Table of Doom
It would be prudent to take a look at our in-depth Radeon HD 4800-series review because many of the advancements made by Radeon HD 4600, over and above HD 3600, are derived from the same design decisions that makes the higher-end SKU so much better than the 38x0 it replaced.
HD 4600, we note, arrives in at least three configurations.
Let's trot out the ubiquitous Table of Doom and highlight the new SKUs' specifications.
| Graphics cards | ATI Radeon HD 4670 GDDR3/DDR3 | ATI Radeon HD 4650 DDR2 | ATI Radeon HD 3650 GDDR3 | ATI Radeon HD 3850 GDDR3 | NVIDIA GeForce 9500 GT GDDR3 | NVIDIA GeForce 9600 GT GDDR3 | NVIDIA GeForce 9600 GSO GDDR3 | |
|---|---|---|---|---|---|---|---|---|
| PCIe | PCIe 2.0 | |||||||
| GPU clock | 750MHz | 600MHz | 725MHz | 666MHz | 550MHz | 650MHz | 550MHz | |
| Shader clock | 750MHz | 600MHz | 725MHz | 666MHz | 1,400MHz | 1,625MHz | 1,375MHz | |
| Memory clock (effective) | 2,000MHz/1,800MHz | 1,000MHz | 1,600MHz | 1,656MHz | 1,600MHz (GDDR3) | 1,800MHz | 1,600MHz | |
| Memory interface, size, and implementation | 128-bit, 512MB/1024MB GDDR3/DDR3 | 128-bit, 256/512MB, DDR2 | 128-bit, 256MB/512MB, GDDR3 | 256-bit, 512MB, GDDR3 | 128-bit, 256MB/512MB, GDDR3 | 256-bit, 256MB/512MB, GDDR3 | 192-bit, 384/768MB, GDDR3 | |
| Memory bandwidth | 32GB/28.8GB/sec | 16GB/sec | 25.6GB/sec | 53GB/sec | 25.6GB/sec | 57.6GB/sec | 38.4GB/sec (card) | |
| Manufacturing process | TSMC, 55nm | TSMC, 65/55nm | TSMC, 65/55nm | TSMC, 65/55nm | ||||
| Transistor count | 514M | 514M | 378M | 666M | ? | 505M | 754M | |
| Die size | 146mm² | 146mm² | 118mm² | 192mm² | ?mm² | ?mm² | ?mm² | |
| DirectX Shader Model | DX10.1, SM4.1 | DX10, SM4.0 | ||||||
| Vertex, fragment, geometry shading (shared) | 320 FP32 scalar ALUs, MADD dual-issue (unified) | 320 FP32 scalar ALUs, MADD dual-issue (unified) | 120 FP32 scalar ALUs, MADD dual-issue (unified) | 320 FP32 scalar ALUs, MADD dual-issue (unified) | 32 FP32 scalar ALUs, MADD dual-issue (unified) | 64 FP32 scalar ALUs, MADD dual-issue (unified) | 96 FP32 scalar ALUs, MADD dual-issue (unified) | |
| Peak GFLOP/s | 480 | 384 | 174 | 426.2 | 134.4 | 312 | 396 | |
| Data sampling and filtering | 32ppc address and 32ppc bilinear INT8 filtering (16ppc FP16), max 16xAF | 32ppc address and 32ppc bilinear INT8 filtering (16ppc FP16), max 16xAF | 8ppc address and 8ppc bilinear INT8/FP16 filtering, max 16xAF | 16ppc address and 16ppc bilinear INT8/FP16 filtering, max 16xAF | 16ppc address and 16ppc bilinear INT8/8ppc FP16 filtering, max 16xAF | 32ppc address and 32ppc bilinear INT8/16ppc FP16 filtering, max 16xAF | 48ppc address and 48ppc bilinear INT8/24ppc FP16 filtering, max 16xAF | |
| Peak GTexel/s (bilinear) | 24 | 19.2 | 5.8 | 10.656 | 8.8 | 20.8 | 26.4 | |
| Peak GPixel/s | 6 | 4.8 | 2.9 | 10.656 | 4.4 | 10.4 | 6.6 | |
| ROPs | 8 | 8 | 4 | 16 | 8 | 16 | 12 | |
| Peak TDP (claimed) | 59W | 48W | 47W | 90W | 50W | 95W | 100W | |
| Power connectors (default clock) | None | None | None | 6-pin | None | 6-pin | 6-pin | |
| Multi-GPU | CrossFireX - four-board | CrossFireX four-board | CrossFireX four-board | CrossFireX - four-board | SLI - two-board | SLI - two-board | SLI - two-board | |
| Outputs (native) | 2 x dual-link DVI w/HDCP, DisplayPort, HDMI 7.1 | 2 x dual-link DVI w/HDCP, DisplayPort, HDMI 7.1 | 2 x dual-link DVI w/HDCP, DisplayPort, HDMI 5.1 | 2 x dual-link DVI w/HDCP, DisplayPort, HDMI 5.1 | 2 x dual-link DVI w/HDCP, mini-DIN, DisplayPort, HDMI | 2 x dual-link DVI w/HDCP mini-DIN, DisplayPort, HDMI | 2 x dual-link DVI w/HDCP, min-DIN, HDMI | |
| Hardware-assisted video-decoding engine | AMD UVD 2 - full H.264 and VC-1 decode | AMD UVD 1 - full H.264 and VC-1 decode | NVIDIA PureVideo3 HD - full H.264 and VC-1 decode | NVIDIA PureVideo2 HD - full H.264 and partial VC-1 decode (no entropy) | ||||
| Reference cooler | single-slot | single-slot | single-slot | single-slot | single-slot | single-slot | single-slot | |
| Retail price (default-clocked model) | £50-£55* | £40-£45* | £35 | £60 | £45 | £70 | £70 | |
* estimated pricing.
