AMD Radeon RX 5700 XT and Radeon RX 5700 Review: New Prices Keep Navi In The Game

[harlan4096]

Our Verdict

Radeon RX 5700 XT stands up better to GeForce RTX 2060 Super than we expected at the same $400 price point, simultaneously eclipsing the performance of Radeon RX Vega 64 in the process. AMD’s own Radeon VII is slightly faster, but also significantly more expensive. Although a lack of hardware for accelerating ray tracing does hurt the Radeon’s value proposition in our opinion, gamers who insist they aren’t interested in the technology have a good alternative for smooth 1440p frame rates.

For

Almost 10% faster than GeForce RTX 2060 Super, on average, through our benchmark suite
Unique card design built with high-quality materials
Much better acoustics than past AMD reference cards

Against

No ray tracing acceleration
Higher power consumption than GeForce RTX 2060 Super
Uncomfortably hot operating temperatures (Tj > 100°C)
Protection mechanisms intervene under certain synthetic workloads

AMD Radeon RX 5700 XT and Radeon RX 5700 Review

11/21/2019 Update: Since the initial launch of of AMD's initial Navi cards in mid-July, we've tested some higher-clocked third-party variants, like the Sapphire Nitro+ RX 5700XT that deliver better out-of-the-box performance and colorful aesthetics.

AMD’s propensity for slowly dribbling out information about upcoming products keeps our news desk buzzing but also serves to illustrate the company’s play book months in advance. That drawn-out tease definitely worked against AMD last generation. By the time Radeon RX Vega landed in our lab, expectations had boiled over beyond what the card could deliver, particularly at its cryptocurrency-affected prices.

This time around, AMD used the Electronic Entertainment Expo in Los Angeles as the launch pad for a Navi deets, pouring out all the architectural details it was willing to divulge with less than 24 hours to write it up, then complicating matters by prohibiting audio or video recordings of the technical deep dives. But even rushing the particulars for reasons unknown couldn’t stop Nvidia from squeezing in a Turing refresh prior to Radeon RX 5700 and Radeon RX 5700 XT availability.

The two Navi-based cards originally took aim at GeForce RTX 2070 and GeForce RTX 2060. Just days before reviews were scheduled to go live, however, AMD found itself staring down the barrel of GeForce RTX 2060, 2060 Super, and 2070 Super. Its hardware was already baked, so the company turned another dial to stay competitive: it dropped the price of Radeon RX 5700 XT to $400, matching GeForce RTX 2060 Super, and lowered Radeon RX 5700 to $350, pulling up alongside GeForce RTX 2060. AMD is clearly feeling good enough about its performance story to go up against the GeForces at identical pricing. Are either of these Radeons among the best graphics cards or is AMD underestimating the appeal of real-time ray tracing support?

Meet Radeon RX 5700 XT

Both AMD Radeon RX 5700-series cards are based on the same Navi GPU. Manufactured on TSMC’s 7nm FinFET process and composed of 10.3 billion transistors, these chips occupy a scant 251 mm². Vega was much larger. Manufactured on GlobalFoundries’ 14nm LPP process, it packed 12.5 billion transistors into a 495 mm² die. For some additional context, Nvidia’s competing GeForce RTX 2060-series cards employ TU106, a 10.8-billion-transistor chip measuring 445 mm² and built using TSMC’s 12nm FinFET process.

We confirmed with AMD that Radeon RX 5700 XT employs a fully-enabled version of the Navi GPU—no part of the chip is turned off to improve yields or leave room for a more resource-rich model in the future. It exposes 40 RDNA Compute Units, each with 64 Stream processors, totaling 2,560 ALUs across the processor. The CUs host four texture units, just as they did in AMD’s Graphics Core Next design, adding up to 160 in a complete Navi GPU. Four render back-ends per quadrant are capable of 16 pixels per clock cycle, yielding 64 ROPs.

That’s clearly a more compact configuration than Radeon RX Vega 64, which featured 64 CUs with 4,096 Stream processors and 256 texture units. And yet our benchmarks will show that Radeon RX 5700 XT averages 15%-higher frame rates than Vega 64. Almost 60% of the architecture’s speed-up comes from performance per clock enhancements, according to AMD. Another 25% is attributable to gains enabled by 7nm manufacturing. The reminder falls under design frequency and power improvement, which includes more effective clock gating.

The specifications for Radeon RX 5700 XT curiously define its base clock rate as “up to 1,605 MHz.” At first, we didn’t think anything of this. After subjecting the 5700 XT to synthetic workloads like FurMark, however, and observing frequencies as low as 1,575 MHz, it appears that the base can be violated under the right (or wrong) conditions, favoring a consistent acoustic experience over strict performance boundaries. AMD also specifies a Game GPU clock of “up to 1,755 MHz” and a Boost GPU clock of “up to 1,905 MHz.” As you might guess, both ratings are subject to certain conditions. In fact, we saw Boost frequencies well above 1,905 MHz at the start of many games. As the card warms up, though, expect to see clock rates closer to the Game GPU clock.

Let’s just throw this out there: We’d prefer that AMD not create a third frequency rating. Because it is fleeting, it’s subject to abuse. In fact, AMD is already using that peak figure to calculate its 9.75 TFLOPS FP32 performance figure. The more sustainable 1,755 MHz Game GPU clock translates to 9 TFLOPS, and that just doesn’t look as thunderous next to GeForce RTX 2060 Super’s 7.2 TFLOPS, right? Navi does carry over support for rapid-packed math, so AMD cites half-precision performance of up to 19.5 TFLOPS.

An aggregate 256-bit pathway is populated by 8GB of GDDR6 operating at 14 Gb/s. This gives Radeon RX 5700 XT up to 448 GBps of memory bandwidth—slightly less than Radeon RX Vega 64’s 484 GBps but significantly more than Radeon RX 590’s 256 GBps. AMD claims other notable improvements throughout Navi’s memory hierarchy, from reduced congestion in its 4MB L2 cache to a new 128KB L1 cache per quadrant that helps reduce latency.
...

Continue Reading