In January of 2012 AMD launched the Radeon HD 7970, their first GPU powered by the GCN family of architectures. Since then GCN has stuck with us through the introduction of Mantle and later low-level APIs, FreeSync, rapid-packed math and new delta colour compression algorithms, as well as the shrink from 28nm to 7nm lithographic processes. Today, and after 7 years, AMD release hardware based on a new approach, one that’s set to take AMD's mainstream graphics lineup through the 7nm era and beyond: RDNA.
Read the AMD Radeon RX 5700-series Launch Review Here!
The RDNA family of architectures debuts with the Radeon RX 5700 and RX 5700 XT, two new graphics cards that incorporate Navi - the first GPU with RDNA inside. Sitting in the enthusiast bracket, the new cards offer substantially better performance and perf. per watt over AMD’s previous designs in the same segment, and are billed to be meaningful alternatives to NVIDIA's RTX 20-series family of cards.
Pricing and relative performance have been discussed since their unveiling at E3 2019 last month, but the RTX 20 Super-series' recent launch has prompted a readjustment. The revised pricing see the Radeon RX 5700 XT come in at an MSRP of $399, while the RX 5700 is a keen $349. These reflect the dynamic competitive landscape of GPUs in the <$500 market, and likely inform AMD’s expectations versus the new cards that are set to go on sale later this week.
The Radeon RX 5700-series are AMD’s workhorse GPUs for the new generation. They’re pitched to be excellent enthusiast options for 1440p gaming with high fidelity image quality – a new baseline which it’s felt better represents gamer expectations as we approach a new decade. In this respect they're replacements for the again RX Vega56 and 64. Naturally, they incorporate many legacy features that are still important in this landscape - notably FreeSync 2 HDR and native support for DirectX 12 and Vulkan – while building on other aspects just beginning to be leveraged such as Rapid Packed Math. And of course they’re not short of a few of their own tricks.
RDNA – A New Approach For Upcoming GPU Generations
Radeon DNA (RDNA) is what’s known as a Fundamental Architecture in AMD’s parlance. Just as GCN was tweaked over the years but retained much of its underlying structure and design ethos, so RDNA is expected to be the basis of AMD GPUs over multiple generations, with modifications where appropriate. Implicit in that AMD need to get it right; a misstep here could have major ramifications for years to come.
RDNA takes forward many of the principles of Zen, without (as yet) transitioning to a chiplet design. It focuses on improving GPU efficiency in both perf/watt and per/chip area, and as a result the first chip released is the 'Navi' (inspired by Gamma Cassiopeiae's alternate name). A Navi GPU is at the heart of every RX 5700 and 5700 XT.
RDNA incorporates multiple changes to the GPUs 'Compute Unit' building-blocks. Each CU incorporates 64 shaders – the same as Vega – but now two Compute Units work in tandem as a Work Group Processor, sharing specific resources including many cache structures. It’s speculated that this WGP could lay the foundation for new hardware-level features in the future, but AMD remain quiet on this score.
Just as the CU has been readdressed, so have many aspects of the GPU cache. Bandwidth between ALU and L0 cache has been reduced and a new L1 cache hierarchy has been implemented. As a whole, the changes significantly improve effective bandwidth and reduce latency, two aspects which serve to bolster ‘IPC’ (or rather its GPU equivalent).
Further improving effective bandwidth is the use of Delta Colour Compression (DCC) throughout the chip rather than isolating it to textures stored in memory. Plus, the DCC algorithm utilised has is now more efficient. DCC has been used for some time within AMD GPUs, but only in recent generations have they highlighted just how integral it’s becoming to their GPU architecture.
Other Important Developments
Navi is manufactured with TSMC’s 7nm process, similar to AMD’s Zen 2 CPUs and the Radeon VII GPU. As a result the GPU is significantly more compact and displays better perf/watt metrics than the those it ostensibly replaces in the AMD performance strata: the Vega56 and Vega 64. If the underlying architecture is scalable beyond 40 Compute Units then 7nm or developments thereof will be critical in keeping TDP under control.
Use of a 7nm node isn’t the only change; the RX 5700-series marks AMD’s first use of GDDR6 memory. Debuting on NVIDIA’s GPU lineup last year, GDDR6 memory chips offer both better bandwidth (now up to 448GB/sec) and perf/watt (60% lower) than GDDR5. The former is particularly important for high resolution and IQ gaming, and in eschewing HBM shares the hallmarks of Polaris AKA RX 500-series principles.
The Radeon RX 5700-series is also the World’s first GPU family to support the brand new PCI-Express 4.0 standard. PCIe 4.0 doubles the per-lane bandwidth available to supported peripherals to 2GB/sec, up to 32GB/sec for discrete x16 lane graphics. Complementing the standard’s inclusion are the Zen 2 processors and X570 motherboards that are releasing concurrently, but it's something of a forward-looking aspect of the card rather than one which will have many immediate applications.
Radeon RX 5700-series Technical Specifications
Cards released today adhere to the reference design not only in GPU specifications but also PCB layout and single-slot ‘blower’ cooling. The AMD reference blower cooler has seem some improvements compared to prior reference designs, including airflow and overall aesthetics, and these models will be the exclusive ones sold during the RX 5700-series launch window.
With this launch AMD are marketing their new series of cards with three different listed clock speeds: Base Clock, Game Clock and Boost Clock.
Base Clock: The minimum operating frequency of the GPU under load. All graphics cards sold should be capable of holding this frequency long-term under a variety of mixed by heavy workloads.
Game Clock: Typical clock speed when gaming. This will vary based on certain conditions, but should be a good comparative guide when partner AIBs are released in the near future.
Boost Clock: Maximum clock speed under light loads. Not particularly representative of gaming performance, and with vary between units depending on the ‘silicon lottery’.
Game Clock: Typical clock speed when gaming. This will vary based on certain conditions, but should be a good comparative guide when partner AIBs are released in the near future.
Boost Clock: Maximum clock speed under light loads. Not particularly representative of gaming performance, and with vary between units depending on the ‘silicon lottery’.
In isolation none of these figures are particularly helpful except in reference to deeper benchmarking and potentially troubleshooting. Comparative figures for Game Clock may be useful when judging out-the-box performance of different partner AIBs that utilise their own cooling and factory overclock. Nonetheless, with dynamic overclocking increasingly the norm for both GPUs and CPUs from each manufacturer, definitions of fundamental operating specifications are now being stretched to the limit in terms of validity and usefulness to consumers.
Additional Card Features
We've discussed some of the new software features exclusively supported by the RX 5700-series at launch elsewhere. Each are enabled in the latest Radeon Software Adrenalin Edition driver where appropriate, and where possible AMD are planning to extend support to older Radeon hardware through later drive releases. Some however utilise new capabilities unlocked by RDNA.
Radeon Image Sharpening (RIS) – Restores clarity to in-game images that have been softened by other post-process effects. An alternative approach to NVIDIA DLSS upscaling/sharpening, it doesn't leverage machine learning nor enforce particular resolutions or aspect ratios. More flexible.
FidelityFX – An open-source developer toolkit available in the coming weeks on GPUOpen that makes it easier for developers to create high-quality post-processing effects with minimal performance penalties. FidelityFX features Contrast-Adaptive Sharpening (CAS) that draws out detail in low-contrast areas while minimizing artefacts caused by typical image sharpening routines.
Radeon Anti-Lag – Radeon Anti-Lag decreases input-to-display response times by up to 31 percent, delivering an experience similar to higher framerates. This tool is particularly useful for esports and other fast-paced competitive gaming environments.
Extreme refresh rates – RDNA enables DisplayPort 1.4 with Display Stream Compression (DSC) to deliver extreme refresh rates, color depth and resolutions up to 8K HDR at 60Hz or 4K HDR at 144+ Hz without chroma-subsampling on cutting edge display
FidelityFX – An open-source developer toolkit available in the coming weeks on GPUOpen that makes it easier for developers to create high-quality post-processing effects with minimal performance penalties. FidelityFX features Contrast-Adaptive Sharpening (CAS) that draws out detail in low-contrast areas while minimizing artefacts caused by typical image sharpening routines.
Radeon Anti-Lag – Radeon Anti-Lag decreases input-to-display response times by up to 31 percent, delivering an experience similar to higher framerates. This tool is particularly useful for esports and other fast-paced competitive gaming environments.
Extreme refresh rates – RDNA enables DisplayPort 1.4 with Display Stream Compression (DSC) to deliver extreme refresh rates, color depth and resolutions up to 8K HDR at 60Hz or 4K HDR at 144+ Hz without chroma-subsampling on cutting edge display
The obvious feature not included in the new series of cards is real-time raytracing. AMD explained that they feel the technology ecosystem and mixed game support doesn’t warrant the increased cost implementing support for hybrid real-time raytracing would entail, including knock-on cost to the consumer. Future RDNA-based GPUs will support the technology when it’s ready (and console partners have already announced ray-tracing features in upcoming products scheduled for 2020), but for now just rasterization and other conventional techniques will be available with the cards. AMD also stated that they have no current plans to roll out a driver to support Microsoft DXR ray-tracing on new or legacy hardware, despite demoing functionality earlier this year.
Concluding Remarks
So that should give you a good top-level overview of AMD’s new RX 5700-series graphics cards – how they are likely to influence AMD’s GPU designs going forward and where they’ll fit within the current market landscape. It’s noteworthy that AMD are not explicitly retiring other GPUs with this launch as the RX 500-series and RX Vega-series will continue to be sold as entry-level through to the <$300/£300 markets. The Radeon VII, based on a 7nm die shrink of Vega, will remain AMD’s flagship model for a little while longer too.
Speaking of pricing, the RX 5700-series will be available from the following price points:
Radeon RX 5700 XT
US MSRP: $399
UK MSRP: £379.99 (inc. V.A.T.)
AUS MSRP: AUS$629
Radeon RX 5700
US MSRP: $349
UK MSRP: £339.99 (inc. V.A.T.)
AUS MSRP: AUS$549
US MSRP: $399
UK MSRP: £379.99 (inc. V.A.T.)
AUS MSRP: AUS$629
Radeon RX 5700
US MSRP: $349
UK MSRP: £339.99 (inc. V.A.T.)
AUS MSRP: AUS$549
The listed MSRPs for the cards are nigh-on identical to the MSRP of NVIDIA’s RTX 2060 and RTX 2060 SUPER, but actual street pricing will of course vary. Realistically, if they seek to compete with NVIDIA’s lineup but not offer real-time ray tracing, they need to bring two things to the table: better rasterization performance, and good cooling. AMD’s pre-launch material does indicate that performance is more than up to par, but only reviews will offer a complete picture.