Review and testing Radeon RX 5700 XT and Radeon RX 5700 by MSI

Summer 2019 is under the banner of AMD. The company introduced a new generation of Ryzen processors based on the Zen 2 architecture and graphics cards from the Navi family. The arrival of the new Radeons is a milestone with major changes to the graphics architecture and numerous improvements. This is an important milestone in the development of AMD graphics solutions and will also mark the emergence of mainstream solutions based on the advanced 7nm process technology. In this review, we will get acquainted with the Radeon RX 5700 XT and Radeon RX 5700, evaluate their thermal and noise characteristics, check the overclocking capabilities, compare with old AMD solutions and some competitive models from NVIDIA.

RDNA architecture

The new AMD GPUs use the RDNA architecture, which inherits the familiar GCN but with a number of improvements. The architecture has undergone changes at the basic level and numerous optimizations that should improve its efficiency and eliminate the weaknesses of GCN. RDNA was also designed with flexible scalability in mind, as the architecture will be used in a wide range of devices, including the next generation of game consoles, Google Stadia servers, and even Samsung mobile devices with licensed AMD graphics.

In addition to the new architecture, the Radeon RX 5700 XT and Radeon RX 5700 have several other important improvements. This is an advanced technological process for the production of GPUs according to 7 nm standards. The first gaming solution based on a 7nm chip was the Radeon VII accelerator, but now we have two video adapters for the mass market. AMD is using GDDR6 memory for the first time. And for the first time on the market, support for the PCI Express 4.0 bus has been implemented.

The presented video cards are based on the Navi 10 processor. It consists of four data processing clusters of 5 Dual Compute Units (DCU) each. The name itself hints that these are twin CU units. Each CU has 64 ALUs (as in the GCN architecture), while the DCU already has 128 ALUs. In total, the chip operates 2560 ALU. Also visible in the block diagram are four asynchronous engines and one geometry processor, which consists of four units. The structure of the cache has been redesigned and its volume has been increased. There are 64 rasterization units, which is similar to the top Vega GPUs, and in general there is a certain continuity with the older chips. Four 64-bit controllers allow you to work with memory over a 256-bit bus.

The difference in the structure of the new and old CU is shown in the bottom illustration. The number of schedulers, scalar blocks has been doubled, there is an additional cache of instructions and scalar data. And we multiply all this by two within the framework of the full Dual Compute Unit. ALUs are organized into two arrays instead of four, because the instruction execution mode has changed. Depending on the type of load, the CU works with Wave32 and Wave64 with SIMD32 data blocks. Wave32 on SIMD32 takes one clock cycle, while on GCN Wave32 on SIMD16 takes four clock cycles.

The Wave32 mode allows you to speed up the work with graphic shaders and better distribute the load across all ALUs. Dual CU allows a single DCU to operate as a powerful processing unit with large cache space.

Hardware changes will require a new approach to code optimization compared to GCN architecture solutions.

Another important change was the updated cache structure.

The bandwidth of L0 memory blocks has been doubled, the L1 hierarchy has been changed. overall latency has been reduced, and the L2 cache has reached 4 MB. It is worth noting that competitive GeForce RTX solutions based on TU102 have the same L2 volume. Reduced delays in accessing external memory.

Color data compression algorithms have been improved, and the GPU can work with them at different levels of the graphics pipeline. This will also have a positive effect on the acceleration of calculations.

Improvements have been announced for asynchronous blocks and geometry processing blocks. Energy efficiency has been improved, and the transition to 7nm itself allows for lower power consumption figures. In terms of performance per watt, the new Radeon Navi is one and a half times better than the Radeon Vega. And the overall performance of new products is higher than the representatives of the Vega families, despite the smaller number of computing units. We will see this for ourselves from the results of our tests.

The full Navi 10 configuration is active in the older Radeon RX 5700 XT graphics card – these are 2560 ALUs and 160 texture units. The Radeon RX 5700 has 2304 active ALUs and 144 texture units. Both graphics cards are equipped with 8 GB of GDDR6 memory with a bandwidth of 14 Gb / s. The younger Radeon RX 5700 card is very similar to the Radeon RX 590 in terms of the number of computing units, but even it successfully competes with the Radeon RX Vega 64. Both new items should occupy a niche between the old Radeon RX Vega and the new flagship Radeon VII, competing with the GeForce RTX 2060/2070.

It’s also worth noting that AMD actively uses 1440p resolution in its informational materials, positioning both cards for this mode. Therefore, in our testing, we will immediately concentrate on a resolution of 2560×1440.

For comparison, let’s summarize the characteristics of new and old AMD video cards in one table.

Separate explanations are required by the GPU frequency designation system. Boost works in the usual way, adjusting frequencies depending on temperatures and power consumption. The base value is the minimum guaranteed level under the heaviest load on the GPU. The Boost frequency shown is the target load frequency, but the Game Clock reflects the average gaming frequency. The Radeon RX 5700 XT has a Boost 1905 MHz and a Game Clock of 1755 MHz with a base value of 1605 MHz, while the Radeon RX 5700 has a combination of frequencies of 1465/1625/1725 MHz.

Reference versions of video cards offer classic turbine-type cooling. Let’s take a closer look at them on the next page. Additionally, it should be noted the presence of the anniversary version of the Radeon RX 5700 XT Anniversary Edition, released in a limited edition. This video card, in addition to a special status, has increased core frequencies and a higher price tag.

Navi video cards were the first to receive support for PCI Express 4.0. At the moment, this interface fully works on AMD X570 chipsets for new Ryzen processors. That is, you will get the most out of an AMD-based configuration. Real growth in games from such an interface is unlikely to be noticeable now. Acceleration is possible in professional applications for certain tasks.

As for the new graphics capabilities, everything is at the level of the previous generation. Navi graphics cards lack hardware enhancements for ray tracing and support for new DirectX 12 features such as Variable Rate Shading. All this remains exclusive to GeForce RTX, but there are still few games that support new features.

But there are new software features. AMD FidelityFX combines Contrast-Adaptive Sharpening (CAS) and Luma Preserving Mapping (LPM) technologies to improve the sharpness of fine image details. This is a clear answer to NVIDIA Freestyle. But if NVIDIA has limited compatibility with certain games, then FidelityFX will be an open platform, which will make it easier for third-party developers to work with technology.

The development of this direction is Radeon Image Sharpening technology, which will improve the image while reducing quality and resolution. It is implemented thanks to a special intelligent sharpness correction algorithm with adaptive contrast adjustment. Performance losses during Radeon Image Sharpening are minimal, at the level of 1-2%.

This technology will be a great addition for games in adaptive resolution mode or when using a lower resolution. Notably, NVIDIA has NVIDIA DLSS technology, which speeds up games with intelligent scaling. But these analogies are superficial. AMD has a post-processing that will reduce the loss in quality when the resolution is reduced in scaling mode, NVIDIA has a deep learning-based hardware method that completes the image and provides an overall performance boost. Both options can have their advantages in different situations. The advantage of AMD’s method is its easy integration, which will provide a wide list of compatible games.

Added a dedicated Radeon Anti-Lag software feature to reduce latency. This happens due to the control of the CPU and GPU during the formation of the frame queue, cutting off unnecessary requests for frame output. This improves responsiveness, which is critical for fast-paced games, especially in esports.

The video card supports HDMI 2.0b and DisplayPort 1.4 HDR interfaces. 4K@240Hz or 4K HDR@120Hz, 8K HDR@60Hz are available. There is compatibility with the Display Stream Comperssion 1.2a standard, which, for example, allows you to connect high-speed 4K monitors over a single DisplayPort cable. FreeSync 2 HDR supported.

Improved built-in media process for video decoding/encoding. H.264 hardware video decoding is supported up to 4K 150 FPS or 8K or 30 FPS, H.265 encoding is supported up to 4K 90 FPS. Supports H.265 encoding/decoding in 4K format. Hardware acceleration of VP9 playback in 4K also works, which is relevant for Youtube and Twitch services.