Nvidia’s new graphics card generation has been eagerly awaited for months and finally we have some official information released by Nvidia itself. Most of the speculation was incorrect, although the architecture will indeed be called Nvidia Turing, the first card to release will not be called the GTX 1180 as popularly speculated but rather the RTX 2080, RTX standing for real time ray tracing. Nvidia revealed the core features of its Turing architecture at SIGGRAPH and teased the naming details for the upcoming gaming cards in a video posted on Youtube entitled ‘be for the game’. With a series of clues in a gaming chat such as the usernames and phrases “NOT 11”, gimme 20 and eight tee, Nvidia reveals the name of the new card.

The Turing architecture includes tensor cores, which were first used in the Volta architecture, and also has RT cores for ray tracing. The presence of tensor cores has only been confirmed for the professional Quadro RTX cards, whether these will be available in the GeForce line remains to be seen. However Nvidia did reveal a new anti-aliasing algorithm called DLAA or deep learning anti-aliasing, which suggests that tensor cores will be enabled on gaming cards. Naturally the gaming cards must have RT cores for ray tracing, otherwise the new branding would not make sense. Microsoft also recently created the DirectX Raytracing API, another big clue that ray tracing is going to be a feature for gamers in the near future. The cards will be manufactured on the TSMC 12nm FinFET process and, as a surprising revelation from the Quadro RTX announcement revealed, the highest end Turing chip will have 18.6 billion transistors and measure 754mm squared. That is a huge chip, much larger than the GP102 used in the GTX 1080 ti and only slightly smaller than the Titan V. In comparison, the GP102 has 11.8 billion transistors and a size of 471mm squared.

The top Turing chip has 4,606 CUDA cores, this is a 20 percent increase over the GP102 and a 29 percent increase over the GTX 1080 ti. It can deliver 16 teraflops of FP32 computatonal cores and has 575 tensor cores capable of 125 teraflops of FP16 performance. If the pattern set by previous generations is to be followed, this chip will be used for the RTX Titan and a cut down version will be used for the 2080 ti, with the 2080 and 2070 receiving chips that are cut down further as is the case in the Pascal lineup with the GP102 and 104 chips. This means the amazing numbers quoted for the top chip won’t actually end up in the 2080 unless Nvidia deviates from its usual pattern and makes the 2080 more expensive than its predecessor and top of the 20 series stack, which is unlikely.  Moving on to memory, there will be 2 chip designs based on information from the Quadro RTX demo, one with 24 and 48GB DDR6 and a 384 bit interface and another with a 256 bit interface and 16GB DDR6.

What about Price and release date? Well, Nvidia will hold a Geforce gaming celebration before Gamescon on August the 20th in Cologne, Germany, which is when we expect to hear the final specifications, pricing and release date. The cards will likely be released in a staggered fashion starting with the 2080 a couple of weeks after the announcement with some sources even suggesting that the first card may be launched immediately. A post on the Chinese website Baidu suggests that the 2080 will cost $649 US Dollars. This is someone who claims to be familiar with a board partner’s plans but it is not an insider source so the information should be taken with a large pinch of salt. The suggested price is $100 dollars above the 1080 launch price but considering the astronomical prices of the Quadro RTX cards, which cost up to 10,000 dollars, the 2080 price actually sounds quite low. How much of this information is actually correct remains to be seen and will probably be revealed on Monday and we will be sure to release a follow up video and article to confirm the final details.