AMD's Radeon HD 6970 & Radeon HD 6950: Paving The Future For AMD
by Ryan Smith on December 15, 2010 12:01 AM ESTAnother New Anti-Aliasing Mode: Enhanced Quality AA
With the 6800 series AMD introduced Morphological Anti-Aliasing (MLAA), a low-complexity post-processing anti-aliasing filter. As a post-processing filter it worked with a wide variety of games and APIs, and in most cases the performance overhead was not very severe. However it’s not the only new anti-aliasing mode that AMD has been working on.
New with the 6900 series is a mode AMD is calling Enhanced Quality Anti-Aliasing. If you recall NVIDIA’s Coverage Sample Anti-Aliasing (CSAA) introduced with the GeForce 8800GTX, then all of this should sound quite familiar – in fact it’s basically the same thing.
Under traditional MSAA, for a pixel covered by 2 or more triangles/fragments, 2, 4, or 8 subpixel samples are taken to determine what the final pixel should be. In the process the color of the triangle and the Z/depth of the triangle are both sampled and stored, and at the end of the process the results are blended together to determine the final pixel value. This process works well for resolving aliasing along polygon edges at a fraction of the cost of true super sampling, but it’s still expensive. Collecting and storing the Z and color values requires extra memory to store the values and extra memory bandwidth to work with the values. Ultimately while we need enough samples to determine colors of the involved triangles, we do not always need a great deal of them. With a few color/Z samples we have all of the color data we need in most cases, however the “hard” part of anti-aliasing becomes what the proper blending of color values should be.
1 Pixel Covred by 2 Triangles/Fragments
Thus we have EQAA, a compromise on the idea. Color/Z samples are expensive, but just checking if a triangle covers part of a subpixel is very cheap. If we have enough color/Z samples to get the necessary color information, then just doing additional simple subpixel coverage checks would allow us better determine what percentage of a pixel is covered by a given polygon, which we can then use to blend colors in a more accurate fashion. For example with 4x MSAA we can only determine if a pixel is 0/25/50/75/100 percent covered by a triangle, but with 4x EQAA where we take 4 color samples and then 4 additional coverage-only samples, we can determine blending values down to 0/12/25/37/50/62/75/87/100 percent coverage, the same amount of accuracy as using 8x MSAA. Thus in the right situation we can have quality similar to 8x MSAA for only a little over 4x MSAA’s cost.
MSAA & EQAA Sample Patterns
In reality of course this doesn’t always work out as well. The best case scenario is that the additional coverage samples are almost as good as having additional color/Z samples, while the worst case scenario is that additional coverage samples are practically worthless. This depends on a game-by-game, if not pixel-by-pixel basis. In practice additional coverage samples are a way to slightly improve MSAA quality for a very, very low cost.
While NVIDIA has had the ability to take separate coverage samples since G80, AMD has not had this ability until now. With the 6900 hardware their ROPs finally gain this ability.
Beyond that, AMD and NVIDIA’s implementations are nearly identical except for the naming convention. Both can take a number of coverage samples independent of the color/Z samples based on the setting used; the only notable difference we’re aware of is that like AMD’s other AA modes, their EQAA mode can be programmed to use a custom sample pattern.
As is the case with NVIDIA’s CSAA, AMD’s EQAA mode is available to DirectX applications or can be forced through the drivers. DirectX applications can set it through the Multisample Quality attribute, which is usually abstracted to list the vendor’s name for the mode in a game’s UI. Otherwise it can be forced via the Catalyst Control Center, either by forcing an AA mode, or as is the case with NVIDIA, enhancing the AA mode by letting the game set the AA mode while the driver overrides the game and specifies different Multisample Quality attribute. Thus the “enhance application settings” AA mode is new to AMD with the 6900 series.
To be honest we’re a bit ruffled by the naming choice. True, NVIDIA did go and have to pick daft names for their CSAA modes (when is 8x not 8 sample MSAA?), but ultimately CSAA and EQAA are virtually identical. NVIDIA has a 4 year lead on AMD here, and we’d just as well use NVIDIA’s naming conventions for consistency. Instead we have the following.
| Coverage Sampling Modes: CSAA vs EQAA | ||||
| NVIDIA |
Mode (Color + Coverage) |
AMD | ||
| 2x | 2+0 | 2x | ||
| N/A | 2+2 | 2xEQ | ||
| 4x | 4+0 | 4x | ||
| 8x | 4+4 | 4xEQ | ||
| 16x | 4+12 | N/A | ||
| 8xQ | 8+0 | 8x | ||
| 16xQ | 8+8 | 8xEQ | ||
| 32x | 8+24 | N/A | ||
AMD ends up having 1 mode NVIDIA doesn’t, 2xEQ, which is 2x MSAA + 2x cover samples; meanwhile NVIDIA has 16x (4x MSAA + 12 cover samples) and 32x (8x MSAA + 24 cover samples). Finally, as we’ll see, just as is the case for NVIDIA additional coverage samples are equally cheap for AMD.
Facebook
Twitter
RSS
168 Comments
View All Comments
versesuvius - Friday, December 17, 2010 - link
Ananke,I am not very knowledgeable about this, but I don't think a modern GPU can fit inside a CPU for now. A better idea would be a console on a card. The motherboards on the consoles are not much bigger than the large graphic cards of today. A console card for $100 would be great. I am sure that there is no technical obstacles that the average electronic wizard cannot overcome, doing that.
Sure, there is a use for everything. I can imagine that every single human being on earth can find a use for a Ferrari, but the point is that even those who do have it, do not use it as often as their other car, (Toyota, VW or whatever). In fact, there is rarely a Ferrari that has more than 20,000 km on it, and even that is put on it by successive owners, not one. The average total an ordinary person can stand a Ferrari is 5000 KM. (Disclaimer: I do not have one. I only read something to that effect somewhere). Having said that, I do have a sense of the "need for speed". I can remember sitting in front of the university's 80286 waiting for the FE program to spit out the results, one node at a time, click, click, ... . You have millions of polygons, we can have billions of mesh nodes, and that even does not even begin to model a running faucet. How's that for the need for speed. I do appreciate the current speeds. However, the CPU deal was and is a straight one. The graphic card deals, today, are not. To be clear, the "and" in "High End"s and "Fool"s is an inclusive one. "Someone will pay for it", was also initiated in the eighties of the last century. By the way, the big question "can it play crysis", will no longer be. Crysis 2 is coming to the consoles.
Quidam67 - Friday, December 17, 2010 - link
"But can it play Crysis" should be in the Urban dictionary as a satirical reference on graphics code that combines two potent attributes: 1) is way ahead of its time in terms of what current hardware can support 2) is so badly written and optimised that even hardware that should be able to run it still can't.In 1000 years time when Organic Graphics cards that you can plug into your head still can't run it smoothly @2560*1600 60fps they will realise the joke was on us and that the code itself was written to run more and more needless loops in order to overwhelm any amount of compute-resource thrown at it.
Iketh - Friday, December 24, 2010 - link
LOLmarc1000 - Friday, December 17, 2010 - link
I swear I've read ALL the comments to see if anyone already pointed it... but no one did.I feel a bit disappointed with this launch too (I have a 5770 and wanted to get 6950 but was wanting a bigger increase %-wise). But one thing interesting it the number of Stream Processors in the new gpus. By the "pure processor" count this number decreased from 1600 SPs on 5870 to 1536 SPs on 6970. But the size of the VLIW processors changed too. It was 5 SPs on 5870 and now is 4 SPs.
So we have:
hd5870 = 1600 SPs / 5 = 320 "processors"
hd6970 = 1536 SPs / 4 = 384 "processors"
if we take that 384 and multiply by 5, we would have 1920 SPs on the new generation (on par with many rumors). this is 20% more shaders. and considering AMD is saying that the new VLIW4 is 10% faster than VLIW5 we should have more than 20% increase in all situations. but this is only true in the minority of tests (like crysis at 2560x1660 where it is 24%, but in the same game at 1680x1050 the increase is only 16%). and at the same time the minimun FPS got better, yet in another games the difference is smaller.
but then again, I was expecting a little more. I believe the 6950 will be a worthy upgrade to me, but the expectations were so high that too much people ended a little disappointed... myself included.
Sunburn74 - Tuesday, December 28, 2010 - link
Well... at least they delivered on time and didn't make you wait 6 more months to simply deliver an equivalent, if not considerably worse, product.Mr Perfect - Friday, December 17, 2010 - link
Yes, the minimums are appreciated when they're included.It would be even better if the framerates was displayed as a line graph instead of a bar graph. That way readers could tell if an average consisted of a lot of high peaks and low valleys, or really was a nice smooth experience all the way through. Some other review sites use linegraphs and while I visit Anandtech for it's timeliness, professionalism, industry insight and community involvement, I go to the other sites for the actual performance numbers.
Quidam67 - Friday, December 17, 2010 - link
There is further rationale for splitting the article. Lets say someone is googling "HD 6970 architecture" perhaps they will pick up this review, or perhaps they won't, but either way, if they see that it is actually a review on the cards, they might be inclined to bypass it in favour of a more focused piece.And again, there is no reason why the Architecture Article can't provide a hyperlink to the review, if the reader then decides they want to see how that architecture translates into performance on the current generation of cards supporting it.
I really hope AT are reading this and giving it some consideration. As you say, they are a great sight and no one is disputing that, but it's not a religion, so you should be allowed to question it without being accused of blasphemy :O)
dustcrusher - Friday, December 17, 2010 - link
It really comes down to how important the mainstream market is. If they are a large enough segment of the market, one company using a simple, easy-to-grasp naming convention would likely grab some market share. Make it easy to buy your product and at least some people will be more likely to do so.If not, then it's fun to talk about but not terribly important. Tech-savvy folk will buy whatever meets their needs price/performance-wise after doing research, even if a card is named the Transylvania 6-9000 or the Wankermeister GTFO. Eager to please tech-naive folk are going to buy the largest model number they can get with the money they have, because "larger model numbers = bigger/better equipment" is a long-established consumer shorthand.
I have a half-baked idea for a model numbering system that's based around the key specs of the card- it's a 5 digit system where the first digit is the hardware platform ID (like what we have now, mostly) and the other four would represent combinations of other specs (one digit could be the lowest memory clock speed and bus width would be 1, the next lowest memory clock speed and lowest bus width would be 2, etc).
No idea if this could actually be implemented- there are probably too many variables with GPU/memory clock speeds, among other things.
Shinobi_III - Saturday, December 18, 2010 - link
If you ever saw Nvidia 4xAA in action, you know it's not as smooth as the radeon implementation (especially in motion) and z-buffer miscalculations has always been a nvidia feature.Go up a hill in Fallout New Vegas and look at Vegas in the horizon, with Nvidia cards it always looks like a disco due to meshes overlapping. Now do the same on Radeon.
TheUsual - Saturday, December 18, 2010 - link
Right now, Newegg has a 6870 for $200 after rebate. Two of these makes for an awesome value at $400. The top tier of cards doesn't give a corresponding increase in performance for the extra cost. Two 6950s costs 50% more but does not give you 50% more FPS. Two GTX 460 1GBs is also a great bang for the buck at $300.Neither of these lets you do triple SLI/XFIRE however. That would be what would be paying extra for.
My hope is that the price will drop on the 6950 by around February. By then the GTX 560 should be out and might drive prices down some. The benchmarks could change some with Sandy Bridge too, if they are currently CPU bound.