GPUs have come a very long way since the early days of blocky, 16-colour graphics. They have developed from traditional graphics accelerators into core compute engines for a broad class of general-purpose applications, and because of the ever evolving technology standards, gamers, miners and researchers now demand new levels of graphics quality and computational power.
Here is what to anticipate from the key players on GPU market in 2018.
NVIDIA: One for Gamers, One for Miners
NVIDIA started at the early phase of 2018 at the base of GPU market. The company revealed the Titan V at the end of the previous year, promoting it as a super powerful consumer card. It is based on NVIDIA’s latest Volta architecture, polished up for high-performance computing and AI applications. The company claims it is nine times faster in computing power than its predecessor. However, with a $3,000 price-tag, it is not very wallet-friendly. This year, Nvidia is expected to release several cheaper models, and the GTX 20-series promises to be amongst the most anticipated launches of 2018.
The next generations of GPU are undoubtedly coming to transcend the former. According to Reuters report, NVIDIA’s new ‘gaming GPU’ is currently called “Turing”. However, the code name, which is the node to Alan Turing, popular English mathematician, computer scientists and cryptanalyst, suggests the card might be designed for miners or researchers.
Another new GPU, the one probably for gamers, is likely to be unveiled in March or April. The leak comes via German tech site 3DCenter. It claims that the new GTX 20 series of graphics cards will debut with the purported GeForce GTX 2080 under the new Ampere GPU family. They will refresh the industry-leading Pascal GPU family released back in 2016.
The site says, Nvidia has suspended the GP102 chip. Therefore, the production of current generation of GTX 1080 Ti and TITAN X will either be stopped or slowed down throughout the rest of the year.. The new GA104 (G = GeForce A = Ampere) graphics processor will be used to replace it. Nvidia’s typical release cadence suggests the GTX 20 Series will be launched with the GeForce GTX 2080 and GeForce GTX 2070. They would be followed by a more affordable GTX 2060 later in 2018 before the flagship GeForce GTX 2080 Ti is subsequently launched in 2019.
Wonder how powerful the GeForce GTX 2080 and GTX 2070 will be? Conservative estimates peg the new generation to have a potential 20–30% performance gain over Pascal. With this report fully confirmed, the GTX 20 series is likely to use Nvidia’s new and improved VRAM technology, which uses a smaller bus than the GTX 1080 Ti, thanks to the arrival of the firm’s GDDR6 technology (which is going to change things at the high-end side of memory bandwidth). In addition, NVIDIA has partnered with Samsung to make the new GPUs with its 14nm process technology.
The Ampere and Turing code names may be used to describe cards for two different markets given the new landscape: gaming and cryptocurrency mining. There are also suspicions that Nvidia will gimp the GTX 20 series cards at a hardware level to make them not suitable for mining and to ensure that the price isn’t skyrocketed. We could probably see the splint into GP104A and GP104B, where A would be a consumer Ampere GPU, and the B — a mining-specific GTX 20 series part, different in design and equipment.
NAVI GPU from AMD
AMD launched their Vega GPU architecture, the first consumer graphics cards to feature HBM2 technology, in May last year. The next generation, Navi-based professional card, is unofficially scheduled for July-August 2018. The company fans were thrilled with the new architecture since it first popped up on roadmaps, with hints of a cutting-edge memory subsystem and a “scalability” option.
Apart from the fact that Navi will be made on the 7nm process, its feature is quite unpredictable. To gain points in competition with Nvidia, AMD would make Navi a modular GPU, by putting in smaller not-so-complicated GPU dies instead of a usual massive GPU die. NVIDIA is also working on the creation of multiple GPU modules on future graphics cards instead of a huge “monolithic” GPU. If AMD were to beat NVIDIA to the modular GPU approach on 7nm, 2018–2019 would be two massive years for the overall GPU technology of the company. .
Later in 2018, AMD plans to start sampling a version on Vega on 7nm aimed at machine learning. However, ‘sampling’ in this case means pre-testing to select partners. This will likely lead to a Radeon Instinct product, initially perhaps with a small die size, being made available in early 2019.
