AMD’s Radeon HD 6970M
We’re glad to see that even AMD is referring to its new mobile modules by the shorter “Radeon HD 6970M” moniker, as the original “Mobility Radeon HD 6970” naming scheme was terribly wordy. We’re even willing to overlook any similarities to the naming practices of its competitor, for the sake of brevity.
Speaking of similarities, Eurocom thought that Nvidia’s GTX 470M SLI would so closely-match the performance levels of AMD’s Radeon HD 6970M in CrossFire that it included a pair of each for today’s tests. If the performance is indeed similar, AMD will take the performance-value win with a colossal $219 price advantage.
A surprisingly-small die comes not from a shrink in die process, but the use of a less complex GPU. A closer look at the Radeon HD 6970M’s specs indicates that its “Blackcomb XT” is nothing more than an underclocked Radeon HD 6850 desktop part, indicating that it's based on the older VLIW5 architecture, and not the company's newer VLIW4 arrangement.
| Desktop vs Mobility Radeon Graphics | |||
|---|---|---|---|
| Desktop Radeon HD 6970 | Desktop Radeon HD 6850 | Radeon HD 6970M | |
| Transistors | 2.64 billion | 1.7 billion | 1.7 billion |
| Engine Clock | 880 MHz | 775 MHz | 680 MHz |
| Shader (ALUs) | 1536 | 960 | 960 |
| Texture Units | 96 | 48 | 48 |
| ROP Units | 32 | 32 | 32 |
| Compute Performance | 2.7 TFLOPS | 1.49 TFLOPS | 1.3 TFLOPS |
| DRAM Type | GDDR5-5500 | GDDR5-4000 | GDDR5-3600 |
| DRAM Interface | 256-bits | 256-bits | 256-bits |
| Memory Bandwidth | 176 GB/s | 128 GB/s | 115.2 GB/s |
| TDP | 250 W | 127 W | 100 W |
AMD isn’t the only one still playing the horrid name game with its notebook parts, as Nvidia’s GeForce GTX 470M appears to be nothing more than an underclocked GeForce GTX 460 SE, priced around 3x higher, without the fancy heat sinks and brackets that a desktop card must include.
| Desktop vs Mobile GeForce Graphics | |||
|---|---|---|---|
| Desktop GeForce GTX 470 | Desktop GeForce GTX 460 SE | GeForce GTX 470M | |
| Transistors | 3 billion | 1.95 billion | 1.95 billion |
| Engine Clock | 607 MHz | 650 MHz | 535 MHz |
| CUDA Cores | 448 | 288 | 288 |
| Texture Units | 56 | 48 | 48 |
| ROP Units | 40 | 32 | 24 |
| Compute Performance | 1.09 TFLOPS | 855 GFLOPS | 634 GFLOPS |
| DRAM Type | GDDR5-3348 | GDDR5-3400 | GDDR5-3000 |
| DRAM Interface | 320-bits | 256-bits | 192-bits |
| Memory Bandwidth | 133.9 GB/s | 108.8 GB/s | 72 GB/s |
| TDP | 215 W | 150 W | 50 W (GPU Only) |
Note that AMD lists TDP for a complete card, while Nvidia lists it for the GPU alone. We expect actual power consumption to be somewhat similar from the complete modules.