AMD's Radeon HD 7970M, Exposed
We already examined Clevo’s P150EM in Eurocom Racer 2.0 Review: Intel's Ivy Bridge Meets Nvidia's Kepler, though Sager, a distributor of Clevo's whitebook designs, renames it to the NP9150 before it gets to Xotic PC. Aside from the logos, it’s the same chassis.
Opening up the machine reveals very similar cooling to what we saw on Nvidia's GeForce GTX 660M. The Radeon HD 7970M has the same basic cooler design as the GeForce, but with custom-matched contact surfaces.
The Radeon HD 7970 GPU is composed of nearly three billion transistors, and it takes up much of the graphics module's real estate, even in light of TSMC's 28 nm manufacturing process. But isn't AMD's Radeon HD 7970 supposed to have 4.3 billion transistors?
| Desktop vs Mobile Radeon Graphics | |||
|---|---|---|---|
| Row 0 - Cell 0 | Desktop Radeon HD 7970 | Radeon HD 7970M | Desktop Radeon HD 7870 |
| Architecture | Tahiti | Pitcairn | Pitcairn |
| Transistors | 4.3 billion | 2.8 billion | 2.8 billion |
| Engine Clock | 925 MHz | 850 MHz | 1000 MHz |
| Stream Processors | 2048 | 1280 | 1280 |
| Texture Units | 128 | 80 | 80 |
| ROP Units | 32 | 32 | 32 |
| Compute Performance | 3.79 TFLOPS | 2.18 TFLOPS | 2.56 TFLOPS |
| DRAM Type | GDDR5-5500 | GDDR5-4800 | GDDR5-4800 |
| DRAM Interface | 384-bit | 256-bit | 256-bit |
| Memory Bandwidth | 264 GB/s | 154 GB/s | 154 GB/s |
| TDP | 250 W | 75 W | 175 W |
The Radeon HD HD 7970M obviously doesn't boast the Tahiti core with which its model name is most commonly associated. Rather, it's a more power-optimized version of the same Pitcairn GPU found in the desktop Radeon HD 7870. It’s not all bad that AMD chose a smaller die for its flagship mobile GPU, we just wish it had picked a more appropriate name. At least they’re both current-generation parts based on a 28 nm node.
| Desktop vs Mobile GeForce Graphics | |||
|---|---|---|---|
| Row 0 - Cell 0 | Desktop GeForce GTX 670 | GeForce GTX 675M | Desktop GeForce GTX 560 Ti |
| Architecture | Kepler (GK104) | Fermi (GF114) | Fermi (GF114) |
| Transistors | 3.54 billion | 1.95 billion | 1.95 billion |
| Engine Clock | 915 MHz | 620 MHz | 822 MHz |
| Cuda Cores | 1344 | 384 | 384 |
| Texture Units | 112 | 64 | 64 |
| ROP Units | 32 | 32 | 32 |
| Compute Performance | 2.46 TFLOPS | 952 GFLOPS | 1.26 TFLOPS |
| DRAM Type | GDDR5-6008 | GDDR5-3000 | GDDR5-4008 |
| DRAM Interface | 256-bit | 256-bit | 256-bit |
| Memory Bandwidth | 192 GB/s | 96 GB/s | 128 GB/s |
| TDP | 170 W | 100 W | 170 W |
By comparison, Nvidia’s GeForce GTX 675M isn’t even part of the same generation. Its GeForce GTX 560 Ti (based on the Fermi architecture at 40 nm) was a top value in its day, but even that card was replaced long ago by a more-powerful 448-core version. We’d like to come down harder on AMD for its naming tomfoolery, but Nvidia is the guiltier party in this specific case.
