AMD Nano

[SteelCrysis]

http://techreport.com/review/28912/tiny-...lop-at-650

The reason for the difference in FPS averages was clear when we looked back at the footnotes of AMD's document supplying the numbers. The firm tested the cards with a very particular formula in order to achieve its results for the Fury X—and it has carried over that formula when generating the R9 Nano results you see above. The basic approach is to test exclusively in 4K with high-quality shader effects and post-process anti-aliasing, but to soft-pedal on texture filtering and mulitsampled edge anti-aliasing.

Here's how AMD says it tested the games in the Nano benchmark results above, based on the footnotes of its presentation:

That's a pretty weird combination of settings, all things considered. 4K is one of the highest resolutions you can get in a display today, and yet we've compromised dramatically on image quality via low-rent texture filtering and edge AA methods.

Here's a quick illustration of the difference in texture quality between 0X and 16X anisotropic filtering:

Without anisotropic filtering


With 16X anisotropic filtering

The difference in quality between post-process AA methods like SMAA/FXAA and more traditional methods like MSAA isn't quite so dramatic, but post-process AA methods aren't aware of sub-pixel geometry, so they sometimes get things wrong. For example:

FXAA (left) cannot resolve the fine geometry that a sub-pixel-aware technique like MSAA or SSAA can (right)

I can't imagine building a high-end gaming rig, getting a 4K monitor and a fancy video card for it, and then tweaking games to compromise image quality in such basic ways.

So why is AMD testing in this fashion? Probably because it plays to the Fiji GPU's strengths—that enormous shader array—while not leaning so hard on its potential areas of relative weakness, like ROP throughput (for MSAA resolve), texturing (for anisotropic filtering), and small triangles (since polys are relatively smaller at lower resolutions). Taking this peculiar path likely puts the R9 Nano in the best possible competitive light. In the absence of independent reviews, those AMD-provided numbers are sure to circulate widely online. Trouble is, AMD could be setting up the R9 Nano to be a disappointment once it gets into the hands of reviewers and consumers.

So there you have AMD's performance numbers with some proper context. My advice: wait for the reviews before you decide to jump on an R9 Nano. It may well be a very good product. Heck, it looks super-cool. But manufacturer-provided performance numbers often tell a different story than one might hope.

http://www.extremetech.com/gaming/212973...x-inch-gpu

AMD is projecting that the Radeon R9 Nano will compare extremely well against the GTX 970 Mini, as shown below:

After some of the issues surrounding the Fury X’s early benchmark leaks, many people will give these results a gimlet eye. According to the footnotes published in AMD’s presentation, Far Cry 4 was tested at 4K at the Ultra High preset with 0x MSAA and 0x AF. AMD used its own Catalyst 15.201 and compared against Nvidia’s 355.60 WHQL driver. All of the additional games were tested at High detail and 4K, with no MSAA or AF forced on or manually selected. FXAA was used in several titles, but it has a minimal impact on GPU performance.