ATI RV570 and Radeon X1950 Pro: Specifications
The RV570 graphics processor is a chip developed specifically for mainstream graphics cards. It incorporates about 330 million transistors and is manufactured at TSMC’s facilities on new 0.08-micron tech process. The optimal (from ATI Technologies’ point of view) number of functional subunits for this product class means there are less transistors in the chip and the die area is smaller as a consequence. The use of the thinner tech process contributes to the same purpose, too. More chips can be made out of one silicon wafer, and so the manufacturing cost of the RV570 is much lower in comparison with that of the R580. Combined with the specially developed PCB, this lowers the self-cost and, accordingly, the price of the RV570-based graphics card which is called Radeon X1950 Pro. The official price of this graphics card, the only product on the new GPU as yet, is $199. This is not much at all for a product that boasts better parameters than the new version of Radeon X1900 GT.
Architecturally, the RV570 is like other chips from the Radeon X1000 family and implements a modular design in which there are no “pipelines” because the pixel shader processors and the texture-mapping units are separated and governed by an ultra-threading dispatch processor that is trying to load them up in a most optimal way.
Following ATI’s 3-to-1 concept, the RV570 has 36 pixel processors, 12 TMUs and 12 raster operators (ROPs). In this aspect, it resembles the cut-down R580 in the Radeon X1900 GT.
It should be noted that the threading dispatch processor has been simplified in the RV570 in comparison with the one in the R580. The number of simultaneously processed threads of shader code is reduced from 512 to 384, which theoretically may have a negative affect on performance of the Radeon X1950 Pro, but this theory can be confirmed or refuted by real-life tests only. For comparison, the RV530, ATI’s first chip to use the 3-math1-operations-per-1-texture-operation concept, uses an even simpler scheduler that can only control 128 threads simultaneously.
There are more transistors in the RV570 than we had expected. More than in the R520 chip. As we found out in our Radeon X1900 XTX review, pixel processors account for but a small percent of transistors in the Radeon X1000 architecture, so the transistor count hasn’t been reduced greatly. The reduction is noticeable, though, and we shouldn’t forget about the new tech process, too!
The Radeon X1950 Pro specifications are almost identical to the original Radeon X1900 GT’s specs, but the new card has one important difference. Its GPU incorporates a Compositing Engine. Until today, CrossFire technology was implemented either with the help of a special Master card that carried a separate compositing engine on board (based on a Xilinx Spartan-3 FPGA) or the data was transferred via the PCI Express bus that had a negative effect on the efficiency of the multi-GPU configuration. By integrating the compositing engine right into the GPU, ATI kills two birds with one stone: they maintain the same high efficiency of hardware CrossFire but do not have to release yet another Master card. The user doesn’t have to bother about purchasing a Master, either.
In its traditional form, CrossFire made use of an ordinary DVI link to transfer data from the Slave card to the Master. This required the Compositing Engine to incorporate two TMDS receivers and two TMDS transmitters, and an external RAMDAC. With this engine now in the GPU, the additional links in the compositing engine design are removed and data flows from one card to another directly, without conversions into DVI format and back again. Otherwise, the CrossFire concept has remained unchanged.
The graphics cards are connected with two flexible cables. As opposed to Nvidia SLI, the cables will be enclosed with the cards, most likely one cable per each sample of Radeon X1950 Pro. You anyway need two such cards to build a CrossFire subsystem. In the new version of CrossFire the cards are both identical. The Master is appointed on the software level. This depends on what mainboard slot the card is plugged into. The two CrossFire connectors may mean bi-directional data transfers or a dual-channel interface or the opportunity to build graphics subsystems with three, four, five or more GPUs. According to ATI Technologies, the bandwidth of the new CrossFire interface allows using graphics modes up to 2560x2048@60Hz which should be more than enough for resolutions like 2560x1600 without dropped frames. By the way, monitors with a resolution of 2560x2048 are as yet a prerogative of health and other specialized industries and are not produced for home or office use.
So, the technical characteristics of the new Radeon X1950 Pro are very optimistic, especially in contrast with the new version of Radeon X1900 GT that has a greatly reduced core frequency. Practice shows that the original version of Radeon X1900 GT is quite capable of challenging not only the GeForce 7900 GS but also the GeForce 7900 GT card that has more pixel processors and TMUs. We can expect the same from the Radeon X1950 Pro, but as always, we can only be sure about that after we’ve done our tests. Let’s take a closer look at the new graphics card now.