Last month, graphics specialist NVIDIA (NASDAQ:NVDA) had some bad news to share with investors. Thanks to the waning of GPU-based cryptocurrency mining, demand for its midrange graphics processors slowed so much that NVIDIA's distribution partners were left with substantial inventories of graphics cards based on the company's GTX 1060 graphics processing unit (GPU).
NVIDIA said in its most recent CFO commentary that its guidance for the current quarter assumed "no meaningful shipments of midrange Pascal GPUs during the quarter." (The GTX 1060 is based on the company's older Pascal architecture.)
Once that channel inventory returns to acceptable levels, NVIDIA will need to start shipping midrange graphics cards to its partners again. According to the website VideoCardz, which cited sources at graphics board maker Gigabyte, NVIDIA will soon launch the successor to the GTX 1060, known as the RTX 2060, based on its newer Turing architecture for the midrange desktop GPU market.
Let's look at some of the details of that card and what it means for that segment of the market.
It's based on the TU106 chip
This is going to get a little technical, but bear with me. Earlier this year, NVIDIA announced three gaming-oriented GPUs: the RTX 2080 Ti, RTX 2080, and RTX 2070. Each of these GPUs is based on a unique chip. The RTX 2080 Ti is based on NVIDIA's largest and most powerful Turing-based GPU, known as TU102, while the RTX 2080 is based on a less powerful chip known as the TU104.
Finally, the company's RTX 2070 is based on an even less powerful chip called TU106.
According to VideoCardz, the RTX 2060, like the RTX 2070, will be based on the TU106 chip. NVIDIA needs to make sure that the RTX 2060 is slower than the RTX 2070 (otherwise nobody would be willing to pay more for the RTX 2070), so, per the leak, NVIDIA will be disabling some of the graphics cores on the TU106 chip to create the RTX 2060 GPU. For those of you who like numbers, the RTX 2060 will reportedly have 1,920 graphics cores (NVIDIA calls them "CUDA cores") enabled out of a total of 2,304.
Also keep in mind that a typical graphics card isn't just a GPU -- it's a GPU mounted to a board. That board includes, among other things, multiple gigabytes of dedicated graphics memory. While the RTX 2070 includes 8 gigabytes of the latest graphics memory, known as GDDR6, the RTX 2060 will reportedly come with only 6 gigabytes.
Since the RTX 2060 cards are supposed to be cheaper than the RTX 2070 parts (although by how much we won't know until pricing becomes public), it makes sense for the former to have less memory than the latter. Cutting back on DRAM translates into lower materials costs and makes the RTX 2070 (which NVIDIA will surely want to up-sell customers to if it can) look more attractive by comparison.
What to expect on pricing
So far, each Turing-based card has been more expensive than its direct Pascal-based predecessor. For example, the RTX 2080 Ti is supposed to start at $999 (street pricing is currently north of $1,200, though), while the older GTX 1080 Ti could be had directly from NVIDIA for $699. The RTX 2080 is supposed to start at $699, while the prior-generation GTX 1080 launched at a manufacturer's suggested retail price (MSRP) of $599 (later lowered to $549). And finally, while the MSRP of the GTX 1070 was $379 at launch, the RTX 2070's MSRP is $499.
I'd expect, then, that the RTX 2060 will be a pricier product than the GTX 1060 cards are. That potentially higher pricing should allow NVIDIA and its partners to launch the RTX 2060 before channel inventories of the GTX 1060 are fully depleted, if necessary, since the GTX 1060 could be sold alongside the RTX 2060 as a lower-cost option.