The outgoing year has clearly shown how important competition is for the discrete graphics market. AMD has been unable to force a battle on Nvidia in the high-end video card segment for a long time, which gives the “greens” complete freedom in pricing new products. Debuted in late summer, GeForce RTX GPUs deliver the same level of performance per dollar as their predecessors in most games. Along with this, the price of the flagship gaming graphics card has exceeded $ 1,000, and real-time ray tracing, the main “chip” of the Turing architecture, is currently used only in Battlefield V.
A lot of food for discussion on the “hardware” forums was brought by the controversial initiative of the GeForce Partner Program, the prices for video cards that increased due to the fault of the miners, the release of the AMD Radeon RX 590 accelerator, and Intel’s intention to release a new discrete GPU by 2020. It is these topics that formed the basis of our second article on the results of 2018 in the world of PC hardware.
A review of the key events of the past 12 months in the discrete graphics segment should begin with the cryptocurrency fever that broke out with renewed vigor last winter. Another relapse led to the fact that prices for video cards rose to unprecedented heights, and almost all adapters that were at least slightly suitable for this task disappeared from the shelves.
Mining farm with liquid cooling from Alphacool
According to analysts’ calculations, in the first three months of this year, “cyber miners” purchased about 1.7 million graphics accelerators.
Average price of MSI GeForce GTX 1060 Gaming X in Ukrainian stores
The subsequent drop in the rates of most popular cryptocurrencies led to the fact that two notorious GPU developers, their AIB partners and many retailers were left with huge stocks of unnecessary video cards or components for them. Most of all, judging by the news background, this problem has affected Nvidia, which still cannot solve the problem with stale GeForce GTX 10-series accelerators.
Meanwhile, the secondary market was flooded with very attractive offers from disillusioned miners seeking to quickly get rid of the “tools of labor” that were no longer needed. Thrifty gamers can get a video card of the last generation at a price two or even three times lower than the store.
GeForce Partner Program
March was remembered for the scandal associated with the Nvidia GPP (GeForce Partner Program) affiliate marketing program. According to the greens, its main goal was to provide complete transparency in the graphics card market so that non-hardware buyers can choose the right product from one of the gaming brands dedicated to GeForce solutions.
In practice, this means that participating companies will have to create a new brand from scratch or, more likely, use an existing one. For example, if the ASUS Republic of Gamers (ROG) brand participates in the GPP, only video cards, gaming PCs and laptops equipped with Nvidia GPUs could be produced under it.
AMD responds to GPP hype with Freedom of Choice video
The Taiwanese vendor would have to promote graphics cards based on AMD chips under a separate brand. And so it happened. Shortly after the announcement of the GeForce Partner Program, ASUSTeK Computer introduced the AREZ gaming brand, under which only Radeon video cards are produced. However, today such accelerators are also available under the good old ASUS ROG.
For GPP participants, Nvidia promised not only to help promote their respective gaming brands, but also to provide access to cutting-edge developments, discount and bonus programs, and to work more closely with the engineering department. What’s more, those manufacturers that opt out of the GeForce Partner Program would be at the end of the line to receive new shipments of GPUs.
Ultimately, Nvidia was forced to curtail the GPP initiative, but this does not mean that the chipmaker has ceased to use its dominant position in the market. According to industry sources, in order to gain access to the Turing GPUs used in GeForce RTX graphics cards, vendors had to buy back large amounts of Pascal GPUs left over from the cryptocurrency rush at the same time. In other words, the chipmaker does not intend to solve such problems on its own, preferring to shift them onto the shoulders of AIB partners.
New this spring
One of the most interesting releases of the first half of the year was the PowerColor Radeon RX Vega 56 Nano compact graphics card. AMD, despite initial promises, did not introduce a successor for the Radeon R9 Nano accelerator, which prompted TUL to take the initiative. Its device is characterized by a length of 170 mm and is based on a stripped-down version of the AMD Vega 10 GPU used in other accelerators of the Radeon RX Vega 56 class.
Heat dissipation from the Vega 10 die, known for its “hot temper”, is provided by a single-fan cooler with a vapor chamber at the base of the heatsink. Given that the card’s rated power consumption is 210 W, this CO does its job well, keeping GPU and memory temperatures within acceptable limits. For its small size, the PowerColor Radeon RX Vega 56 Nano video card quite expectedly paid off with turbo frequencies, which is why its larger relatives turn out to be a little faster.
At the end of March, the Taiwanese company ASRock, which has earned a reputation in the production of motherboards, joined the ranks of AMD’s AIB partners. Its debut line of Phantom Gaming (X) graphics cards initially included Radeon RX 500-series accelerators, and a few months later it was supplemented with Radeon RX Vega 56 and RX Vega 64 adapters in the reference version.
The OEM for ASRock graphics adapters is reportedly Colorful. This is evidenced by the striking similarity of Phantom Gaming (X) video cards with previously released products from the Chinese vendor. In addition, already in the summer, Colorful decided to stop producing graphics cards based on AMD chips, which were previously available under the Colorfire subsidiary brand.
Meanwhile, in the “green” camp, they were busy expanding the already rich line of accelerators based on GPU Pascal. For example, in March, a new version of the GeForce GT 1030 video card was introduced, equipped with DDR4 chips instead of the more familiar GDDR5 chips. The resulting device has three times less memory bandwidth than the usual GeForce GT 1030, which negatively affected its gaming performance.
The main complaint about the new GeForce GT 1030 is that Nvidia did not introduce fresh markings for a stripped-down version (for example, “GeForce GT 1030 MX”), releasing it under the same name as the version with GDDR5 memory. Because of this, unprepared buyers will not be able to distinguish a full-fledged GeForce GT 1030 from its less fast sister.
A large inventory of Pascal GPUs led Nvidia to the idea of releasing a GeForce GTX 1050 graphics card with three gigabytes of GDDR5. The device received a 96-bit memory bus, operates with 768 CUDA cores and occupies a niche between the GeForce GTX 1050 Ti and GTX 1050. Unlike the GeForce GTX 1060 5GB, the new product has not become exclusive to the Chinese market and is officially available worldwide.
Rays that are not
Hungry for new graphics accelerators, gamers and PC enthusiasts have been eagerly waiting for the moment when Nvidia will show the world the successors to the GeForce GTX 10-series graphics cards. The “green” chipmaker officially introduced the line of GeForce RTX gaming adapters in the second half of August, opening it with three consumer accelerators based on the Turing architecture – GeForce RTX 2070, RTX 2080 and RTX 2080 Ti. Note that a week earlier, Nvidia announced professional cards of the Quadro RTX series, but we will not dwell on them today.
The debuted devices not only surpass their predecessors in the number of active CUDA cores, but also boast GDDR6 chips, which made it possible to significantly increase the memory bandwidth with a similar bus width. Nvidia Turing GPUs use 12nm FFN (FinFET Nvidia) process technology developed by TSMC specifically for the new GPUs. If we talk about the frequency potential, then despite the use of more advanced technological standards, it remained plus or minus at the level characteristic of 16-nm Pascal chips.
Real-time ray tracing has become a key feature of Nvidia GeForce RTX graphics cards. To do this, Turing GPUs have special functional blocks – RT (Ray Tracing) cores. For example, the flagship GeForce RTX 2080 Ti has 68 such cores. In addition, the new GPUs inherited tensor cores from the Volta architecture, which are used in machine learning tasks.
Turing Streaming Multiprocessor (SM) Schematic
With hardware-accelerated ray tracing through the DirectX Raytracing (DXR) software component, video game developers can incorporate realistic shadows, reflections, and volumetric lighting into their projects, partially or completely abandoning the many ways to simulate these effects. Our material on the features of the Nvidia Turing graphics architecture can be found here.
The main disadvantage of GeForce RTX video cards was their recommended cost. Architectural innovations in tandem with the lack of competition in the high-end segment of accelerators have led to the fact that the ratio of performance and price in existing games without Ray Tracing support turned out to be at the level of solutions of the previous generation. The flagship GeForce RTX 2080 Ti outperforms the well-known GeForce GTX 1080 Ti by 35-40%, but it turns out to be just as much more expensive.
Another issue is the lack of games that play to the strengths of Nvidia’s GeForce RTX graphics cards. For example, to date, the best tracing has been implemented only in Battlefield V, and in the initial iteration, its performance left much to be desired.
It is also worth mentioning the rather frequent failures of new video cards (especially the GeForce RTX 2080 Ti Founders Edition from the first batches), however, Nvidia and its AIB partners take the brunt here, promptly exchanging faulty adapters under warranty.
In the first days of this month, the announcement of the Nvidia Titan RTX video accelerator took place. It is based on the full version of the 12nm TU102 GPU, operates with 24GB of GDDR6 video memory with a 384-bit interface, and its performance level in FP32 operations is an impressive 16.3 Tflops. The recommended price of a new “titanium” is $2,500.
Third iteration of AMD Polaris
Today, unfortunately, there is no need to talk about competitors for GeForce RTX video cards from AMD. Currently, the “red” manages to effectively compete with Nvidia only in the middle and lower price segments of graphics accelerators. Here AMD feels quite confident and continues to expand the range of video cards addressed to budget buyers. Another “new” was the Radeon RX 590 adapter, built on the basis of a 12-nm Polaris 30 crystal.
This is the third and hopefully final incarnation of the Polaris graphics architecture. If last year AMD showed the world several video cards on the updated 14nm GPU at once, then this year it limited itself to the only accelerator designed to occupy a niche between the commercially available Radeon RX 580 and Radeon RX Vega 56.
Thanks to the transition to the 12nm process technology, Radeon Technologies Group engineers managed to significantly increase the operating frequencies of the older Polaris GPU, albeit at the expense of increased power consumption. Ultimately, the Radeon RX 590 turned out to be more productive than both the Radeon RX 580 and the closest “green” competitor in the face of the GeForce GTX 1060.
Gigabyte GeForce GTX 1060 G1 Gaming D5X
Nvidia was forced to respond to the threat from AMD and prepared a new modification of the GeForce GTX 1060 equipped with GDDR5X memory. It is based on the GP104 crystal, familiar from GeForce GTX 1070 and GTX 1080 level cards, in a configuration with 1280 CUDA cores. The novelty was received by Nvidia’s AIB partners without much enthusiasm, which, apparently, is easily explained by the imminent release of relatively inexpensive GeForce RTX cards.
Intel’s plans to enter the discrete GPU market became known last year. Then the company managed to lure the former head of the Radeon Technologies Group, Raja Koduri, who headed the newly formed division of the Core and Visual Computing Group. This summer, at the SIGGRAPH 2018 conference, the chipmaker openly announced its intention to release a new discrete GPU by 2020.
However, before entering the fight for a place under the sun in the independent graphics processor market, Intel is going to increase the speed of the video cores integrated into the CPU.
Integrated graphics of the 11th generation (Gen11), which will soon replace the current Gen9.5, will not only increase the number of computing modules, but also boast a number of architectural innovations. The latter include tile rasterization, an advanced multimedia block, and support for VESA Adaptive-Sync (AMD FreeSync) synchronization technology. The next generation iGPU will be available in 10nm processors scheduled for release next year.
Intel Gen11 to be replaced by X graphics architecture in 2020e (or Xe). It will be used in both integrated and discrete video cores, which can be found in a variety of products. Intel will likely focus most of its attention on AI accelerators and data centers, where Nvidia currently dominates. As for the consumer market, Intel intends to produce GPUs for graphics cards of all price ranges.
The chipmaker, unfortunately, is in no hurry to share details about the plans for the development of its graphics architecture, limiting itself to statements that 10-nanometer standards will be used for the production of discrete video cores. It remains to wish Intel success in the development of a new line of business.
Fan concept art of an Intel gaming graphics card