From 26 to 29 March, the NVIDIA GTC conference will be held, dedicated to GPU technologies. Many expect high-profile announcements from it, and there are persistent rumors that NVIDIA will finally show a product line for its main audience – gamers – of the GTX series based on the new Ampere architecture. Under the hood, a 12-nm process technology and a new GDDR6 memory are announced. All known details read below.
Recently, NVIDIA has somewhat forgotten its role in the gaming community: new video cards came out to work with neural networks, professional use in graphics and scientific calculations and gamers were content with the series GTX 10 based on the microarchitecture Pascal. Released in 2017, NVIDIA GTX Titan V primarily intended for machine learning and working with artificial intelligence, the same game series, it appears that the future architecture of Ampere should become. At the moment it is not clear whether Ampere will develop the Volta and Pascal foundations, or become a truly new product.
The only thing that is currently known with high probability is that the chips will be produced at TSMC’s 12nm process capacity, and the main memory type will be the inexpensive HBM2 or GDDR5X, and the GDDR6 standard developed by Samsung. Its bandwidth (up to 16 Gb / s per contact) is twice as high compared to the previous generation, GDDR5. The first in the line of GPUs will be the “consumer” GA104 and will replace at once two of its predecessors: GP104 (used in GTX 1080/70 and 70 Ti) – in terms of chip size and positioning in the market and GP102 (GTX 1080 Ti) in performance. Apparently, the production of the current generation has already been stopped, and the video cards of the current series are assembled from the remaining chips left in the warehouses.
Most likely, products based on Ampere will show the same leap in speed with graphics operations as it was in the transition from Maxwell’s architecture to Pascal, when the consumer segment models easily outperformed the results of the flagship solutions for enthusiasts of the past year.
It should be noted that there are still no official schemes for the name of the video cards released under the GeForce GTX Ampere, and the “2080” and “2070” flashing in the network are used only for the convenience of notation. Will the company change the first digit in the title to demonstrate the “new generation of GPU”, or will it remain true to the old scheme, and we will get the 11th series, we will learn together with the release of the novelty.
Another important debut for NVIDIA is the Turing graphics architecture, specially designed and optimized for artificial intelligence and machine learning. Previously, NVIDIA used the same architecture, optimized for different markets. It is the differences in chips and configurations, for example, the limitation of performance in FP64 mode, the use of ultrafast HBM memory, it was determined whether this is a game graphics card or an accelerator, grounded on industrial and scientific applications. If the rumors are confirmed, then 2018 will be the first year for NVIDIA, when they will release two completely different products for games and computing, and not just different chips based on the same architecture.
The niche of gaming graphics cards has long been under pressure from crypto-currency miners. It would be nice if large companies really decided to completely separate these two areas, limiting the performance of gaming models in non-target tasks.