Technically this is only half of the announcement so many have been waiting for, but it is still an important half. NVIDIA announced three new graphics cards using its Turing GPU architecture, but these are all targeting the professional workspace. It is still the first official glimpse at the capabilities of the Turing architecture, and so we can get an idea of what NVIDIA is working with for its next generation of consumer-aimed GeForce GPUs.
The three cards announced are the Quadro RTX 8000, Quadro RTX 6000, and Quadro RTX 5000. One of the features of Turing these cards all share are RT Cores, as NVIDIA calls them, which are responsible for the accelerated ray tracing the manufacturer is pushing very heavily. According to NVIDIA, both the RTX 8000 and RTX 6000 are capable of 10 GigaRays/sec while the RTX 5000 reaches 6 GigaRays/sec. This lines up with the CUDA and Tensor core counts also stated, with both the RTX 8000 and RTX 6000 having 4608 CUDA core and 576 Tensor cores, while the RTX 5000 has 3072 CUDA and 384 Tensor cores. NVIDIA is claiming the 4608 CUDA cores will allow the Turing GPU to reach 16 TFLOPS and 16 IOPS (integer operations per second). All three cards will use Samsung-produced GDDR6 memory, with the RTX 8000 sporting 48 GB, RTX 6000 halving it at 24 GB, and the RTX 5000 coming in at 16 GB. Utilizing NVLink to connect two RTX 8000 cards though gets you up to 96 GB with up-to 100 GB/s data transfer.
The estimated price of these cards is $10,000 for the Quadro RTX 8000, $6300 for the Quadro RTX 6000, and $2300 for the Quadro RTX 5000. They are expected to be available in Q4 of this year. Unfortunately, we do not have any specific information on the next round of GeForce cards, although NVIDIA has shared a teaser video (below) that supposedly contains hints, and the belief is an announcement is coming at Gamescom 2018, which starts next week.
Back to original news post