Diamond ATI Radeon HD4850 Review :: ATI Radeon 4850 Features

06-20-2008 · Category: Hardware - Video Cards

By Ben Sun
  • DirectX 10.1
  • 800 Stream Processing units
  • 128-bit FP precision throughout
  • 160 texture fetches per clock
  • 128 textures per pixel
  • DXTC and 3Dc+ texture compression
  • 55nm
  • GDDR5 memory
  • 160W TDP
  • 1 TeraFLOP
  • Enhanced Geometry Shader
  • Tessellation
  • 240 GigaFLOPS Double Precision processing
  • CrossFireX up to 4 cards
  • 8x AA
  • Up to 24x CFAA
  • 16x AF
  • Unified Video Decoder 2
  • On-chip HDCP
  • HDMI
  • 965 million transistors
  • PCI Express 2.0 x16 bus interface
  • 256-bit GDDR3/GDDR5 memory
  • DirectX 10.1
  • Shader Model 4.1
  • 32-bit floating point texture filtering
  • Indexed cube map arrays
  • Independent blend modes per render target
  • Pixel  coverage sample masking
  • Read/Write multi-sample surfaces with shaders
  • Gather4 texture fetching


Brand Name

Diamond

Part Number

ATI Radeon HD4850 512MB

Graphics chip

RV770

Core Clock

625MHz

Shader Clock

625MHz

SPs

800

Fabrication Process

55nm

Transistors

965 million

Memory Clock

1GHz/2GHz

Memory Interface

256-bit

Memory  Bandwidth

64GB

Memory size

512

ROPs

16

Texture Filtering Units

80

Texture Fillrate

20 Gigatexels/second

HDCP Support

Yes

HDMI Support

Yes (Via adapter)

Connectors

2x Dual-Link DVI, TV-Out

RAMDACs

400MHZ

Bus

PCI Express 2.0

Form Factor

Single Slot

Power Connectors

6-pin

ATI’s new card is based upon their RV770 chip that is their second 55 nanometer chip. The HD4850 has 965 million transistors, a noticeable jump over that of the RV670 (HD3870) which had 666 million transistors. The die size on the HD4850 is 260mm2 which is amazing considering that the GT200 die size is around 600mm2 and the previous generation HD3870 has a die size of 190mm2.

There are 800 Stream Processing units on the HD4870, compared to the 360 found on the HD3870 from the last generation, a 2.5x increase. The 800 Stream Processors when combined with a clock speed of 625MHz for the core clock brings a total of 1 TeraFLOP of computing power on a single graphics card. This is the first TeraFLOP graphics card, if you don’t count the overclocked GTX280 from EVGA which also reaches this goal by virtue of the 1.498GHz Shader clock x3 (SMs)x 240 SPs.

Double Precision Floating Point calculations were introduced with the HD3870 series last year. The HD4870 does 240GFLOP DP calculations per second. NVIDIA’s GTX280 has a maximum DP of 90 GFLOPS per second. As the GTX280 is NVIDIA’s first DP GPU this is to be expected. The HD4850 is about 4x faster than the fastest quad-core CPU according to ATI and is fully IEEE-754 compliant precision. The move to Stream Processing allows the graphics card to do more on the GPU than simply playing games.

NVIDIA is really pushing Physics on their card with the PhysX brand and the GTX280. ATI is partnering with Intel to push the Havok gaming engine on their HD4850 and older ATI cards. Havok has been used in games like Half Life 2, World In Conflict, and Hellgate London on the PC. There are many other titles under development. The point is the HD4850 does hardware physics and with the 1 TeraFLOP of performance it should rock. Hopefully we’ll see ATI use PhysX as well, as there are a lot of games that require it for the best visuals.

Graphics DDR (GDDR) memory has been in use on graphics cards since 1999 with the release of the GeForce DDR video card. ATI is using GDDR5 memory with their HD4870 video card launching next week. This has the advantages of higher clock speeds, higher memory bandwidth for the same memory bus and lower power draw over GDDR3 memory. The 4850 has 512MB of GDDR3 memory clocked at 1.0 GHz or 2.0 GHz effective. This brings a total memory bandwidth to the 4850 of 64GB/second.

DirectX 10.1 is the latest API update from Microsoft for graphics cards. 10.1 has features like 32-bit texture filtering, independent blend modes per render target , pixel coverage sample masking, read/write multi-sample surfaces with shaders and Gather4 texture fetching. It is an improvement over 10.0 in that some applications can run in fewer rendering passes, but is mostly a minor update for games. ATI of course is touting the benefits of 10.1 while NVIDIA is downplaying it. The HD3870 was the first ATI video card to support 8x MSAA in hardware. The HD4850 is ATI’s second card to support 8x AA. NVIDIA cards also support up to 8 sample anti-aliasing but they do it in two passes instead of one as in the ATI cards. Generally, the more samples used in anti-aliasing, the better the image quality. Note, that NVIDIA’s true 8x MSAA is called 8XQ not 8x. ATI has the ability to do tent filter which improves the image quality the further the image is from the center of the pixel.


Diamond ATI Radeon HD4850 Review No AA

No AA

Diamond ATI Radeon HD4850 Review 2x AA

2x AA

Diamond ATI Radeon HD4850 Review 4x AA

4x AA

Diamond ATI Radeon HD4850 Review 8x AA

8x AA