[H]ard|OCP

Navigation

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

ATI Radeon X1900 Series Evaluation

ATI’s new series of high-end desktop gaming video cards are poised to provide a more immersive gaming experience. We have many comparisons to show you and some surprising results. If you are a gamer, don’t miss this evaluation.

Introduction:

As odd as it seems, here I am writing about ATI’s latest generation of products only three months after ATI’s last generation of products. Technically, ATI’s latest family of video cards is merely a refresh of their previous series, but this is the quickest refresh I’ve ever seen, rivaling the GeForce 5900 FX refresh of the GeForce 5800 FX series. It wasn’t poor performance that made ATI kick out this refresh so fast; instead it was the delay in the Radeon X1800 series product launch that caused this. Originally, ATI was aiming to launch the Radeon X1800 series in the summer of 2005, but because of some pesky transistors in the R520 GPU, that product was severely delayed. Instead of pushing back their refresh for the R520, ATI decided to keep on track with what they had originally planned for this new refresh and go ahead and get it out the door now. In fact, they are getting it out of the door so fast you can already buy them!

ATI really wanted to do a hard launch with this new series, so they geared up to prepare for immediate availability on launch day. This was great news, but what was even better news was that people were buying these video cards before the official launch day! Yes, there were cards in stock at online stores and people were buying them the weekend before the launch! This is absolutely amazing; we have never seen a launch so hard that the product is actually available BEFORE they announce it! Kudos to ATI for keeping their word and making this the hardest launch we have ever seen. Hopefully, availability will keep steady and we won’t see the situation we saw with the GeForce 7800 GTX 512 where cards disappeared out of stock within the first few weeks.

As we mentioned earlier, today’s announcement is a refresh of the Radeon X1000 series technology. The new GPU is known as the R580, and it takes everything the R520 was and improves on it. You can read all about X1000 technology here; we are not going to go into detail about it since everything in there applies to the R580, with the exception of some simple additions and clock speed changes.

Radeon X1900 XT/XTX and XT CrossFire:

As you can see from the title, ATI is announcing three new video cards today, the Radeon X1900 XT, X1900 XTX, and Radeon X1900 XT CrossFire Edition. First, let’s talk about what is new because there are not a lot of new features being introduced with the X1900 series. Take everything you know about the X1800 architecture and add to it what we are fixing to discuss. From the outside, you wouldn’t be able to tell an X1800 XT apart from an X1900 XT, it is on the inside that counts.

These new video cards use the same die process used on the Radeon X1800 XT, which is 90nm and made at TSMC. However, there are more transistors in the new R580 GPU—over 380 million—because there are some improvements in the architecture.

Not Pipelines, Shader Processors!

We’ve been saying this over and over again during 2005 and now that it is 2006 we mean it even more: don’t get caught up in the number of “pipelines” in a video card. The architecture of video cards is moving away from a distinct 1:1 ratio of the various pipelines; instead, we are moving into the world of processing units. Back in the old school days, you could say a card had a certain amount of pipelines and that would help you figure out how it might perform. Well, things are getting so diverse in the architecture now that all the different units in a 3D architecture no longer have a 1:1 ratio. Different parts of the engine now have different numbers of each part and they mean different things, and there are more parts. The X1900 XT and XTX are very good examples of how the architecture is evolving.

ATI has set out to improve shader performance, specifically pixel shader processing performance. Instead of creating more Texture Mapping Units (often referred to as “pipelines”), they have increased the amount of “shader processors” or Arithmetic Logic Units (ALUs). ALUs handle arithmetic operations and ATI’s GPU uses these for pixel shader processors. Texture operations are handled by Texture Mapping Units (TMUs), which is usually what people refer to as “pipelines.” This is why you would call the Radeon X1800 XT a “16 pipeline video card.” As you can see, that takes into account the TMUs but leaves out the ALUs. In addition, we can’t forget about vertex units, which are also very important to performance. You can’t just look at one area of a GPU these days; you have to look at the big picture.

ATI could have very well increased the TMUs to make a 24 or 32 “pipeline” video card. At this stage, ATI felt that leaving the Texture Units alone was the right choice and they decided to concentrate on what makes Pixel Shaders perform better, and that is the ALUs.

The Radeon X1800 XT (R520 GPU) has 16 TMUs (Texture Mapping Units), 16 pixel ALUs (also called shader processors), 16 ROP’s (Render Outputs, or Raster Operators) and 8 vertex units.

With the Radeon X1900 XT (R580) ATI has increased the number of ALUs (shader processors).

The Radeon X1900 XT/XTX (R580 GPU) has 16 TMUs, 48 pixel ALUs (shader processors), 16 Render Outputs (ROPs), and 8 vertex units.

ATI created a 3:1 ratio of ALUs to TMUs for a total of 48 ALUs and 16 TMUs. This increase in pixel shader processors creates three times the arithmetic processing power of the R520. This increase in ALUs accounts for the 20% more transistors that had to be added to the GPU.

It is important to note that these are not unified processors; there are still 8 separate vertex processors in the Radeon X1900. The goal of these shader processors is to improve pixel shader performance. This is the largest improvement in the Radeon X1900.

Fetch me a Fetch4

ATI has also added support for a new feature that the Radeon X1800 XT and XL did not have, Fetch4. Fetch4 is simply a feature that can help speed up Shadow Mapping.

It works by exploiting the fact that most textures are composed of color values, each consisting of four components (Red, Green, Blue, and Alpha or transparency). The texture units are designed to sample and filter all four components from one texture address simultaneously. However, when looking up different types of textures with single-component values (such as shadow maps), Fetch4 instead allows four values from adjacent addresses to be sampled simultaneously. This effectively increases the texture-sampling rate by a factor of 4.

With Ultra-Threading technology providing fast flow control and Fetch4 providing fast texture lookups, the Radeon X1900 can render attractive soft shadows at speeds approaching those of traditional hard-edged shadow mapping techniques.

Fetch4, like Dynamic Flow Control is a content driven feature, which means game content developers will have to code their games to utilize these features. This new feature is not an automatic feature that will improve your games by default; the games have to be written to take advantage of Fetch4.

Higher Z

ATI also gave their Hierarchical Z support a boost in performance with the R580.

The Radeon X1900 incorporates 50% more on-chip memory for Hierarchical Z than the Radeon X1800. This ensures that its performance does not drop off precipitously at very high resolutions.

That’s the extent of the architecture additions.