Tuesday, August 21, 2018

On NVIDIA GeForce RTX and the state of competition in the graphics card market

On August 20, 2018, NVIDIA launched a new line of graphics cards based on their new Turing architecture. Branded as GeForce RTX, these new cards boast several significant new features, including:
  • Fixed-function ASIC ray-tracing units, enabling real-time ray-tracing for the ultimate in realism (more on that below)
  • Tensor cores for AI. There are myriad applications for this, one of which is faster and smarter anti-aliasing.
  • Datacenter-grade NVLink connectivity for SLI, providing as much as 50 times the bandwidth of conventional SLI connections.
  • Redesigned shader cores with new features like Variable Rate Shading for higher performance.
  • More CUDA cores and higher efficiency enabled through the TSMC 12FFN process designed specifically for NVIDIA.
Undoubtedly, the most important new feature is the real-time ray-tracing capability. Ray-tracing enables the most realistic lighting effects possible by actually computing the paths that light rays travel. This process is extremely computationally expensive, which means it's long been been limited to movies and other media which do not require real-time graphical rendering. Instead, video games use simpler rasterization techniques which provide approximations of lighting effects. NVIDIA's new RTX graphics cards contain specialized hardware designed specifically to accelerate ray-tracing, enabling cinematic lighting effects that would otherwise be impossible to implement in video games and other real-time applications. NVIDIA has also developed RTX middleware, built on the DirectX Raytracing API, to enable game developers to take advantage of this functionality more easily.

However, there is a more sinister side to these new graphics cards.

The new GeForce RTX graphics cards are the most expensive gaming-oriented graphics cards NVIDIA has produced yet. The GeForce RTX 2080 Ti starts at US$999, with NVIDIA's factory-overclocked Founders Edition costing as much as $1,199. The GeForce RTX 2080 starts at $699, with the Founders Edition at $799. For comparison, the previous generation of GeForce graphics cards, the GeForce GTX 10 Series, topped out at $699 at launch.

This is, in part, enabled by what is effectively a monopoly in the high end of the gaming graphics card market. As of this writing, NVIDIA has very little competition from the upper midrange of the market and no meaningful competition at the far high end. AMD's most advanced graphics cards, the Radeon RX Vega series, suffered from poor retail availability due to extreme demand from the blockchain market as well as poor power efficiency and high cost for the level of performance it delivers (partly because it uses expensive HBM2 memory instead of the more conventional GDDR5). The Radeon RX Vega 64 performs similarly to a GeForce GTX 1080, but requires substantially more power and tends to be more expensive to the GeForce solution.

AMD's recent products have been variable in their competitiveness over the years. For the last several years up until 2017, AMD's processors have generally been inferior to those made by Intel. Furthermore, since the introduction of NVIDIA Maxwell architecture GPUs in 2014, AMD has been unable to keep up with NVIDIA's advances in power efficiency and performance. This has resulted in rapidly declining sales, greatly weakening the company's finances and limiting its ability to invest in R&D and produce competitive products. It took a complete redesign of its x86 processor architecture, known as Zen, for AMD to become competitive in processors again, and the limited resources available to the Santa Clara-based semiconductor company resulted in their much-hyped Vega graphics cards being only marginally competitive. Despite the dramatic success of their Zen-based Ryzen processors, due to AMD's financial constraints, it is unlikely AMD will be able to produce a graphics card with hardware-accelerated ray-tracing capabilities similar to what is provided in the Turing architecture quickly enough in order to remain competitive in the high-end graphics card market. We note that NVIDIA spent ten years of research and development and 10,000 man-hours working on RTX and have invested into a manufacturing process at semiconductor manufacturer TSMC called 12FFN made just for them. It is exceedingly unlikely that AMD would have the resources or money to produce anything resembling these technologies within a reasonable amount of time.

Furthermore, NVIDIA has a history of anticompetitive practices which have received an increasing amount of attention in recent years. In 2013, ExtremeTech reported that NVIDIA GameWorks, a developer support program and suite of middleware designed to simplify the implementation of advanced graphical effects, imposed restrictions on game developers which effectively barred them from optimizing these features for AMD graphics cards. When GameWorks effects were enabled, both NVIDIA and AMD graphics cards saw significant drops in performance, but AMD cards fared substantially worse than NVIDIA cards. The amount of performance lost on NVIDIA's own GPUs was significant enough that it was reasonable to suspect that NVIDIA was using GameWorks to artificially depress the performance of AMD's graphics cards, even if it meant sacrificing performance on their own products. Related to this was the excessive use of tessellation, a graphical feature designed to make surfaces look more realistic. Because NVIDIA graphics cards tended to perform better in tessellation than AMD graphics cards, some NVIDIA-sponsored games would use an excessive level of tessellation or tessellate objects or level geometry that were not visible to the player in order to make AMD graphics card perform worse than they should.

More recently, on March 1, 2018, NVIDIA announced the highly controversial GeForce Partner Program (GPP). At AMD's request, [H]ardOCP reported that this program, marketed under the guise of promoting consumer choice, imposed significant restrictions on participating graphics card manufacturers and OEMs. NVIDIA's stated goal was to ensure that gaming brands were "aligned exclusively with GeForce". This carried the critical implication that existing, well-established gaming brands like ASUS Republic of Gamers or Gigabyte AORUS were to be associated solely with GeForce graphics cards and not AMD Radeon products. OEMs swiftly but silently removed their branded Radeon products as a result. But what if an OEM refused to participate in GPP? They would have been denied critical resources from NVIDIA, including marketing funds and priority allocations of GPU chips for use in graphics cards. Despite claims to the contrary, GPP was effectively mandatory as a result of the benefits afforded to GPP members which would be refused to non-members. Needless to say, this was an anticompetitive practice, and it would not be unreasonable to state that GPP violates the Sherman Antitrust Act 1890. The law states that "[e]very contract [...] in restraint of trade or commerce among the several States [...] is declared to be illegal", and the terms that GPP imposed upon OEMs would effectively restrain the trade of Radeon graphics cards in interstate commerce. At least one OEM, ASUS, would launch a separate brand (AREZ) for AMD Radeon products to get around this limitation. In the face of ongoing bad press, NVIDIA eventually relented and shut down the program on May 4, 2018.

Given NVIDIA's history, it is a given that they are encouraging developers to implement RTX in their games in order to drive adoption of GeForce RTX graphics cards, which would disadvantage AMD Radeon graphics cards as they lack hardware-accelerated ray-tracing capability. As significant a technological advance this may be, I have very good reason to believe that this will further weaken the already low level of competition in the discrete graphics card market. In fact, I fear that NVIDIA has the financial and technical ability to, and may attempt to, drive adoption of RTX in upcoming games so aggressively that AMD graphics solutions will be effectively unusable in these games. This would prevent AMD from being able to compete in any meaningful way and compel AMD to exit the GPU market altogether, creating a total monopoly for NVIDIA in the discrete graphics card market. Although this scenario is unlikely due to the potential for legal action by regulators and competitors, the fact that this is even possible should send chills down the spine of any computer enthusiast.

If this worst-case scenario happens, NVIDIA will be able to price graphics cards arbitrarily high, potentially far beyond even the $1,199 price for the GeForce RTX 2080 Ti Founders Edition. In fact, this is not the first time something like this has happened in the consumer computing world; the Intel Core i7-6950X launched at upwards of $1,700 because there was no meaningful competition to this processor at the time. The only other company capable of competing with NVIDIA on a level playing ground is Intel, whose upcoming discrete graphics solution is not expected to be available until 2020.

In conclusion, while real-time ray-tracing capability is one of the most significant technological achievements in video game history, NVIDIA has the ability to leverage their RTX technology to cripple competition in the graphics card market, harming consumers by enabling them to set prices as high as the market will bear.

Draco

1 comment:

  1. To be clear, I am not stating that consumers should not buy NVIDIA products. In fact, I expect to continue to buy GeForce graphics cards because they are more often than not the only solution which meets my requirements. However, I put up this blog post because the potential for monopolization is there and I have a good reason to be concerned. Ultimately, it is my opinion that you should choose the products that best meet your requirements rather than refuse to buy a particular brand over reasons like the ones described in this post.

    ReplyDelete