GeForce RTX 2080 Ti

The GeForce RTX 2080 Ti is the flagship graphics card based on the Turing architecture, and is targeted at PC enthusiasts that like to game with large 4K NVIDIA BFGD (big format gaming display) monitors, which due to their big size and lower PPI, could warrant anti-aliasing, which isn't needed for more prevalent 4K monitor sizes such as 28-inch.

Like its predecessor, the RTX 2080 Ti doesn't exactly max out the silicon it's based on. It has 68 out of 72 streaming multiprocessors (SMs) physically present on the chip enabled, which works out to 4,352 out of 4,608 CUDA cores, 544 out of 576 Tensor Cores, and 68 out of 72 RT cores. The memory bus is 352-bit wide and holds 11 GB of memory, just like its predecessor, but the memory standard is advanced to GDDR6.



NVIDIA did not mention the reference clock of this GPU in the material we're allowed to share as of now, but the rated GPU Boost frequency is 1545 MHz for the reference, and 1635 MHz for the Founders Edition SKU. The memory ticks at 14 Gbps, which is a staggering 27 percent higher than the 11 Gbps the GTX 1080 Ti enjoyed. This works out to a memory bandwidth of 616 GB/s.

NVIDIA is pricing the RTX 2080 Ti at $999, with the Founders Edition SKU going for $1,199. NVIDIA partners can sell custom-design SKUs, and so, it's unlikely you'll find cards at the base price.

GeForce RTX 2080

The GeForce RTX 2080 is the second-fastest card (for now) and could be your ticket to 4K gaming at 60 Hz. It has enough muscle to experience all the RTX goodness you're paying $699 for (at least). As we mentioned earlier, even the RTX 2080 doesn't max out the silicon it's based on. Just 46 out of the 48 streaming multiprocessors physically present on the chip are enabled.

The chip hence has 2,944 out of 3,072 CUDA cores, 46 out of 48 RT cores, and 368 Tensor Cores. Not maxing out the TU104 gives NVIDIA room to explore a slightly premium "RTX 2080+" in the future, or it could just be a means to improve yields.



The GPU ticks at up to 1515 MHz nominal and 1710 MHz GPU boost in the reference design, with a slightly elevated boost clock of 1800 MHz on the Founders Edition SKU. The memory runs at 14 Gbps, and you get 448 GB/s of memory bandwidth.

NVIDIA is launching the RTX 2080 at $699 (nominal price) and $799 for the Founders Edition SKU. These prices rival those of the previous-generation flagship GTX 1080 Ti, and better be worth it.

GeForce RTX 2070

The GeForce RTX 2070 is the latest in a long line of performance-segment SKUs NVIDIA sold tons of, be it the GTX 1070, GTX 970, or GTX 670. You get a high-end feature-set at a much lower price than the higher SKU. Unlike its predecessors, though, the RTX 2070 does not share the same silicon as the RTX 2080. It uses the Turing TU106 silicon.

As we explained on the previous page, NVIDIA was looking for a physically smaller chip for the RTX 2070, rather than wasting manufacturing costs on a larger chip with disabled SMs that are otherwise functional. The TU106 is a big step up from the GP106. It's half of the TU102 (big chip), while its predecessors have traditionally been half the second-big chip (GP106 being half of the GP104, for example).



You get the full 256-bit wide GDDR6 memory interface, same 8 GB memory amount, and same 448 GB/s memory bandwidth as the RTX 2080, but exactly half the number of CUDA/Tensor/RT cores as the RTX 2080 Ti, with 2,304 CUDA cores, 288 Tensor Cores, and 36 RT cores. The RTX 2070 is ideal for sub-4K resolutions such as 1440p, or even certain UIltrawide resolutions such as 3440x1440.

The core of the RTX 2070 is clocked at 1410 MHz, with 1620 MHz GPU Boost, which is dialed up to 1710 MHz in the Founders Edition SKU. When it launches some time this October, the RTX 2070 will start at $499, with its Founders Edition SKU going for $599.

Speculative SKUs

While NVIDIA is launching the Turing family with just the RTX 2080 Ti, RTX 2080, and RTX 2070 in the visible roadmap, the way it marshaled its TU102, TU104, and TU106 assets leads us to speculate certain possible SKUs. This is purely speculation on our part, and no information from NVIDIA suggests these even exist. Some of these may never exist.

• TITAN X Turing: Low-hanging fruit first. The TITAN X Turing, which we speculate will be priced at $1,499, if not more, could max out the TU102 silicon, including the full 384-bit memory bus, 12 GB of memory, and higher clocks.
• GeForce RTX 2080+: NVIDIA choosing not to max out the TU104 leads us to speculate that a new SKU positioned slightly above the RTX 2080 Founders Edition could be on the horizon, which has all components on the TU104 enabled, and probably higher clocks than the Founders Edition still. There's also a price-gap between the RTX 2080 and RTX 2080 Ti to fill.
• GeForce RTX 2070 Ti: What does NVIDIA do with all the TU104 chips that have fewer than 46/48 streaming processors functional? Carve out a new SKU that's positioned halfway between the RTX 2070 and RTX 2080 in terms of CUDA core count and clock speeds.
• GeForce RTX 2060 series: TU106 gives NVIDIA the freedom to not just power the RTX 2070 in its maxed out form, but also the "sweetspot" segment RTX 2060 series. This lineup may see the light of day either towards the end of 2018 or once NVIDIA is convinced current-generation inventories are digested well enough.