Is the RX 5700XT really 40 percent slower than the 2070 Super?


    Reviews for the 2070 and 2060 Super have started arriving. It’s safe to say that these are some of Nvidia’s most competitive enthusiast parts since Maxwell. The 2070 Super offers RTX 2080 level performance at a sub-$500 price-point, while the 2060 Super one-ups the RTX 2070 at a substantial discount. The only potential spanner in the works is the arrival next week of AMD’s RX 5700 series. Just how fast is Navi 10? And is the cut-down TU-104 die at the heart of the 2070 Super really enough to play spoiler to AMD’s next-gen architecture?

    The past few weeks, leaked benchmarks of both Navi and the Super series made their rounds of the internet, well before the July 2nd Super soft launch. We now have a good idea of where RTX 2070 Super and RTX 2060 Super stand in terms of performance. The 2070 is generally 1-3 frames per second slower than the 1080 Ti—easily made up with a mild overclock—while the RTX 2060 Super delivers performance almost exactly on par with the original 2070, and slightly better than the vanilla 1080.

    In terms of raw performance in the enthusiast segment—and not added raytracing chops—this is what many had expected Turing to deliver: Good 4K performance at affordable price points. What we don’t know right now is exactly how well the Navi parts—RX 5700XT and RX 5700 will perform, particularly in comparison to the new Super parts. However, looking at AMD’s official benchmarks and the whole range of Navi leaks, it is possible to read between the lines and get a bead on Navi’s real-world performance profile.

    AMD’s official benchmarks: Take with a pinch of salt?

    AMD released its own benchmarks during the initial Navi announcement. These pitched the RX 5700XT as a stand-in replacement for Vega 56. AMD’s benchmarks show the 5700XT handing in 1440p framerates comparable to Vega 56’s 1080p output. If this is true, it’d mean that Navi is a tremendous step up, but does the math pan out? Let’s see. At 2560×1440, you’re pushing 36,86,400 pixels onscreen, which is approximately 77 percent higher than 1080p. Now, this is a problem. Vega 56 is no slouch and, at 1440p, the only card on the market that’s 77 percent or more faster than it is the RTX 2080 Ti.

    There’s simply no way the RX 5700XT can deliver comparable performance to the 2070 while being 77 percent faster than Vega 56. How did AMD likely reach these results? Here’s what we think: The Vega 56, as we’d mentioned earlier, is a powerful piece of kit, averaging in between the 1070 and 1080 in most games. At a modest 1080p, there are plenty of games where a Vega 56-equipped PC will start hitting CPU bottlenecks. Moreover, many of the titles in AMD’s test suite, such as AC: Odyssey, are notorious for being CPU-intensive. Even the 2080 Ti can’t deliver minimums above 60 FPS, not because the card’s incapable of doing so but because of CPU-side limitations. At 1080p, a hypothetical RX 5700XT that’s about 30 percent faster than Vega 56 (roughly in line with the 2070), will deliver comparable performance at 1080p as both cards brush against CPU bottlenecks. But while Vega 56’s performance at 1440p will halve, the 5700XT will incur a much lesser penalty from the bump in resolution. This would mean that the 5700XT at 1440p performs comparably to Vega 56 at 1080p, even if the former is only 25-30 percent more powerful.

    AMD’s other benchmarks show the 5700XT posting single-digit leads against the RTX 2070, though this widens to a spectacular 22 percent in Battlefield V. Dice titles have historically run better on AMD hardware, so it’s comfortable to assume that this is an outlier. What’s more interesting are AMD’s results for Metro Exodus and AC: Odyssey. These are both titles that heavily favor Nvidia GPU, to the extent that the old flagship R9 Fury X runs Exodus worse than a 6GB 1060. AMD’s benchmarks show the RX 5700XT posting wins against the RTX 2070 in both titles, with a 2 percent margin in Odyssey and a much wider 15 percent margin in Exodus. While it’s always important to take vendor benchmarks with a pinch of salt, it’s hard to reconcile these results with any benchmark shenanigans on AMD’s side, short of their dropping quality settings on the 5700XT (which’d probably open them up to a lawsuit in certain jurisdictions). The only conclusion we can draw here is that the 5700XT does indeed perform comparably to the RTX 2070, even in Nvidia-friendly games, meaning that a 10-15 percent advantage in other titles isn’t beyond the realm of possibility.

    What do the synthetics say?

    Synthetic benchmarks like 3DMarks aren’t a great predictor of real-world GPU performance. Case in point: My overclocked RX 580 hands in comparable 3DMark scores to my old Fury, but the Fury performed comprehensively better in games, so long as VRAM wasn’t a limiting factor. Nevertheless, synthetics do have their uses. They are a good predictor of ballpark performance: Two GPUs with similar 3DMark scores will generally perform within a few FPS of each other in most games. 3DMark scores for the 5700XT put it in the same ballpark as the RTX 2070 and this is crucial in understanding where the 5700XT is positioned. It lends credence to AMD’s claim that the card is a 2070-killer. Looking at the 3DMark scores in the context of the benchmark leaks, it’s likely that the 5700XT will deliver performance somewhere between the 2070 and 2070 Super.

    What about the FFXV bench?

    Ah, the infamous Final Fantasy XV benchmark leak. The clickbait byline of the day last week was “RX 5700XT is 40 percent slower than the RTX 2070.” Final Fantasy XV has a fairly comprehensive benchmark and it is true that leaked figures for the 5700XT showed it performing significantly below the 2070, and actually within the 2060’s performance envelope.

    But there’s a major caveat here: Final Fantasy XV is a notoriously Nvidia-friendly title. FFXV plays host to a range of exclusive GameWorks features, and runs poorly on, well just about everything AMD. But enabling GameWorks makes the framerate tank hard, especially on red hardware. The leaked FFXV benchmarks were at 1080p with the High present enabled. Ordinarily, you’d think that High would only enable standard features but in the FFXV benchmark, High enables a whole raft of framerate-crushing GameWorks functionality.

    Considering that GameWorks is a black-box, proprietary solution, enabling it incurs only a reasonable penalty on Nvidia hardware. But, with existing AMD cards, GameWorks in FFXV cuts framerates in half. An RX 580 averaging near 60 FPS with Gameworks disabled will see framerates tumble to the 30 FPS range, with minimums in the 20s. Nvidia cards only incur a ~25 percent penalty. With GameWorks on, the Vega 64 performs comparably to the GTX 1060 two tiers down. If the RX 5700XT is slightly faster than the RTX 2070, it is entirely feasible that it’d trail the latter by 40 percent in the FFXV bench, simply because of the impact of GameWorks.

    Overclocking: Can it cover the gap?

    So far, benchmarks indicate that the 5700XT will perform somewhere in the vicinity of the RTX 2070, likely a bit faster than it, but notably slower than the 2070 Super. We’re talking about stock speeds, though. Ever since Maxwell gave gamers an almost ludicrous amount of overclocking headroom (remember those 1500 MHz 980 Tis?), Nvidia’s made it a point to limit user-controlled overclocking, while placing a greater emphasis on its GPU Boost tech. In practice, Pascal and Turing parts offer as little as 5-8 percent of additional OC headroom. AMD’s well aware that the price-performance ratio is where it needs to be competitive.

    All modern AMD GPUs are overclocking-friendly, with almost all functionality being user-tweakable, from core and memory voltages to power limits. At launch, Vega 56 was seen as a disappointment because it only offered performance in line with the GTX 1070. But as many Vega 56 owners know, unlocking the power limit and tweaking voltages has a massive impact on performance, to the point where a tweaked Vega 56 is competitive with the GTX 1080. We fully expect AMD to leave Navi’s overclocking functionality wide open. Considering that it’s built on the 7nm process, it’s entirely possible for Navi to hit 2.1 GHz at reasonable voltages. How it’ll perform then, relative to the 2070 Super, remains to be seen.

    Navi’s just a few days away from soft launching. Nevertheless, the Nvidia Super series has managed to steal a lot of its thunder. Leaks are one thing: What we’re looking forward to is putting the 5700XT through its paces in our benchmarking suite and finding out exactly how it compares to Nvidia’s enthusiast parts.

    If you found this story interesting, follow us on Facebook and Instagram.

    Further Reading:

    Leave a Reply