It is really happening, folks. NVIDIA is just days away from announcing a new generation of graphics cards for gamers, presumably based on its recently unveiled Turing architecture. If the leaks and hints to this point prove true, NVIDIA will roll out a GeForce RTX 2080 graphics card with ray-tracing capabilities, and beefy specs for glorious 4K resolution gaming.
At this point, we pretty much know NVIDIA's next gaming card will be called the GeForce RTX 2080. NVIDIA dropped some strong hints of this in a recent teaser video. And if that is the case, then ray-tracing will figure prominently in its release, at least from a marketing standpoint, just as it is a point of focus with NVIDIA's new Quadro RTX graphics cards for professionals. But what else can we expect?
Everything else is speculation at this point, but assuming NVIDIA's gaming cards share the same Turing architecture as the Quadro RTX parts, we can reasonably surmise what the specs might be. NVIDIA is billing its Turing architecture as the "greatest leap since the invention of the CUDA GPU in 2006." What we have to go on right now are the following specs:
- Quadro RTX 8000: 4,680 CUDA cores, 576 Tensor cores, RT cores, 48GB GDDR6 memory
- Quadro RTX 6000: 4,608 CUDA cores, 576 Tensor cores, RT cores, 24GB GDDR6 memory
- Quadro RTX 5000: 3,072 CUDA cores, 384 Tensor cores, RT cores, 16Gb GDDR6 memory
The two higher end models both sport a 384-bit memory bus, while the Quadro RTX 5000 has a 256-bit memory bus. Each one is also equipped with ray-tracing cores.
There are two different Turing GPUs at play above. Assuming the GeForce RTX 2080 is based on the same one as found in the Quadro RTX 5000, it will have 3,072 CUDA cores and ray-tracing cores (hence the RTX branding), but probably not any Tensor cores. We would also expect 192 TMUs, 64 ROPs, and likely 8GB of GDDR6 memory on a 256-bit bus, for 448GB/s of memory bandwidth.
It is also possible that the GeForce RTX 2080 would be a toned down version of the Quadro RTX 5000, with 2,944 CUDA cores instead. Rumored clockspeeds are all over the place, but in either case, we could be looking at around 1,800MHz for the base clock and 2,000MHz for the boost clock.
Let's assume NVIDIA goes with the higher end configuration. If so, here's how things shake out compared to the GeForce GTX 1080 and 1080 Ti:
- GeForce RTX 2080: ~12.2 TFLOPS FP32, 448GB/s memory bandwidth, 170-200W TDP
- GeForce GTX 1080 Ti: 11.3 TFLOPS FP32, 484 GB/s memory bandwidth, 250W TDP
- GeForce GTX 1080: ~8.9 TFLOPS FP32, 320GB/s memory bandwidth, 180W TDP
Real-time ray-tracing powered by NVIDIA's Turing architecture
The GeForce RTX 2080 would be faster than both cards, though it would not blow the GeForce GTX 1080 Ti out of the water. However, NVIDIA leaves itself room for a GeForce RTX 2080 Ti with a higher power profile. In addition, the ray-tracing cores would pay big dividends in games and applications that can actually use them, which we hope will grow in number once NVIDIA's new cards are out in the wild.
NVIDIA's GeForce RTX 2080 should also excel at 4K resolution gaming, similar to (and slightly better) than the GeForce GTX 1080 Ti. The only question is, how much will it cost? We will find out soon enough—NVIDIA is expected to unveil its new cards on August 20, a day ahead of Gamescom.