TABLE OF CONTENTS
June 6-10, 2006: Computex'2006: ATI
We have already reviewed eminent hardware manufacturers, but what about chipmakers? That is those companies, which do not manufacture end-user products, but only cores and processors. Including GPUs.
Of course, there is nothing special, nothing revolutionary new. And there shouldn't be. There are no announcements, and there won't be any. But ATI actually prepared interesting material, food for thought.
Let's start with the stand.
In fact, there were two locations. The first one was the stand in Hall One, the second one was a large hall in Hayatt hotel. The stand was small and half empty. Most events happened in the hotel. ATI had no official new products, so there was nothing much to show on the stand. But micro-seminars in the hotel offered much interesting information. Moreover, the hall in the hotel was decorated even better than the stand.
That is the exposition was available in two places: at the stand and in the above mentioned hotel hall. Strange as it may seem, 80% of the exposition was concentrated in the hotel, where people could enter only with business cards to visit seminars. The stand was nearly empty, though it would have been more logical to demonstrate everything to the public at large.
So, the exposition consisted of various video cards, manufactured by Canadian partners. However, I repeat that ATI and NVIDIA share the same policy concerning Hi-End: to manufacture cards on their own (placing orders to third-party plants, as these companies do not have their own serial production lines) and then sell them to their partners. That's why you cannot really say that Sapphire or HIS or ASUS are manufacturers of X1900 XTX or 7900GTX cards. They just bought these cards, applied their labels and packed them in their boxes. And then these cards were sold through their channels.
There were plenty of professional FireGL accelerators. Unlike NVIDIA, ATI manufactures these cards on its own. That's probably why FireGL cards are not popular (according to various sources, 90% of the market of such video cards is taken by NVIDIA). These cards are mostly created for demonstrations at expositions, but they are not actually popular. However, what else can you expect, when CATALYST drivers still cannot detect 3D modes, if an application runs in a window, not full screen (yours truly had been screaming about it to the software authors for three months, but in vain). I just want to note here that all professional 3D packages work in a window mode. Of course, software for professional cards is written by other people. But I suspect that these products are not so popular because of these very software problems. Or due to the eternal marketing problem of the Canadian company :(. That's my personal opinion, mind it. Nothing more. I just see no other reasons why FireGL cards are not popular.
These are weak points. But there are a lot of strong points as well. Mobile solutions in the first place. ATI is the strongest here. There were a lot of such examples at the exposition. And of course we get back to 3D graphics for desktop solutions. The company demonstrated new features of CrossFire, drivers, etc. The X1900 XTX card is still the most expedient Hi-End accelerator among single-chip video cards. With its lower cost, it possesses anisotropy of higher quality and allows to use HDR together with AA.
But the most interesting information was provided at seminars. Everybody knows that some manufacturers are concerned with computing physics in games and some other applications. Here is the point: modern CPUs are so heavily loaded with computations that it's impossible for them to compute sterling physical processes. That's why we sometimes see primitive object interactions in games.
Here is the main point: capacities of modern accelerators for processing 3D graphics are getting excessive, especially in case of such tandems as CrossFire. Why not use one of the accelerators to compute physics? In fact, this idea belongs to NVIDIA. It was the first to announce that one of SLI accelerators could be used to compute physics.
But ATI develops this idea. Why only CrossFire? It can be done with any configuration (even asymmetric), even if you install the third video card (in case of available slots) for physics.
Come to think of it, a 3D accelerator is a perfect candidate for physics computations. This, and flexible shaders, and multiple pipelines, etc. Some effects can actually be emulated by shaders:
Multiple pipelines, now called "Data parallel processor", can be used well to compute physical parameters. GPU is actually just a calculator (though very fast and powerful), which receives some values, processes them by a given algorithm, and outputs other values. In case of 3D graphics, they may be pixel colors, etc, in physics - flight coordinates of an object, relative weights, speeds, etc.
So, ATI suggests using one accelerator of your tandem or even trinity (as we can see on the above photos of the testbed on the seminar) to compute physical phenomena. This is done in cooperation with Havok, where entire physics will be computed by the software Havok FX engine. That is there actually appeared a new API. We know about Ageia with its PhysX. But the Canadians preferred their own innovations (using an additional video card instead of a stand-alone physics accelerator) together with Havok FX. The point is that configurations can be asymmetric and any accelerator (up to X1600) can compute physics.
Havok FX engine currently computes collisions and movements of small particles only.
You can see on the diagram that many 3D objects have complex surface relief, obtained with shaders, etc. Formally, these are rectangles or cubes, etc. Of course, you cannot make these cubes touch with their visible borders instead of their invisible ones. So you have to cover the surface of a real visible object with additional coordinates and use them to compute collisions. It requires additional resources, so you need an additional accelerator.
What can you do? Compute coordinates each time an object changes its position? - Even five powerful accelerators may be insufficient for this task. The company offers to split a scene into parts and then select parts with objects that do not participate in a collision (they don't have to be recomputed) in order to reduce computations. It resembles the tile architecture and hierarchical Z-buffer, doesn't it?
Let's sum it all up. ATI states that even the X1600 XT card can compute physics twice as fast as Ageia PhysX or a fast modern CPU, such as Intel Core Duo 2.66 GHz. The X1900 XTX card can do it 9 times as fast. The branching unit of NVIDIA GeForce 7xxx is too weak for computing physics. That is the company declares that hardware from ATI is the best solution here.
Of course, we should check it up, but we don't have the hardware yet. All those demos we were shown at the seminar were just dreams, plans, nothing more. There are no games to see it in action. The company promises such games by the end of the year, as this method and Havok does not require overhauling game engines, because computing physics is a bonus from ATI, an addition to the existing order in a game. That is, everything will work as usual without Havok FX and an additional ATI accelerator. But if you add them, you will have ADDITIONAL effects. It all looks great, even admirable. But we'll see how it will be implemented. We all remember too well TruForm and its enthusiastic demos, but it died unborn.
On the whole, the seminar was very useful, it was very interesting to hear about company's innovations. If they do everything they planned, CrossFire will get the second wind. I'd like to thank Nikolay Radovsky for this seminar and for the warm and cordial welcome at ATI's stand.
As usual, we conclude this part with some photos.
Read the next part, devoted to NVIDIA and a number of other companies.
Write a comment below. No registration needed!
Copyright © Byrds Research & Publishing, Ltd., 1997–2011. All rights reserved.