Introduction

We have with us the "new" NVIDIA GeForce RTX 2060 12 GB. This is for all intents and purposes a new GPU model released by the green team just a few weeks ago, at the start of December 2021. The GeForce RTX 2060 12 GB is based on the 12 nm Turing graphics architecture that debuted way back in 2018. This isn't just a memory capacity bump for the RTX 2060 that originally shipped with 6 GB, but also a cross between the original RTX 2060 and newer RTX 2060 SUPER, in that it has the GPU core configuration of the latter, but with the memory bus width of the former.

The RTX 2060 12 GB is based on the same 12 nm TU106 silicon as the original. This GPU chip physically has 2,304 CUDA cores and a 256-bit wide GDDR6 memory bus, which are maxed out on the RTX 2070. The original RTX 2060 was carved out by enabling just 1,920 CUDA cores and reducing the memory bus width to 192-bit. Faced with competition from AMD's RX 5700, NVIDIA refreshed its product-stack with the RTX 2060 SUPER, which had 2,176 CUDA cores and a 256-bit wide memory bus. The new RTX 2060 12 GB in this review has 2,176 CUDA cores, which is the same as the RTX 2060 SUPER, but with a 192-bit wide memory bus that now holds double the memory amount, at 12 GB.



Having the same CUDA core count as the RTX 2060 SUPER means the RTX 2060 12 GB has more RT cores (34 vs. 30) and Tensor cores (272 vs. 240) than the original RTX 2060. Interestingly, the card's typical board power is higher, at 185 W, as opposed to 175 W with the RTX 2060 SUPER. This probably has to do with the double-density memory chips, or NVIDIA simply giving it more boosting headroom. The maximum boost frequency is set at 1650 MHz, which is the same as the RTX 2060 SUPER.

Why NVIDIA had to resort to this move is anyone's guess. AMD's Radeon RX 6600 XT and RX 6600 offer nice levels of performance compared to the current-gen RTX 3060 Ampere, and at least on the desktop segment, NVIDIA didn't further cut down its 8 nm GA106 silicon. The company probably had some 12 nm inventory to go around, and 16 Gbit 14 Gbps GDDR6 memory chips are probably priced similarly to what 8 Gbit chips cost back in 2018. The PCB complexity is kept to a minimum with just six memory chips across the 192-bit bus, while enabling a handful more CUDA cores costs practically nothing for NVIDIA.

Also, it may seem like a 3-year old graphics architecture, but Turing is still contemporary. It meets the full DirectX 12 Ultimate feature-set, including real-time ray tracing. We've been testing the RTX 2060 and RTX 2060 SUPER through all these years and find them to be formidable at 1080p with ray tracing, or even 1440p with fairly high settings, on today's games.

NVIDIA has no reference-design or Founders Edition card for the GeForce RTX 2060 12 GB. The card was stealthily launched on December 8, 2021 through its board partners. We have with us a ZOTAC GeForce RTX 2060 12 GB graphics card we bought in retail. The card sticks to NVIDIA-reference clock speeds and has a typical board power of 185 W, which lets Zotac make do with a single 8-pin PCIe power connector, a configuration rated for 225 W. ZOTAC's card features a compact board design with a simple twin-fan cooling solution. Pricing of the card varies. I bought mine in retail for €599 including 19% VAT, so that converts to roughly $599.