[H]ard|OCP

Navigation

• [H] Forums
• Archives
• Search
• [H] Folding

GeForce3 HRAA High Resolution Anti Aliasing

What can the GeForce3 do for you TODAY?

Certainly one of the things the GeForce3 has brought to the table is High Resolution Anti-Aliasing, or HRAA, as NVIDIA calls it. As we have explained before, we do not intend to do a full review of the Vid Card until we can do one that will show off the new features of the GeForce3. It is not fair to NVIDIA or YOU to show the card NOT doing all of the things it is capable of. To answer the question above, HRAA is one thing it can do for you NOW with the current DX7 and OpenGL games you play today. We have had the VisionTek GeForce3 on the test bench for about five days now. However, we still do not have any official drivers for it. Rather, we are trapped into using NEDs (NVIDIA Experimental Drivers) that we have found leaked onto the Net. Although they are buggy in many games, we did get the GeForce3 working with some applications.

If you are not familiar with what the GeForce3 will have to offer in the coming months, please check out our GeForce3 Technology Overview. It will help get you up to speed with what kind of heat the card is packing.

This article is going to be a simple look at the GeForce3's performance when using HRAA. More specifically, we will be looking into NVIDIA's new Quincunx AA mode, or QCAA for short. What is interesting here is the new way in which NVIDIA does their AA. To explain it very simply, NVIDIA's new method, QCAA, draws information from surrounding pixels to fill in the jaggies. Here is a simple graphic that will give you an idea of how this is done, if you are not very familiar with AA.

Here is a quote from our Technology article that will give you a few more details.

NVIDIA will no longer be "blowing up" your 800x600 screen to 1600x1200 and then shrinking it back down in hardware to do their antialiasing in their new Quincunx mode. From what I am hearing, this is the mode you will most likely be using if you are an AA fan. To give you a basic rundown, it gives you 4X AA quality at the 2X price. Quincunx AA provides as many pixels to smooth out an edge that you would normally see in a 4X supersampling, while only needing the GPU to do the work that is required for 2X supersampling. Now if you compare Quincunx mode with true 4X AA side by side, you will see a difference. The edges of images using Quincunx mode are a bit softer compared with the harder edges of the 4X mode. If you are wondering what a Quincunx is... quin·cunx (kwin'kungks) - It is an arrangement of five objects, with one at each corner of a rectangle or square and one at the center. This is a reference to the way they sample the pixels that are used to fill in the jaggies.

Now what does all of this Quincunx stuff really mean in terms of bandwidth savings and the DDRam used on your GeForce3 card?

Assuming that we have a GeForce3 with even 64Megs of memory on the card, you can still see where this operation requires a tremendous amount of memory. Remember that while all of this is going on, to do HRAA, all of the other operations that the card has to perform are happening also. Considering that our data path between the GPU and DDR are only so wide, you can see why it can become congested very quickly and cause slowdowns. With that said, even the Quincunx still has a pretty big performance hit just as much of the AA we have seen in the past. Just this time we have a lot more power to push the game engine with.

Even a trained monkey can see that while QCAA claims to give us near 4XAA quality, the bandwidth needed is identical to 2XAA. This little chart shows how NVIDIA can claim to get 4XAA at the 2XAA price. This is also a big red flag which shows us that even with NVIDIA's advanced technology, memory bandwidth limitations are still the biggest roadblock - even on premier 3D graphics cards.

While this is all great to sit around and theorize about, what really happens in terms of LOOKS when we use it? If you have not had a good experience with AA in the past, you just might be surprised.