Nvidia's mobile Kepler benchmarked -

Hot on the trail of the Acer-Nvidia Ultrabook announcement yesterday that prominently featured a Kepler-powered machine, someone has handed over the goods to PC Perspective which was prompt to rush a quick graphics benchmark session with the device.

The ‘Timeline M3 Ultra’ from Acer, apart from being an Ultrabook, features Nvidia’s Geforce GT 640m discrete mobile GPU. Not a rebadged 500-series, but the real deal. While most review sites are still beavering away on the desktop parts, the Timeline Ultra M3 is already on sale in Asia, which has warranted a quick headline-grabbing review on the graphics.

The Geforce GT 640m carries 384 CUDA cores, according to PC Perspective, and this particular model sports 1GB of DDR3 memory. Other details were still tied down, we guess, by the NDA-like constraints.

There were a few benchmarks run: 3DMark 06, 3DMark 11, Battlefield 3, Just Cause 2 and Dawn of War 2. Overall it provides reasonable frame rates at a resolution of 1366x768 on ‘Low Detail’, for the games. This isn’t quite the smackdown one would expect, but it is a mainstream part that will squeeze 30fps-like performance when playing Battlefield 3 on high-detail.

The graphics chip also seems to be highly-optimised for DX 11 gaming, and it shows. The GT 640m can outpace a GT 555m, but if the CUDA core count is correct, there is something seriously off. Worst case scenario, the GT 555m has 2.5 times less CUDA cores than the 640m, yet only slightly less performance. This will definitely require some looking into.

Heat seems to be a problem on this fancy thin toy. Under heavy load, the heat on the bottom side of the laptop shell will reach 55 celsius, which will be unacceptable for consumers shelling out for an Ultrabook. Nvidia has suggested this is Acer’s choice and if anything, people should be ringing them up asking why on earth it was released to market.

This expensive ultratoy with state-of-the-art 28nm Kepler graphics core isn’t shaping up to be more than happy coincidence in benchmarking that might go south for Kepler. It is far from impressive, when compared to the Fermi-based GT 555m, and still bleeds heat like a radiator.

We’re not quite sure what kind of reaction this will get from the choice few who are hamstrung with NDA units in their labs currently testing Kepler on desktop, but considering the lengths Nvidia goes to keep things under wraps, we wouldn’t be surprised if Nvidia PR is having some heavy-handed conversations as we speak.