I ask me if this is linked to the Unreal Engine 4 temporal AA in addition
AFAIK, Maxwell is not dual issue.
Yeah, they do that a lot. It makes sense: most people don't upgrade every year. And what a nice upgrade it is.
Riddle me this: in the Anandtech review, for most games, the delta between the GTX980 and the 290X for 4K, supposedly the strong point for the latter, is larger than for smaller resolutions.
I give up...
sheepdogexpress
Newcomer
AMD is screwed from an architectural stand point. 4k isn't a problem anymore and it does extremely well for a card on a 256bit bus.
Those 64Rops being used efficiently is what is doing it. It seems compute performance is increase as well in openCL. Nvidia is going to scoop back that apple mac pro contract if AMD doesn't sweeten the deal. Efficiency and P/w is everything for the 400w powersupply in that unit.
Considering their size and power usage is similar where does this does the big brother of Tonga and Maxwell.
Fiji vs gm200. With the leaks from the fiji cooler, it looks like they want to bring a really really big knife to a gun fight.
They seem to be using the reference cards, maybe at 4k the cards are downclocking more.
Well you could say that, but it is a bit more sophisticated in implementation than you think. From Tech Report:
"DSR brings supersampling back by letting users select higher resolutions, via in-game menus, than their monitors can natively support. For instance, a gamer with a 1080p display could choose the most popular 4K resolution of 3840x2160, which is exactly four times the size of his display. The graphics card will then render the game at a full 3840x2160 internally and scale the output down to 1920x1080 in order to match the display. In doing so, every single pixel on the screen will have been sampled four times, producing a smoother, higher-quality result than what's possible with any form of multisampling."
"DSR goes beyond traditional supersampling, though. Rather than just sample multiple times from within a pixel, it uses a 13-tap gaussian downsizing filter to produce a nice, soft result. The images it produces are likely to be a little softer and more cinematic-feeling. This filter has the advantage of being able to resize from intermediate resolutions. For instance, the user could select 2560x1440, and DSR would downsize to 1080p even though it's not a perfect 2:1 or 4:1 fit."
Again, MFAA is more sophisticated in implementation than you think. From Tech Report:
"MFAA seeks to achieve the quality of 4X multisampling at the performance cost of 2X multisampling. To do so, it combines several elements. The subpixel sample points vary from one pixel to the next in interleaved, screen-door fashion, and they swap every other frame. The algorithm then "borrows" samples from past frames and combines them with current samples to produce higher-quality results—that is, smoother edges."
"More interesting than MFAA itself is the fact that Maxwell has much more flexibility with regard to AA sampling points than Kepler. On Maxwell, each pixel in a 4x4 quad can have its own unique set of subpixel sample points, and the GPU can vary those points from one frame to the next. That means Maxwell could allow for much more sophisticated pseudo-stochastic sampling methods once it's been in the hands of Nvidia's software engineers for more than a few weeks."
Again compared to the rumor mongering trash tabloid sites it is. However....
Touche; I completely forgot about DX12. Other than that spare me from videocardz or wwftech or whatever it's called as anything but a "reliable" source. You could call it plagiarism at best and both are lucky they haven't faced yet legal consequences for their repeated violations.
I've seen in real time for the past years more than one alternative antialiasing methods from both IHVs, which either ended up as not particularly convincing or no one actually cared to use them. Coverage sampling wasn't entirely without potential; in past architectures it was supposed to reach in sample count the alpha test MSAA amount of samples. Ie enabling 32x CSAA should have given also 32x samples for transparencies. Hw was capable of it, never enabled in drivers probably because no one ever cared to benchmark with CSAA.
For the above unless I see it in real time, the sound of the gaussian filter alone gives me already shivers...meaning nothing else but "we'll see".
We do test with the reference card, but this is why we also include the "uber mode" tests, to demonstrate how 290X performs without any throttling whatsoever.
Some reviews indeed mention it's a bit blurry though luckily the filter is easily tunable. Don't know though if it's really any different to a standard bilinear filter if you tune smoothness to 0% thus indeed making this 100% equivalent into a very standard downsampling without a fancy name ;-)
Compute seems „fixed“ with Maxwell (not that it was ever broken in the first place).
http://www.pcgameshardware.de/Gefor...-Geforce-GTX-980-970-GM204-Maxwell-1136157/6/
But recent drivers seem to play into the hands of Maxwell, making it shine a little bit more than if the competition would be utilizing their utmost potential.
http://www.pcgameshardware.de/Gefor...-Geforce-GTX-980-970-GM204-Maxwell-1136157/6/
But recent drivers seem to play into the hands of Maxwell, making it shine a little bit more than if the competition would be utilizing their utmost potential.
Not a surprise considering 48 Kb L1 data cache, a separate 96 Kb shared memory and added hardware shared mem atomics
Wouldn't a LOD slider in the driver control panel just do the trick?
----------------------------------------------------------------------
By the way just read Damien Triolet's review for GM204 SKUs. I wish I could read french, cause the online translation sucks badly. 4 FP64 units after all per SMM according to him and I severely doubt he's wrong about it.
The most impressive about GM204 is that it beats Hawai and Tonga in every single metrics, even the ones where AMD is traditionally strong like DX level (11.3 and 12 for Maxwell) and video (HDMI 2.0 + HVEC enc + hybrid HVEC decode + triple DP).
Finally, when you know that a custom $340 AIB board like the ones from MSI, ASUS and Gigabyte, give us nearly R290X performance at roughly half the power is the last nail in the coffin
For sure, an uber scary moment for AMD...
and BTW, what will be AMD answer ? I don't talk about price reduction, it's a fact in the short time, but in terms of new product and when ?
Finally, when you know that a custom $340 AIB board like the ones from MSI, ASUS and Gigabyte, give us nearly R290X performance at roughly half the power is the last nail in the coffin
For sure, an uber scary moment for AMD...
and BTW, what will be AMD answer ? I don't talk about price reduction, it's a fact in the short time, but in terms of new product and when ?
