Graphics specialist NVIDIA (NVDA -1.58%) formally announced the GeForce GTX 1060, the company's mid-range graphics processor based on its new Pascal architecture, on July 7. According to the company's disclosures, the GTX 1060 would only come in variants with 6 gigabytes of memory, contrary to rumors that there would also be a lower-cost version with 3 gigabytes of memory.
According to the generally reliable BenchLife.info, NVIDIA and its add-in-board partners plan to release presumably lower-cost GeForce GTX 1060 models. These are expected to come with with 3 gigabytes of memory and slightly reduced performance and should come approximately two to three weeks after the July 19 release of the GTX 1060.
Less memory, reduced performance
The GTX 1060 that NVIDIA announced comes with 6 gigabytes of memory and a full GP106 graphics chip with 1280 processing cores, which NVIDIA terms CUDA cores. The 3-gigabyte model of the 1060, according to BenchLife, has 128 CUDA cores disabled for a total of 1152 CUDA cores.
It's worth noting that NVIDA's graphics processors are made up of things called streaming multiprocessors (SMs). Each SM in NVIDIA's Pascal architecture is made up of 128 CUDA cores, so, as BenchLife points out, the cut-down GTX 1060 has a single SM turned off.
NVIDIA has said that its manufacturing yields on Taiwan Semiconductor Manufacturing Co.'s (TSM 0.15%) 16-nanometer manufacturing process, upon which all of the company's Pascal chips are built, are "good," but obviously not all produced chips are going to make the cut. This seems like a reasonable way for NVIDIA to salvage otherwise unsaleable chips -- a common strategy in the semiconductor industry.
What price point should NVIDIA aim for?
It will be interesting to see how NVIDIA prices this product. From a pricing perspective, the company needs to choose a price point that entices customers who aren't willing or able to pay for the full GTX 1060 to buy the product. At the same time, though, the company needs to price this product in such a way that customers with the means can easily justify, and find value in, buying the higher-priced 6-gigabyte version.
To that end, I think that a manufacturer suggested retail price of $219 would be appropriate for this product. It would noticeably cheaper than the full GTX 1060, which should make it more accessible to those on tight budgets. At the same time, for those with larger budgets and/or find value in getting more performance/longevity out of their graphics cards, the $30 isn't an unreasonable premium to pay.
I also think that the GTX 1060 with 3 gigabytes of memory could be a bit more expensive to manufacture than the previous generation GTX 960 (more on that another time), so the $20 premium to the $199 GTX 960 could also help NVIDIA and its partners preserve gross profit margin generation-on-generation.
What about the 1050 Ti?
If NVIDIA prices the 3-gigabyte GTX 1060 at around $219, then I have to wonder how the company and its add-in-board partners will price the GeForce GTX 1050 Ti that's also expected to come out at some point.
My guess is that the 1050 Ti will serve as the direct successor to the GeForce GTX 750 Ti. The 750 Ti launched at $149, but saw a price cut to $119 when the GeForce GTX 950 (a cut-down GTX 960) hit the market.
If NVIDIA prices the GTX 1060 with 3 gigabytes of memory at $219, then the 1050 Ti could occupy a price point between $149 to $179. Given that NVIDIA seems to be pricing its current-generation Pascal products at higher levels SKU to SKU, I wouldn't be surprised to see this product priced at the higher end of that range.