The Radeon RX 5700 and Radeon RX 5700 XT graphics cards represent a fresh start and a bright future for AMD, brimming with technologies that have never been seen in GPUs before.
The Radeon RX 5700 series are the first mainstream GPUs built using the bleeding-edge 7nm manufacturing process (hence their July 7 launch). They’re the first graphics cards packing the ultra-fast PCIe 4.0 interface. They’re the first graphics cards crafted with AMD’s all-new “RDNA” architecture, which delivers a massive power efficiency boost that Radeon’s been needing for years. They’re packing fresh display technologies to enable 4K, 144Hz monitors without the need for messy chroma subsampling. They’ve upgraded to GDDR6 memory. Heck, AMD even managed to tame their blower-style cooler design, learning from the mistakes of the frankly unpleasant-to-be-around Vega 64.
Nvidia tried to spoil their launch with the surprise release of the excellent GeForce RTX 2060 Super and RTX 2070 Super, but some aggressive pre-launch price cuts by AMD foiled that plan. The $350 Radeon RX 5700 and $400 Radeon RX 5700 XT are great graphics cards, full stop.
Radeon RX 5700 and 5700 XT: Specs and features
Warning: With so much new technology baked into the Radeon RX 5700 series, this is going to be a long review. We’re going to kick it off by talking about what’s fresh under the hood. Feel free to use the table of contents at left to hop between sections if you’d like. Got it? Good. On to the cool stuff.
Let’s kick things off by examining the hardware, before delving into some new software features later. Here’s an AMD-supplied list of technological specifications for the Radeon RX 5700 and Radeon RX 5700 XT, alongside the same stats for the Radeon RX Vega GPUs they’re replacing in AMD’s product stack. (Radeon boss Scott Herkelman confirmed that Vega will be disappearing in The Full Nerd interview embedded above.)
While that table provides a helpful overview, note that you cannot simply compare the number of compute units and stream processors between the Vega GPUs and the “Navi” GPUs in the Radeon RX 5700 series. Compared to AMD’s long-lasting GCN architecture (which Vega is based on), the new RDNA architecture introduces several radical changes to the underlying GPU design, unleashing significant hardware-level overhauls to everything from the cache to the graphics engine to the compute units themselves. These new graphics cards perform tasks differently from their predecessors at a fundamental level.
We’re not going to get into the weeds of redesigned cache hierarchies and SIMD Wave cycles here. If you want an insightful yet understandable explanation of the major RDNA architecture changes and have 23 minutes to spare, I highly recommend watching the Gamers Nexus interview with technology analyst David Kanter embedded below. It’s great.
AMD says that between the RDNA tweaks and the shift to the 7nm manufacturing process, its new cards are much more powerful and power-efficient than before—claims that bear fruit in our testing. (Spoiler alert: Navi is even more power-efficient than Nvidia’s Turing architecture, a monumental reversal of the norm with GCN.)
The company’s architecture deep-dive for reporters touted some impressive numbers for Navi: 25 percent more performance per clock versus Vega, and a whopping 50-percent performance increase overall at the same power levels. Impressive. Those performance-per-clock improvements also make the TFLOPS comparisons in the chart above misleading, as these Navi GPUs do more with each teraflop.
Speaking of clocks, the Radeon RX 5700 and 5700 XT hit much, much higher clock speeds than AMD’s previous cards, surpassing Vega’s best by hundreds of megahertz. No matter how you measure it, that’s a good thing. AMD also introduced new clock speed terminology to more clearly express estimated performance levels.
The company’s providing three clock speeds for Navi GPUs: Base and boost are what you’d expect, defining the upper and lower limits. But AMD’s also listing a “game clock,” that represents “the minimum expected GPU clock when running gaming applications.” It warns that those speeds aren’t guaranteed or even set in the VBIOS, but is instead simply “a guide used to set expectations with gamers.” Nvidia handles its listed Boost clocks in the same manner, so if you want to compare GeForce cards against the Radeon RX 5700’s speeds, look there.
Note that it says “minimum expected GPU clock,” and that they were obtained using AMD’s blower-style reference cards. In the Full Nerd interview at the top of this article, Radeon boss Scott Herkelman acknowledged that overclocked, custom-cooled models by AMD partners could very well push gaming speeds closer to the Boost clock rating, as you’d expect.
The Radeon RX 5700 and RX 5700 XT also change the underlying memory configuration. The Radeon VII and Vega GPUs utilized second-generation high-bandwidth memory stacks, continuing the HBM push that AMD spearheaded with the Fury X in 2015. AMD switched to traditional GDDR6 memory for these new cards, matching Nvidia’s current GeForce lineup.
For Nvidia, the switch from GDDR5 to GDDR6 provided enormous throughput benefit thanks to GDDR6’s much faster speeds. It’s more of a lateral move performance-wise for AMD, because HBM2’s 2,048-bit wide memory interface allowed for excellent memory bandwidth speeds. The Navi cards we’re reviewing today hit 448GB/second of memory bandwidth thanks to their 8GB capacity and 256-bit bus.
“We are not ruling out HBM2 in our roadmap,” Herkelman told me in our Full Nerd interview. “We will continue to support that in markets that make sense.” For these Radeon RX 5700 cards in particular, GDDR6 lets AMD “hit better price points [and] still get great performance” while being more widely adopted than HBM2, whose pricing he called “volatile.” The power-hungry Vega cards needed HBM2’s energy efficiency, but AMD says GDDR6 offers 60-percent-improved performance per watt versus GDDR5.
Faster, more efficient clocks and faster, more efficient memory both provide tangible benefits to PC gamers. One of the flagship features of the Radeon RX 5700 series, PCIe 4.0 support, doesn’t—at least in normal gaming scenarios. Modern graphics cards simply don’t saturate the PCIe 3.0 x16 interface typically used in most gaming PCs. I’d wondered whether the move to PCIe 4.0 could benefit performance in CrossFire setups, because those need to tap into a narrower x8 connection. Herkelman said any benefits would be “corner case, game-dependent, and situational.” He said he wouldn’t recommend people buy into PCIe 4.0 just for potential improvements to multi-GPU gaming.
But there are advantages to having the Radeon RX 5700 and 5700XT ride the bleeding-edge PCIe 4.0 interface. Herkelman says a large percentage of gaming GPU buyers also run content creation tasks, which makes sense with Twitch and YouTube engulfing the world. “There is much more of a benefit in creative workloads today for that new standard,” Herkelman told PCWorld.
I wasn’t able to validate the claim, as I don’t have a PCIe 4.0-compatible system. To tap into the Radeon RX 5700 series’ PCIe 4.0 capabilities, you’ll need to buy a Ryzen 3000 series processor and an X570 motherboard, both of which are also launching Sunday. AMD’s hardware barrage marks the debut of PCIe 4.0 in PCs, while Nvidia and Intel’s hardware still uses PCIe 3.0. If you don’t have a Ryzen 3000 CPU, fear not, as PCIe 4.0 hardware is backward-compatible with PCIe 3.0. The Radeon RX 5700 GPUs will work in any PC.
Both the Radeon RX 5700 and RX 5700 XT return to blower-style cooler designs, even as Nvidia’s Founders Edition cards shift to dual-axial fans. Herkelman says it was a conscious decision: Blower-style coolers expel hot air out of the rear of your system, while dual-axial fans dump it back into your rig, requiring your PC to have adequate air flow from case fans. Sticking to a blower-style design eliminates potential user error from the cooling equation and provides a more universal baseline experience.
The decision makes sense on paper, but anyone who suffered through a reference Vega 64 roaring hot fire next to them might be cringing right now. It was unpleasant. The new cooler isn’t. Herkelman claims AMD invested a lot of time and effort into improving the Radeon RX 5700 series reference coolers, and those tweaks manifest in practice. These coolers still aren’t quiet, but neither have they crossed the line in being noisy or distracting. While the reference Navi GPUs run at higher temperatures than Nvidia’s rival Founders Edition models, the heat doesn’t affect performance or usability. (More on that in the benchmarking sections.)
AMD’s new GPUs feature different shroud designs. The Radeon RX 5700 looks like a sturdier, cleaner upgrade to the stark red-on-black design ethos introduced with AMD’s Radeon RX 400-series, while the Radeon RX 5700 XT changes things up with a series of closely spaced lines that run the length of the black shroud, and a red band that intersects with the illuminated Radeon logo on the side of the card. It also packs an intriguing wave-like deformation on the edge of the card. It’ll probably prove divisive, but I don’t mind it. Both cards require 6-pin and 8-pin power connectors, with the Radeon RX 5700 demanding 185 watts of total board power and the Radeon RX 5700 XT needing 225W.
The Radeon RX 5700 XT ships with a backplate; the Radeon RX 5700 does not. The RX 5700 XT includes additional venting at the rear edge of the card, while the cheaper Navi GPU gets a solid back edge.
The cards ship with an HDMI 2.0b port and a trio of DisplayPort 1.4 connections. The DisplayPorts support Display Stream Compression 1.2a, which lets the GPU power fearsome monitors over a single cord, maxing out at 4K/240Hz, 4K HDR at 120Hz, and 8K HDR at 60Hz. More significantly, Display Stream Compression lets you run those high-speed 4K HDR monitors without leaning on chroma subsampling, which can affect the readability of fine text in some scenarios. Asus teased the “world’s first Display Stream Compression monitor” at AMD’s E3 event. With 4K, 144Hz speeds and HDR 1000 support, it appears poised to challenge Nvidia’s vaunted G-Sync Ultimate (previously G-Sync HDR) monitors, like the superb Acer Predator X27.
Finally, the Radeon RX 5700 series introduces a new version of the Radeon Media Engine, with better hardware accelerated support for VP9 decoding, and both encoding and decoding with H.264 or H.265.
You’ll notice two glaring omissions from this rundown of upgrades: support for ray tracing and variable rate pixel shading, both of which were introduced in Nvidia’s GeForce RTX 20-series, and both of which were embraced (agnostically) by Windows 10’s DirectX 12. You can hear Herkelman explain why AMD made the decision to omit dedicated ray tracing hardware from the Radeon RX 5700 series at the 39:20 mark in the video interview near the top of this article. Quick tl;dr: AMD will enable ray tracing when there are more games available that support it, and when the performance hit becomes less drastic. “Don’t get me wrong: It is definitely the future,” he says. “But it needs to be done right.”
That’s it for the hardware side of things. But AMD’s latest graphics cards pack some new software tricks, too.
Next page: Radeon Software improvements