PowerColor Radeon RX 5700 XT Red Devil Review 40

PowerColor Radeon RX 5700 XT Red Devil Review



PowerColor Logo

PowerColor today launched the Radeon RX 5700 XT Red Devil, its flagship custom-design graphics card based on the new 7 nm AMD RX 5700 XT Navi graphics processor. This card arrives hot on the heels of AMD's add-in board partners launching custom-design Radeon RX 5700-series graphics cards a little over a month following the July 7th launch of Navi. The Red Devil is positioned as an extremely premium offering, right in the middle of NVIDIA's RTX 2060 Super and RX 2070 Super and priced at $440 ($450 for a premium package with more accessories, which we cover in this review). The PowerColor RX 5700 XT Red Devil tops Navi with a factory overclock, a gargantuan cooling solution that looks a class above, and has many enthusiast features you'd expect from much pricier cards.

The Radeon RX 5700 XT is AMD's first true performance-segment graphics card in over two years since the RX Vega series. It's based on the brand new "Navi" architecture that leverages the 7 nm silicon fabrication proces and brand new number-crunching machinery AMD calls RDNA compute units. These constitute the biggest update to AMD's GPU design since the very first Graphics CoreNext (GCN) architecture circa 2013. Together with clock speeds, RDNA is designed to bring about massive IPC improvements over GCN. The silicon also has a number of architectural changes. An interesting series of price adjustments and product launches ensures that even at its starting price of $399, it offers a bit more price-performance than NVIDIA.

AMD had originally planned to launch the Radeon RX 5700 XT at $449 and the RX 5700 at $399, with the two cards beating the $499 NVIDIA RTX 2070 and $349 RTX 2060, respectively. This forced NVIDIA to refresh its lineup with the new RTX 2070 Super at $499 and the RTX 2060 Super at $399. The RTX 2060 Super in particular was carefully crafted not to cannibalize the RTX 2070. AMD seeped into this imbroglio of NVIDIA and slotted the RX 5700 XT at $399 and the RX 5700 at $349, at which prices they outclass the RTX 2060 Super and original RTX 2060, respectively. NVIDIA didn't adjust prices of its RTX 2060 Super or RTX 2070 Super any further, and we hence have a fair bit of headroom between the RTX 2060 Super and the RTX 2070 Super in which AMD's board partners can launch custom-design RX 5700 XT cards with factory-overclocked speeds and other goodies, such as quieter coolers.

At the heart of the Radeon RX 5700 XT is the 7 nm "Navi 10" silicon with an impressive 10.3 billion transistors crammed into a 251 mm² die. Unlike the "Vega 20", Navi is a more traditional GPU in that the package only has the GPU die and is surrounded by memory chips. AMD opted for cost-effective 256-bit GDDR6 memory over exotic design choices such as HBM2. At a memory frequency of 14 Gbps, Navi enjoys a healthy memory bandwidth of 448 GB/s. It also features the latest-generation PCI-Express gen 4.0 x16 host interface with full backwards compatibility for older generations of PCIe, so you can pair it with AMD's new Ryzen 3000 processors on an X570 chipset motherboard. The buzz-words "7 nm" and "PCIe gen 4.0" are extensively used in AMD's marketing, as if to suggest that Navi is a generation ahead of NVIDIA's Turing, which is built on 12 nm and has PCIe gen 3.0.

The PowerColor Radeon RX 5700 XT Red Devil features a custom-design, triple-slot cooling solution that features a massive compound aluminium fin-stack heatsink that uses several copper heat pipes which pass through multiple fin stacks ventilated by a trio of fans. This overkill cooling solution not only gives you idle fan stop, but also supports the card's increased power limits and factory-overclocked speeds of up to 1905 MHz boost. You also get several features, such as Dual BIOS, with one of two BIOSes activating a "Quiet Mode" that significantly reduces fan noise when gaming, and an accessory package that includes an RGB mousepad. All this for a $50 premium over the $400 baseline.

Radeon RX 5700 XT Market Segment Analysis
GTX 1070 Ti$4502432641607 MHz1683 MHz2000 MHzGP1047200M8 GB, GDDR5, 256-bit
RTX 2060$2901920481365 MHz1680 MHz1750 MHzTU10610800M6 GB, GDDR6, 192-bit
RX 5700$3502304641465 MHz1625 MHz1750 MHzNavi 1010300M8 GB, GDDR6, 256-bit
GTX 1080$5002560641607 MHz1733 MHz1251 MHzGP1047200M8 GB, GDDR5X, 256-bit
RTX 2060 Super$4002176641470 MHz1650 MHz1750 MHzTU10610800M8 GB, GDDR6, 256-bit
RX Vega 64$400 4096641247 MHz1546 MHz953 MHzVega 1012500M8 GB, HBM2, 2048-bit
GTX 1080 Ti$7003584881481 MHz1582 MHz1376 MHzGP10212000M11 GB, GDDR5X, 352-bit
RX 5700 XT$4002560641605 MHz1755 MHz1750 MHzNavi 1010300M8 GB, GDDR6, 256-bit
PowerColor RX 5700
XT Red Devil
$4502560641770 MHz1905 MHz1750 MHzNavi 1010300M8 GB, GDDR6, 256-bit
RTX 2070$4402304641410 MHz1620 MHz1750 MHzTU10610800M8 GB, GDDR6, 256-bit
RTX 2070 Super$5002560641605 MHz1770 MHz1750 MHzTU10413600M8 GB, GDDR6, 256-bit
Radeon VII$6803840641802 MHzN/A1000 MHzVega 2013230M16 GB, HBM2, 4096-bit
RTX 2080$6302944641515 MHz1710 MHz1750 MHzTU10413600M8 GB, GDDR6, 256-bit

Architecture: Navi and RDNA

We've been hearing the moniker "Navi" for years now, and AMD threw another one at us this Computex, "RDNA", so let us demystify the two first. "Navi" is the codename for the family of silicon the GPU is based on. RDNA is a new architecture introduced by AMD to succeed Graphics Core Next (GCN). It prescribes the GPU's component hierarchy and, more importantly, its main number-crunching machinery, the compute units.

Another example of this distinction would be "Vega". Vega 10, Vega 20, and Vega 12 are pieces of silicon from the same family, while the GPU follows the 5th generation Graphics Core Next architecture governing even its compute units. Over many years, AMD made incremental updates to GCN, but this time, it claims that RDNA is sufficiently different from GCN to not be considered a new version, but rather a new hardware component that brings with it massive IPC gains over the previous generation.

The Radeon RX 5700 series is built around "Navi 10", an elegant little piece of silicon engineered on the 7 nm process at TSMC with 10.3 billion transistors crammed into a die measuring just 251 mm². The chip features a PCI-Express 4.0 x16 bus interface and a 256-bit wide GDDR6 memory interface. Infinity Fabric, which debuted on AMD's Ryzen CPUs, is extensively used as an on-die interconnect linking the various major components.

The bulk of AMD's engineering effort with RDNA has been to increase the number of dedicated resources to avoid starvation by fewer components waiting for access to a resource. The "Navi 10" silicon has two Shader Engines sharing a centralized Command Processor that distributes workloads, a Geometry Processor, and ACEs (asynchronous compute engines).

Each Shader Engine is further divided into two Graphics Engines. A graphics engine shares render backends, a Rasterizer, and a Prim Unit among five Workgroup Processors. This is where the core of RDNA begins. AMD figured it could merge two compute units (CUs) to share schedulers, scalar units, a data share, instruction and data caches, and TMUs. The Workgroup Processor, or "dual-compute unit" as shown in the architecture block diagram, is for all intents and purposes indivisible in that individual CUs cannot be disabled.

An RDNA compute unit packs 64 stream processors for vector operations and double the number of scalar units for localized serial processing. The stream processors in a CU are split into groups of two, each equipped with a scalar unit. According to AMD, this greatly reduces latency and improves the overall IPC of the compute unit. It also more efficiently utilizes local caches.

The vector execution units, or stream processors, is where much of the GPU's parallel processing happens. Due to the redesigned compute unit, two scalar processors pull two SIMD32 vector units made up of 32 stream processors, each, instead of a single scalar processor pulling four SIMD16 vector units. How is this important? On GCN, the way SIMD units are laid out, all items in a Wave64 operation get to do work once every four clocks due to hardware interleaving. With RDNA, Wave32 work items can do work every clock cycle. In all, RDNA minimizes wasted clock cycles by more efficiently and uniformly utilizing the hardware resources.

AMD examined previous generations of its graphics architecture to locate bottlenecks in the graphics pipeline. Besides increasing the number of dedicated resources, the company reworked the chip's cache hierarchy by cushioning data transfers at various stages. Each workgroup processor has dedicated 32 KB instruction and 16 KB data caches, which write back to a 128 KB L1 cache dedicated to each Graphics Engine.

These L1 caches talk to 4 MB of L2 cache. The introduction of the L1 cache and doubling in bandwidth between the various caches contributes greatly to IPC as it minimizes memory accesses, which are much slower than cache accesses. AMD is also using faster (lower latency) SRAM that reduces cache latencies by around 20 percent on die and by 8 percent at the memory level. AMD also introduced new features to the ACEs that include async-compute tunneling.

AMD summarizes the benefits of RDNA in a 25 percent IPC gain over the latest version of GCN, and an effective 50 percent performance gain for the GPU when taking into account IPC, the 7 nm process, and gains from the frequency and power management (ability to sustain boost frequencies better).

Elsewhere on the silicon, AMD updated the Display Engine and Multimedia Engine to keep up with the latest display and video standards. The Display Engine now supports DSC 1.2a (display stream compression) along with output standards HDMI 2.0 and DisplayPort 1.4 HDR to support display formats as bandwidth-intensive as 4K 240 Hz or 8K 60 Hz over a single cable, and support for 30 bits per pixel color depth. The multimedia engine supports VP9 and H.265 decoding at up to 8K 24 Hz, or 4K 90 Hz, and hardware-accelerated H.265 encoding at up to 4K 60 Hz.

Features: FidelityFX and Anti-Lag

With each new graphics architecture, gamers expect new image quality enhancement features. NVIDIA introduced DLSS, and AMD's response to that is FidelityFX, a combination of content-specific and image-specific quality enhancements. The first part of this is contrast-adaptive sharpening, which brings out details in a scene by enhancing their contrast. To work best, it requires game developers to declare which parts of the image are to be sharpened (like the HUD and on-screen texts). Details such as wear-lines on the slick tires of a race-car or hexagonal patterns on a wall come to life. We will test this feature later in a separate article.

AMD wants to improve its adoption by professional e-Sports gamers by addressing a key bottleneck with modern high-end graphics: mouse lag. This would be the amount of time taken for a click to register and a response to be rendered by the GPU. Radeon Anti-Lag is a CTR (click-to-response) enhancement that reduced mouse lag by roughly a third across various popular e-Sports titles. This setting is effectively identical to "pre-rendered" frames on NVIDIA. Modern GPUs calculate one or two frames ahead, so they can better time sending them to the monitor to avoid stuttering. Of course, this results in input lag because any input information that comes in only makes it to the screen one or two frames later.

Packaging and Contents

Package Front
Package Back

You will receive:
  • Graphics card
  • Documentation
  • Driver disc
  • Red Devil sticker
  • PowerColor RGB mousepad

The Card

Graphics Card Front
Graphics Card Back

PowerColor has engineered a large cooler that uses different shades of gray, and the metal backplate has an RGB illuminated Red Devil logo and uses black as dominant color. Dimensions of the card are 30.0 x 13.5 cm.

Graphics Card Height

Installation requires three slots.

Monitor Outputs, Display Connectors

Display connectivity options include three standard DisplayPort 1.4a and an HDMI 2.0b.

AMD took the opportunity to update the display controllers handling these outputs by leveraging DSC 1.2a (display stream compression), which unlocks very high resolution and refresh-rate combinations over a single cable. Among the single-cable display modes supported are 8K 60 Hz (which took two DP 1.3 cables until now), 4K 240 Hz, and 1080p as high as 360 Hz. On top of these, the outputs support HDR and 30 bpc color-depth for better color accuracy in creative applications.

Graphics Card Power Plugs

The board uses two 8-pin power connectors. This input configuration is specified for up to 375 watts of power draw.

Multi-GPU Area

AMD's Navi generation of GPUs no longer supports CrossFire. DirectX 12 does include its own set of multi-GPU capabilities, but implementation requires game developers to put serious development time into a feature only a tiny fraction of their customers might ever use.

In this area, you'll also find a dual-BIOS switch, with the default setting being "OC" and the other "Silent". It not only runs a quieter fan curve, but also comes with slightly lower clocks and voltages, which helps the card stay cool at even reduced fan speeds.

Our Patreon Silver Supporters can read articles in single-page format.
Discuss(40 Comments)