Nvidia’s Maxwell architecture was kicked off on the desktop by the release of the GeForce GTX 750 and GTX 750 Ti, and now we’re seeing the first hints of what the new architecture can achieve with the GTX 860M. A series of leaked benchmarks have surfaced on the internet which show off rather impressive performance numbers for Nvidia’s Maxwell over the older Keplar architecture.
From the initial impressions, it would seem we’ll be dealing with a very similarly specced card as the GTX 750 Ti, which bodes well for those wishing to game at 1080P. We’ll be seeing the card with 640 CUDA cores, 40 TMU (Texture Mapping Units) and 16 ROPS. Unfortunately we’re unable to get a reading on what the GPU’s core clock speed is for the GTX 860M as GPU-Z isn’t able to accurately take its reading. The card is running at 1253MHZ for the memory though, or 5012 MHZ effective.
5339 is what the GPU is hitting in 3d Mark 2011, which is slightly better than the GTX 770M. On the extreme mode however, performance does dip a little and the card is hitting slightly below the 770M’s performance target. This is possibly due to the narrower bus width (128-bit rather than 192 bit). The GTX 860M only offers 80GB/s memory bandwidth, which is fairly ‘tight’. It’s likely that the 2GB of RAM will be more than sufficient for most games at the 1080P resolution however.
It’s likely (if the desktop’s GTX 750 Ti is anything to go by) that the card will be highly overclockable. Hopefully we’ll see a situation where we can push the memory to a higher frequency. Hitting 90GB/s (requiring around 5500MHZ effective) would certainly help the bottleneck imposed by the tight bus. That’d go somewhat against the power savings offered by Maxwell, when you consider the GTX 860M is only a 45 watt part, compared to the 770M’s 75W.
It makes you wonder what type of performance we’ll be seeing from the Maxwell’s equivalent of the GTX 780 Ti, does it not?