew through triple the power for multi monitor setups which negates any short term price difference in purchase price to the tune of around £12 p.a.

Now it isn't the quantum per se it is the fact of inefficiency. I could be less annoyed if it because of additional hardware overhead giving longevity as asked about in the post #9 above. £12 per annum is cheaper than a new GPU after 2 or three years.

It's exactly this, GCN architecture has a lot of hardware onboard that simply goes unused in DX11 as it's not setup to utilise the command system of DX11 efficiently, AMD played the long game a d hedged on people utilising their hardware sooner, it never really happened for DX11. The hardware still requires power for the gpu to work and that power is effectively wasted.

Nvidia has not bothered with a lot of these hardware features and setup their hardware for the here and now, this has let them be a lot more streamlined and efficient for DX11 and require Less power as they dont have the hardware onboard wasting it.

However this does mean their dx12 performance and Vulcan is not as great as they have to tackle it with software only where as AMD hardware gains an upper hand, as their hardware finally gets to stretch its legs and the results speak for themselves mostly.

Nvidia ultimately will have to design these features into their hardware or lose ground on AMD cards, which means they will become less efficient as they add extra hardware that requires power.

I will say look at Polaris vs Pascal in DX12 and Vulcan, if we compare the 480 to the 1060 you will see the 1060 wins most DX11 and uses less power, the 480 reverses this trend mostly in DX12 and especially in Doom Vulcan, although the power remains higher on AMD. You are probably correct you say ultimately it will even the playing field once Nvidia build for DX12 their efficiency will take a hit but performance will improve.

As far as I can tell Polaris is the first true DX12 architecture, Pascal is still designed for DX11, Volta will probably be Nvidias first DX12 focused Hardware, but I expect Vega to be out before then and that will brute force close the DX11 gap and widen the DX12 gap

 

[tommybhoy]

Lost count of the devs that's reported Async being disabled on Maxwell./

@Op, toss up between efficiency or long term driver support-you make the choice as you can't really count on non bias info in here.

 

[SiDeards73]

Lost count of the devs that's reported Async being disabled on Maxwell./

@Op, toss up between efficiency or long term driver support-you make the choice as you can't really count on non bias info in here.

Yes this is also true, there are a lot of pro Nvidia fans on these boards and they will shout down anything they don't like or paints Nvidia in a bad light. Many devs have said that Maxwell does not have hardware Async and its support is switched off in drivers, there's a good possibility that Pascal does not either, but of course these Nvidia fans would have you believe otherwise.

At the end of the day if many devs say they can't support a feature on xxx hardware it's because xxx hardware cannot support it.

 

[V F]

As far as I can tell Polaris is the first true DX12 architecture, Pascal is still designed for DX11, Volta will probably be Nvidias first DX12 focused Hardware, but I expect Vega to be out before then and that will brute force close the DX11 gap and widen the DX12 gap

Hmm, I don't know now. :s I've been edging to a GTX 1080 but if it isn't a good DX12 card and wait for a 1080Ti... I don't want to get a card then 3 - 6 months down the line saying, oh crap wish I had waited. Or even March.

 

[Bonjour]

Where iis the link which states that? Who said that Kepler and Maxwell are getting legacy driver?

You know better then Nvidia and their employees?
ManuelG
NVIDIA Rep



http://forums.guru3d.com/showthread.php?t=409017

That tells me that they aren't actively optimising the drivers for their cards that are EoL, but that some of the optimisations they make are architecture agnostic so should still benefit the older cards.

 

[SiDeards73]

1080 is fine for dx12 it's brute force power will keep it ahead of any current AMD cards, playing the waiting game is good IF you upgrade every 3 or more years, if you upgrade yearly or even every 2 years then buy a 1080 now, it will batter anything you play.

AMD do not currently have a high end card and Vega won't be out til the end of the year at the earliest.

Also there is always going to be a better card around the corner

 

[SiDeards73]

Personally I wouldn't buy a current Nvidia Pascal card if I only upgraded every 3 or more years. I would wait to see Vega and then decide, as I say 1080 is fine for DX12 right now and prob the next year or two, but it's going to get outclassed by the next Nvidia and AMD cards which is the natural evolution of things.

I see Pascal as the last of Nvidias DX11 architecture

 

[V F]

When is the next NVIDIA architecture due? I'd rather be happy with the next 2 - 3 - 4 years rather than 6 - 9 months.

 

[TheRealDeal]

When is the next NVIDIA architecture due? I'd rather be happy with the next 2 - 3 - 4 years rather than 6 - 9 months.

Most likely 2018 for Volta but it could come sooner. I would say late 2017 at the earliest but i am only guessing. I think the Nvidia road map suggested 2018 but these are also open to change.

If i was you i would at least wait for a 1080ti to hit the market as that should last longer than the gtx1080. Vega is also interesting so if you are not in a rush decide between Vega and 1080ti if you want something to last.

 

[V F]

Roger that. Suppose I could hold out till the New Year.

 

[N19h7m4r3]

Roger that. Suppose I could hold out till the New Year.

That's when AMD's Vega is hopefully going to rear its head. Fingers crossed for another 290X vs Titan scenario; excluding the extremely hot reference AMD card of course

 

[nashathedog]

I keep swerving towards a 1060 but after everything I've read I'm sure the 480 would be the correct decision for longevity and future-proofing. I just can't find one of the blasted things that's available for a reasonable price. I'd buy a reference card now, but they no one seems to rate them next to the customs.

I'd usually say go Sapphire but the XFX coolers look nice this gen and I'm sure they'll do a decent job on the 480, The PowerDevil looks beefy too.

 

[nashathedog]

Wrong, DX12 multi-engine queues are computed in hardware in both Maxwell and Pascal.

I've seen this being claimed by certain people on this forum before, I specifically mentioned async compute so if it's wrong as you're saying can I have a source please I'd like to read up on it.

 

[titaniumx3]

With regards to async compute, I reckon the whole situation with that will unfold a lot like tesselation did back in the HD5870 days. I can recall ATI/AMD making a lot of noise about them being the first to the market with tesselation and that they had been doing it for a long time, ATI trueform or something like that? In any case, once nvidia caught up they did it way better and did at a time when games were starting to actually made use of tessellation.

Personally I buy GPUs that provide the best performance for the here and now, not 3 years down the line, when it's already been eclipsed by better hardware anyway.