The R700, RADEON HD 4800 X2 series

• Codename: R700 (two RV770 GPUs)
• Process technology: 55 nm
• 2 x 956 million transistors
• Unified architecture with an array of common processors for streaming processing of vertices and pixels, as well as other data
• Hardware support for DirectX 10.1, including new Shader Model 4.1, geometry generation, and stream output
• 2 x 256-bit memory buses, four 64-bit controllers each, supporting GDDR3 and GDDR5
• Core clock: 750 MHz (for HD 4870 X2)
• 2 x 10 SIMD cores, including 2 x 800 scalar floating-point ALUs (integer and floating-point formats, support for FP32 and FP64 in compliance with IEEE 754)
• 2 x 10 enlarged texture units supporting FP16 and FP32 formats
• 2 x 40 texture address units
• 2 x 160 texture fetch units
• 2 x 40 bilinear filtering units that can filter FP16 textures at full speed, trilinear and anisotropic filtering for all texture formats
• Dynamic branching in pixel and vertex shaders
• 2 x 16 ROPs supporting antialiasing with programmable sample patterns (over 16 samples per pixel), including FP16 or FP32 formats of the frame buffer. Peak performance is up to 32 samples per cycle (including MSAA 2x/4x and FP16 buffers), 128 samples per cycle in Z only mode
• Writing results up to eight frame buffers simultaneously (MRT)
• Integrated support for two RAMDACs, two Dual Link DVIs, HDMI, HDTV, DisplayPort

RADEON HD 4870 X2 specifications

• Core clock: 750 MHz
• Unified processors: 1600 (2 x 800)
• 2 x 40 texture units, 2 x 16 blending units
• Effective memory frequency: 3600 MHz (4*900 MHz)
• Memory type: GDDR5
• Memory: 2 x 1024 MB
• Memory bandwidth: 2 x 115 GB/sec
• Maximum theoretical fillrate: 2 x 12.0 gigapixel per second
• Theoretical texture sampling rate: 2 x 30.0 gigatexel per second
• CrossFireX connector support
• PCI Express 2.0 x16
• 2 x DVI-I Dual Link, 2560x1600 video output
• TV-Out, HDTV-Out, support for HDCP, HDMI, DisplayPort
• Power consumption: 286W (8-pin and 6-pin connectors)
• Two-slot design
• MSRP: $549

RADEON HD 4850 X2 specifications

• Core clock: 625 MHz
• Unified processors: 1600 (2 x 800)
• 2 x 40 texture units, 2 x 16 blending units
• Effective memory frequency: 2000 MHz (2*1000 MHz)
• Memory type: GDDR3
• Memory: 2 x 1024 MB
• Memory bandwidth: 2 x 64 GB/sec
• Maximum theoretical fillrate: 2 x 10.0 gigapixel per second
• Theoretical texture sampling rate: 2 x 25.0 gigatexel per second
• CrossFireX connector support
• PCI Express 2.0 x16
• 2 x DVI-I Dual Link, 2560x1600 video output
• TV-Out, HDTV-Out, support for HDCP, HDMI, DisplayPort
• Power consumption: 230W (8-pin and 6-pin connectors)
• Two-slot design
• MSRP: $399

R700 architecture

Let's proceed to the boring part. We have nothing to add to what we have already published. RV770 hasn't been modified. Its architecture has been described in detail in the corresponding review. Let's dwell on some peculiarities of using multiple GPUs. In the times of HD 3870 X2 we mentioned that it did not support PCI Express 2.0, even though it was one of the key innovations in the RV670. That graphics card used a special PCI Express bridge (PLX PEX 8547) to connect its two GPUs, which supported 48 lanes of PCI-E 1.1. Back then AMD had planned to integrate this PCI Express bridge into future GPUs.

Let's see what has been done in HD 4870 X2. The dispatcher is still installed between two GPUs to link them. To all appearances, this chip is from the so-called second generation of PCI Express bridges manufactured by the same PLX Technology, one of the leading manufacturers of PCI Express solutions for Input/Output systems and internal wiring. As the company has no products with a similar designation (see the photo in the next part of the review), this chip might have been manufactured specially for AMD. Its characteristics resemble PEX8647, which is proved by documentation published on the official web site.

The second generation of PLX bridges was announced in the beginning of 2008. It's notable for relatively low latencies (140 ns), very low power consumption (about 3.8W), and flexible configuration. This bridge provides three PCI-E ports compliant with Specs 2.0, 16 lanes each. Sixteen PCI-E lanes are dedicated to connect to each RV770, and just as many lanes are used to exchange data between a motherboard and a graphics card. The new model differs from PEX8547 (used in HD 3870 X2) in PCI Express 2.0 support, smaller package dimensions, and better power saving (the old solution consumed about 5W).

Let's analyze the diagram of links between the GPUs and a motherboard, provided by AMD. It compares HD 3870 X2 and HD 4870 X2:

You'll notice right away theoretical values of general bandwidth, which differ three-fold in these generations: 21.8 GB/s versus 6.8 GB/s (that's the total in all directions, of course). Well, let's find out where this difference comes from. Firstly, it's the effect of the bridge supporting PCI Express 2.0 instead of 1.1. This feature provided a greater part of bandwidth, doubling it in all directions. External link with the bandwidth of 0.9 GB/s hasn't changed. But the secret sideport, mentioned in early diagrams of RV770, turned out to be an additional port (similar to a 16-channel PCI-E 2.0, judging by its bandwidth) solely for data exchange between the GPUs, by-passing the bridge and a motherboard.

Will it provide any serious advantages? Unlikely. Just because the main part of limitations has to do not with bandwidth between GPUs, but with peculiarities of used algorithms. Well, theoretically, data exchange (off-screen render buffers) must go faster, but it's hardly the main limitation of AFR performance. It would have been interesting to compare performance of a single HD 4870 X2 and two HD 4870 cards (with 1 GB of video memory and operating at the same frequencies) in CrossFire, in order to evaluate contribution of this interchip channel. But in its documents, AMD prefers to compare the speed of the new card with 512 MB modifications of HD 4870 in CrossFire, obtaining performance advantages, as the latter cards suffer from insufficient video memory in heavy modes.

HD 4870 X2 itself operates as a dual-GPU system, but ATI CrossFireX technology allows to connect two such cards on a single motherboard. According to AMD, such a system provides high efficiency of multi-GPU rendering in high resolutions with antialiasing and anisotropic filtering, performance gains from each GPU reach 75-80%.

Write a comment below. No registration needed!