Including 8K resolution
So, Nvidia decided to share new information about its Ampere video cards. Let’s start with the performance.
As you can see in the diagrams, GeForce RTX 3070 is 60% faster than its predecessor, RTX 3080 is twice as fast as RTX 2080, and RTX 3090 is about 50-80% faster than Titan RTX, depending on the task.
The GA102 GPU has a chip area of 628.4 mm2. It is this GPU that underlies the RTX 3090 and RTX 3080. Recall that it contains 28 billion transistors. RTX 3090 contains 82 SM multiprocessors with 10,496 cores of CUDA. The configuration also includes 328 tensor cores and 82 RT cores.
The GA104 graphics processor, which is the basis for RTX 3070, contains 17.4 billion transistors and has a crystal area of 392 mm2. This card includes 46 SM multiprocessors with 5888 cores of CUDA. There are 184 tensor cores and 46 RT cores.
As you can see, there are fewer tensor cores compared to Turing maps and more RT cores. At the same time, both of them are better. In particular, second-generation RT cores have received a component of time scale triangle interpolation in coordination with the triangle intersection block. According to Nvidia, it should be useful during motion blur effects when tracing beams in real time.
Nvidia also published a quality image of the GeForce RTX 3080 graphics card circuit board.
We knew before that it had a “wrong” shape, but now we can look at all the details. This includes a very tight layout, a 20 phase power subsystem and an unusually located auxiliary power connector.
You can also see what the owner of RTX 3090 can count on in 8K resolution. Recall that Nvidia called this adapter the first to provide enough performance in games at such a huge resolution.
However, as can be seen from the diagram, in some games, the novelty can do only through DLSS.