I was making my usual rounds of the Zauba database when I found something very interesting: an entry for the GP106 GPU (aka the Nvidia Geforce GTX 1060). Not only is this the absolute first confirmation that the GP106 is going to end up in a mainstream consumer format (it is thought to power the Drive PX2) very soon but it also contains another twist. The GTX 1060 will double the bus width available to its predecessor for a total of 256-bits worth of bus width. The same amount available to the GTX 1080 and GTX 1070 graphics cards from Nvidia.
Nvidia Gefoce GTX 1060 spotted on Zauba - supports a 256 bit bus with high density GDDR5 memory
The shipping entry over at Zauba mentions only the insurance value of the chip (meaningless for determining MSRP), the specifications and also the memory type. The GP106 will be using high density GDDR5 (256MX32 configuration). Since the bus width is 256-bits, we are looking at either 4GB or 8GB of vRAM. I would like to point out here that the entry only mentions the GP106 GPU, so the information contained in the manifest can pertain to either the GTX 1060 or the GTX 1060 Ti. However, I believe it is more likely that we are talking about the former, considering Nvidia's historical trend with the "Ti" nomenclature which usually tend to land after the vanilla offering. According to earlier (unconfirmed) reports, the GP106 powered graphics card will be landing in the fall. The launch schedule is however, subject to change, since Nvidia is no doubt waiting for AMD to make a move.
Not much information is known about the GP106 at this point apart from the few scattered die shots given below (courtesy of VideocardZ). The die size appears to be around 200mm² and if the GPU has two Global Processing Clusters then we can expect around 1280 CUDA cores (256 CUDA Cores per Streaming Multi Processor and 5 SMPs per GPC). The Nvidia GTX 1060 should work only on power provided by the PCIe by design but will probably have an optional 6 pin connector added by AIBs to ensure overclocking headroom. Even though the GP106 is near the bottom of the barrel of the Pascal lineup, we have already seen the high clocks that pascal can achieve, allowing it to pack a lot of punch thanks to the use of 16nm FinFETs. The time has come when entry level actually means affordable gaming performance and not a downright horrible gaming experience.
The x60 point in Nvidia's lineup has usually fallen prey to one bottleneck or the other, the last example being the GTX 960 which only had a 128-bit bus width. The low bus width of the 960 meant that despite the color compression technologies introduced in Maxwell, high resolution gaming was something that was a difficult proposition at best. With the Pascal based GTX 1060 however, things will take a drastic turn. Not only is Pascal able to achieve a very high clock rate, but VR friendly technologies like Simultaneous Multi Projection means that we are looking at a card that will be able to offer very decent gaming performance as well as compatibility with VR HMDs like the Oculus Rift and HTC Vive.
To give an example, the GTX 960 had a theoretical peak performance of 2.4 TFLOPS (1024 CUDA Cores * 2 Operations Per Second * 1178 Mhz Boost Clock) and a bandwidth of just 112 GB/s. The VR minimum spec is thought to be around the 3.5 TFLOPs mark. If the GTX 1060 is clocked anywhere near the GTX 1070/1070, you are looking at performance levels of 4.0 TFLOPs+ (Assuming a 1280 CUDA Core Count and a conservative clock speed of 1600 Mhz) with a bandwidth of ~224 GB/s. Add Nvidia's new SMP technology on top of this and you have one amazing deal in terms of a VR-Ready graphics card. We don't really have any idea about the price point at this time, but I think we are going to be looking at price levels around the $249-$300 mark. AMD aims to expand the Total Available Market for VR with their Polaris GPUs, it looks like this is Nvidia's answer to the same.