The Table Tells All
ATI's Radeon HD 4770 becomes the standard-bearer in the $99 (£85) space. Based on established Radeon 4K-series technology it mixes in some old with new. Cutting right to the chase, a table will help explain the new GPU's architecture.| Graphics cards | ATI Radeon HD 4770 | ATI Radeon HD 4670 | ATI Radeon HD 4830 | ATI Radeon HD 4850 | NVIDIA GeForce 9800 GT | NVIDIA GeForce 9600 GT |
|---|---|---|---|---|---|---|
| PCIe | PCIe 2.0 | |||||
| Manufacturing process | 40nm | 55nm | 55nm | 55nm | 55nm/65nm | 55nm/65nm |
| Transistors | 826m | 514m | 956m | 956m | 500m |
500m |
| Die size | 172mm² | 146mm² | 255mm² | 255mm² | 230mm² (55nm) | 240mm² (65nm) |
| GPU clock | 750MHz | 750MHz | 575MHz | 625MHz | 600MHz | 650MHz |
| Shader clock | 750MHz | 750MHz | 575MHz | 625MHz | 1,500MHz | 1,625MHz |
| Memory clock (effective) | 3,200MHz | 2,000MHz | 1,800MHz | 1,986MHz | 1,800MHz | 1.800MHz |
| Memory interface and size (usual) | 128-bit, 512MB, GDDR5 |
128-bit, 512MB, GDDR3 | 256-bit, 512MB, GDDR3 | 256-bit, 512MB, GDDR3 | 256-bit, 512MB, GDDR3 | 256-bit, 512MB, GDDR3 |
| Memory bandwidth | 51.2GB/s |
32GB/s | 57.6GB/s |
63.5GB/s |
57.6 |
57.6 |
| DirectX/ Shader Model | DX10.1, 4.1 | DX10.1, 4.1 | DX10.1, 4.1 | DX10.1, 4.1 | DX10, 4.0 | DX10.0, 4.0 |
| Vertex, fragment, geometry shading (shared) | 640 FP32 scalar ALUs, MADD dual-issue | 320 FP32 scalar ALUs, MADD dual-issue | 640 FP32 scalar ALUs, MADD dual-issue | 800 FP32 scalar ALUs, MADD dual-issue | 112 FP32 scalar ALUs, MADD dual-issue + MUL (unified) | 64 FP32 scalar ALUs, MADD dual-issue + MUL (unified) |
| GFLOPs throughput | 960 | 480 |
736 | 1,000 | 504 |
312 |
| Fillrate (GT/s) | 12 | 6 |
9.2 |
10 | 9.6 |
10.4 |
| Multi-GPU | CrossFire - four-board | CrossFire - four-board | CrossFire - four-board | CrossFire - four-board | SLI - three-board | SLI - two-board |
| Hardware-assisted video-decoding engine | AMD UVD 2 - full H.264 and VC-1 decode, plus dual-stream decode | NVIDIA's PureVideo HD - full H.264 decode and partial VC-1 decode, plus dual-stream decode | ||||
| Reference cooler | dual-slot | single-slot | dual-slot | dual-slot | dual-slot | single-slot |
| Claimed TDP | 80W | 59W | 110W | 110W | 140W | 105W |
| Current etail price | £85 (estimated) | £57 | £75 | £110 |
£85 |
£70 |
40nm manufacturing process
By dint of its nomenclature, the Radeon HD 4770 should slot in-between the HD 4670 and HD 4830/4850 GPUs, but it's not as simple as that in performance terms. ATI's keen to cut costs but still retail a potent architecture, so the HD 4770 is the first consumer card to be based on a 40nm manufacturing process. The benefits to ATI are clear, as it can ship considerably more GPUs on to a given-sized wafer than if fabricating them on the established 55nm process, and the estimated die-size of 172mm² is significantly smaller than the Radeon HD 4830/4850's.
We see the 40nm move as a dip in the water for upcoming high-end GPUs, so don't be surprised to see the next iteration of ATI hardware - RV8xx - baked on the same process.
Half-node 40nm also marks the first time that GPUs have been manufactured on a smaller process than CPUs, which are at 45nm right now. It shows that the evolution of the GPU has been forced at a faster pace than their CPU counterparts, if nothing else.
Shaders and clocks
Look a little further down the table and you'll see that the GPU ships with 640 steam processors that operate at 750MHz. Therefore it beats out a Radeon HD 4830 on pure GLOPS throughput through higher clock-speed, but gives way, just, to the 800SP HD 4850. Agnostic to the shaders, fillrate is the highest of all the cards in the line-up.
Radeon HD 4750
ATI will add to the HD 47x0 brand with the release of a slower-clocked card to be known as Radeon HD 4750. Sharing the same architecture traits as its bigger brother we reckon it will debut for around £75.
