There’s no getting away from ray tracing in video games, and that’s a fact. In 2018, Nvidia’s initiative to endow Turing GPUs with blocks responsible for accelerating ray tracing was met with rather cool reception. Which is not surprising, because advanced graphic effects could only be appreciated in Battlefield V. Two years later, the list of games was significantly expanded, and AMD Radeon graphics adapters and even game consoles can now boast Ray Tracing support.
This fall, two major GPU developers have released new generations of gaming graphics cards. Nvidia brought to market the GeForce RTX 3000 line based on 8nm Ampere cores, and AMD responded with the release of a trio of Radeon RX 6000 graphics cards. They were based on the 7nm Navi 21 graphics chip, also known as “big Navi “. We will talk about these and other innovations in 2020 from the field of discrete graphics, including Intel GPUs, in this material.
Race to the front
As part of the January CES 2020 exhibition, the presentation of the Radeon RX 5600 XT video card took place. This model is positioned by AMD as a product for “ultimate 1080p gaming” and, roughly speaking, is a stripped down version of the Radeon RX 5700 with a suggested price of $279. The main difference between the novelty is a 192-bit memory bus instead of a 256-bit one and, as a result, a smaller video buffer – 6 GB. Otherwise, these video cards are almost indistinguishable and often use the same combination of PCB and cooling system. Our review can be found at the link.
Both the announcement of the Radeon RX 5600 XT and Nvidia’s reaction to it attracted attention. The very next day, information appeared that, as a response to the new AMD product, the “green” plan to reduce the recommended price of the GeForce RTX 2060. And so it happened. In mid-January, the junior member of the GeForce RTX family officially dropped to $299. Along with this, part of the GeForce RTX 2060 cards performed by Nvidia’s AIB partners was transferred to a stripped-down TU104 crystal, a GPU rejection for older models.
The sparring between the two video core developers did not end there. Immediately after the GeForce RTX 2060 price cut, AMD allowed its partners to experiment with factory overclocking of the Radeon RX 5600 XT. Vendors began to rush to release new VGA-UEFI firmware, increasing the performance of the video card by 10-11%. This happened a few days before the start of sales of the Radeon RX 5600 XT, and the first batches of 3D accelerators have already been shipped to retail chains. Their owners were asked to update the firmware themselves by downloading it from the official website of the manufacturer.
The third player in the GPU market. Almost
Back in 2018, Intel announced plans to enter the discrete graphics adapter market, competing with AMD and Nvidia. Shortly before this, the processor giant had lured a lot of specialists from the field of GPU construction to itself, and the newly formed division was headed by the former head of the Radeon Technologies Group, Raja Koduri. The published teaser showed an expansion card for desktop PCs, but in the course of work, Intel had to reduce its ambitions.
Starting in January, Intel has been fueling interest in the graphics processor with the uncomplicated name Discrete Graphics 1, the first carrier of the Xe architecture. This GPU has not become something outstanding against the background of AMD and Nvidia solutions, however, it feels quite confident in its niche. The first device based on it was the Iris Xe Max laptop 3D accelerator.
In terms of characteristics, this graphics adapter is close to the iGPU of Tiger Lake processors. All the differences come down to increased operating frequencies and a 4-gigabyte LPDDR4-4266 video buffer with a 128-bit bus. On the other hand, in the benchmarks, the Iris Xe Max proved to be a worthy competitor to entry-level mobile video adapters like the Nvidia GeForce MX350. For the time being, one cannot count on more. The Intel DG1 graphics processor also served as the basis for the 3D accelerator for H3C XG310 servers, which uses four of these GPUs at once.
If in the desktop segment AMD can still offer rivals for GeForce video cards, then in the field of laptop gaming, Nvidia rules the show. In April, Jensen Huang’s team added GeForce RTX 2070/2080 Super solutions to the already rich range of mobile 3D accelerators. These graphics adapters offer ~10% better performance than last year’s products and have found their way into a wide variety of 10th Gen Intel Core processor-based laptops.
Meanwhile, AMD has been trying to establish itself in this segment with its Radeon RX 5000M graphics cards. Back in January, it was announced that the Radeon RX 5500M released in the fall will be joined by the Radeon RX 5600M and RX 5700M. The first 3D accelerator was adopted by several manufacturers of gaming laptops, such as MSI and Dell, but the release of the flagship took place only on paper. Apparently, AMD decided not to waste time and resources on its release, given that the next generation of GPUs are only a few months away.
GeForce RTX 3000
Autumn pleased gamers and PC enthusiasts with new video cards, and this time not only Nvidia, but also AMD could offer high-end products. On September 1, the green team led by Jensen Huang announced the GeForce RTX 3000 graphics adapters. They are based on GPUs with the Ampere microarchitecture, previously tested on the A100 computing accelerator. However, for gaming video cards, the structure of streaming multiprocessors (Streaming Multiprocessors) has been seriously redesigned.
The Nvidia GA102 graphics chip, which served as the heart of the GeForce RTX 3090 and RTX 3080 graphics cards, physically contains 10,752 FP32 compute units (CUDA cores) and has 28.3 billion transistors, which is a record for a consumer-grade GPU. Please note that we are talking about the maximum configuration, while Nvidia uses stripped-down versions of the crystal in end devices.
Key features of Ampere gaming GPUs also include next-generation tensor and RT cores, which provide a performance boost for machine learning and ray tracing tasks. For the release of graphics processors, Samsung’s 8nm technology is used.
A distinctive feature of the GeForce RTX 3080/3090 is a video buffer based on GDDR6X chips. This type of memory is manufactured by Micron Technology and has an effective frequency in the region of 19-19.5 GHz. Nvidia equipped the GeForce RTX 3090 with a 24 GB video buffer with a bandwidth of 936 GB / s, while the GeForce RTX 3080 is content with 10 GB of GDDR6X memory at 760 GB / s.
Nvidia GA102 GPU in GeForce RTX 3090
For the GeForce RTX 3070 and RTX 3060 Ti, the market debut of which took place in November-December, the chipmaker has prepared the GA104 core. Like the “big brother”, it is produced at Samsung’s facilities according to 8nm standards, but at the same time it contains 17 billion transistors with 6144 CUDA cores. In these cards, Nvidia decided to limit itself to slower GDDR6 memory, which is equipped with an 8-gigabyte video buffer with a bandwidth of 512 GB / s.
In the second generation of GeForce RTX video cards, the green ones surprised not only with a record number of CUDA cores, but also with the outstanding design of the Founders Edition models. For the GeForce RTX 3080/3090, the chipmaker has developed an air cooler with fans on opposite sides. Also worth mentioning is the 12-pin auxiliary power connector that all Founders Edition cards are equipped with. According to Nvidia, thanks to him, the company’s engineers were able to free up space on the PCB, making it more compact, and optimize the power supply system for graphics adapters.
The gaming performance of the GeForce RTX 3000 graphics cards was at its best. The average superiority of the flagship GeForce RTX 3090 over the GeForce RTX 2080 Ti is close to 50%, but its recommended price is one and a half times higher: $1499 versus $999. On the other hand, the 24 GB video buffer makes it the successor to the Titan RTX ($2499), and its performance is noticeably higher.
A more rational purchase would be the GeForce RTX 3080. It offers a third more performance than the previous generation flagship, starting at $699. At the same time, the GeForce RTX 3070 was the equivalent of the GeForce RTX 2080 Ti at half price ($499), and the GeForce RTX 3060 Ti ($399) became a full replacement for the GeForce RTX 2080 Super ($699). Perhaps the only drawback of Nvidia Ampere video cards is the terrible availability and, as a result, one and a half to two times increased retail prices.
With the RDNA architecture, which debuted with the Radeon RX 5000-series graphics cards, AMD has succeeded in bringing the Zen of the world of 3D accelerators to life. If the 1st generation Navi GPUs could not yet impose a fight on Nvidia in all segments, then with the Navi 2X cores, the Reds made a triumphant return to the high-end segment of gaming graphics cards. Roughly speaking, the RDNA 2 microarchitecture repeated last year’s Zen 2 success in the processor market.
The flagship GPU based on the 2nd generation RDNA architecture is AMD Navi 21, also known as “big Navi”. It is produced at TSMC’s facilities using 7nm technology and has 26.8 billion transistors versus 10.3 billion for last year’s Navi 10. More than 2.5-fold growth in the number of elements is due not only to twice the number of stream processors (5120 versus 2560 units), but also the appearance of Ray Accelerators blocks responsible for hardware acceleration of ray tracing. Do not forget about 128 MB of L3 cache with the marketing name Infinity Cache, which also consumes the transistor budget.
The AMD Navi 21 crystal has found application in three video cards at once. The flagship Radeon RX 6900 XT is equipped with a full version of the crystal, the variant cut down to 4608 shader ALUs is used in the Radeon RX 6800 XT, and the Radeon RX 6800 is content with a configuration with 3840 stream processors. In all cases, the video buffer is represented by 16 GB GDDR6 with a bandwidth of 512 GB / s, which distinguishes AMD’s new products against the background of 8- and 10-GB solutions from Nvidia.
Like Nvidia, AMD paid a lot of attention to the cooling system of the reference modifications. The chipmaker abandoned the standard “turbine” hated by many in favor of a cooler with three 80mm fans. By design, it resembles the CO of the Radeon VII video card, in particular, an evaporation chamber is used, and a graphite thermal interface is applied to the graphics processor. In terms of temperature and noise characteristics, reference solutions are not inferior to many non-reference models.
In games without ray tracing, the performance of Radeon RX 6000 graphics cards is fully consistent with the recommended cost. For example, the Radeon RX 6800 is noticeably faster than the GeForce RTX 3070, but is priced at $80 more. The Radeon RX 6800 XT model is inferior to the GeForce RTX 3080 by a few percentages of fps, but in the reference version it will cost $ 50 less, and the flagship Radeon RX 6900 XT with a declared price of $ 999 is located between the GeForce RTX 3080 and RTX 3090.
AMD Radeon RX 6800 XT Reference PCB
The weak point of AMD Navi 21 3D accelerators is ray tracing. Here, the Radeon Technologies Group was able to catch up with the GeForce RTX 2080 Ti, released more than two years ago. On the other hand, even this result does not give up on ray-traced effects, but for a comfortable frame rate, you will have to limit yourself to WQHD resolution (2560×1440 pixels).
The older Radeon RX 6000 graphics cards suffered the same trouble as the GeForce RTX 3000. If Nvidia Ampere 3D accelerators are available in stores at an inflated price, then AMD Navi 21 graphics adapters have not even appeared in many large venues. For the most part, the release of the Radeon RX 6000 took place only on paper. It remains to be hoped that the company will solve the supply problems in the first months of next year.
Overview of the main events of 2020. Processors and platforms
Overview of the main events of 2020. Consoles, memory and new trends