Review GeForce GTX 680M vs. Radeon HD 7970M

The specs of the GeForce GTX 680M are a monster. Essentially a notebook model of the desktop GTX 670 (GK104 chip), the 680M houses 1344 CUDA cores, 32 ROPs, 112 TMUs, and a 4096 MByte GDDR5 video RAM connected via a 256-Bit bus. The GeForce GTX 660M (GK107), which is also based on the modern 28 nm Kepler architecture, “only” features 384 Shaders, 8 ROPs, 32 TMUs, and 2048 MByte GDDR5 VRAM with a 128-Bit interface.

Compared to the desktop GTX 670, however, Nvidia have significantly decreased the clock rates of the 680M. Instead of 915 MHz for the chip and 3004 MHz for the memory, the mobile GTX 680M is only clocked at 720 MHz and 1800 MHz, respectively. Subsequently, the bandwidth has fallen from 192.2 GB/s to 115.2 GB/s, the pixel fill rate from 29.3 to 23.0 GPix/s and the texture fill rate from 102.5 GTex/s to 80.6 GTex/s. Fortunately, the bus type (PCI-E 3.0) and DirectX 11.1 API are identical between the two models.

Similar to the GTX 580M (aka. GTX 675M) of which the 680M succeeds, the TDP is rated at 100 Watt. However, it is worth noting that the GTX 680M lacks a unique feature of the new desktop Kepler generation - the automatic core overclocking called GPU Boost.

Unlike what the name would otherwise suggest, the Radeon HD 7970M in fact does not resemble its namesake HD 7970, but is instead more closely related to the Radeon HD 7870. The 7870 and 7970M each feature 1280 shaders, 32 ROPs, 80 TMUs, and 2.8 billion transistors. The die size of the 28 nm mobile chip is 212 mm² and the interface of the 2048 MByte GDDR5-VRAM is 256-Bit wide with a Memory throughput of 153.6 GB/s.

Unfortunately, AMD has also modified the core clock rate just like what Nvidia have done to the 680M, although not as severely. While the memory clocks at 1200 MHz, the core clock has been decreased from 1000 MHz to 850 MHz. As expected, the GCN-architecture (Graphics Core Next) supports all modern features and DirectX 11.1. Its rated energy consumption of 100 Watt TDP also matches that of the GTX 680M’s.

In comparison to the old HD 6990M, the core performance of the 7970M has been significantly improved. The 40nm DirectX-11.0-GPU features only 1.7 billion transistors, a core clock of 715 MHz and a memory clock of 900 MHz with a bandwidth of 115.2 GB/s.

Nvidia takes a clearer win in terms of temperature. While the HD 7970M reaches about 83 °C  in Battlefield 3 and a maximum of 95°C in the stress test (Furmark + Prime), the 680M runs cooler at 70 °C and 85 °C, respectively.

Furthermore, the AMD GPU generally appears to drive the notebook to its limits. When under full load, the CPU often operates at its base clock rate (2.6 GHz @ Core i7-3720QM) while the CPU in the Nvidia notebook model can always use its Turbo Boost (3 GHz or more). Even worse: According to HWiNFO (GPU-Z 0.6.2 does not record correctly), the core of the HD 7970M sometimes throttles from 850 to 580 MHz. However, we did not experience any throttling during games.

Since the fans work slightly more intensively with the 7970M equipped, the surfaces of the notebook stay cooler with the Radeon GPU. Instead of 34 and 29 °C (full load / idle), the GTX 680M reaches higher averages of 35 and 31 °C.

As the screenshots on the right depict, the energy saving mechanisms are quite different on battery. Despite heavy load, the GTX 680M clocks down to 135 MHz (core) and 324 MHz (memory) during the BatteryEater Classic test. The HD 7970M idles higher at 450 and 1200 MHz, respectively, which will reduce battery life accordingly. From our tests, the Nvidia-equipped notebook achieves 100 minutes compared to the paltry 56 minutes of the AMD model. Just the opposite is true when connected to an outlet: The clock rate of the GTX 680M is surprisingly high at low loads compared to the AMD card. We suspect that this could be a bug and may be patched at a later date.

Nvidia are asking for much more money for the GTX 680M than are AMD for the HD 7970M. An almost 300 Euro ($380 USD) surcharge (see configurator) can be pretty pretty steep and seems excessive given that the Radeon GPU can perform equally as fast in most games, if not faster. The price differences for the HD 6970M / HD 6990M compared to the GTX 485M / GTX 580M were similarly large in the past.

In an attempt to avoid losing potential buyers, we believe that Nvidia should decrease the price by about 200 Euro (~$250 USD). From a price-performance perspective, the HD 7970M is significantly more appealing, at least for now.

It’s a fact that Nvidia have produced the overall better GPU with the GeForce GTX 680M. Its waste heat and energy consumption are both lower compared to the Radeon HD 7970M, leading to quieter system noise and longer battery life.

The manufacturer with the green logo continues to take the lead in supported features. In particular, PhysX and 3D Vision currently do not have any serious competition. Furthermore, AMD’s graphics switching technology is almost abysmal. Poor detection rate, few configuration options, and noticeable performance losses are just a few of the annoying problems. Nvidia’s Optimus definitely deserves praise for its much friendlier interface and better functionality.

On the bright side, the Radeon HD 7970M is not too far behind the 680M in gaming performance. In fact, both are about on par in games and the GTX 680M is even significantly slower in professional applications. However, the biggest advantage of the HD 7970M is its more reasonable price. Hardly any other high-end notebook GPU model can achieve better FPS-per-dollar ratio than this Tahiti core.

If AMD can improve upon the still-young Enduro technology, then we would recommend the 7970M without hesitation. But for now, the 680M offers the most balanced, reliable, and most versatile performance of the two.