Topps Blog: AMD Radeon RX 5600 XT Review: Testing With Third-Gen Ryzen

AMD is launching its new RTX 5600 XT today ($279) and we’ve decided to examine the GPU in what’s become a rather unusual method for primary hardware reviews: We’re going to examine AMD’s performance on an AMD X570 motherboard and compare against Nvidia cards on that basis.

Ryzen has been a solid CPU for gaming ever since it launched; our 1080 Ti review showed the gap between Broadwell-E and first-generation Ryzen was 1 percent at 4K (and 8 percent at 1080p). One of the significant places where third-generation Ryzen improved on the original microarchitecture was in gaming at lower resolutions. While Intel still has an edge in space, AMD’s rising desktop market share means more players are going to be wanting to see AMD-specific results.

I’m not planning to formally move my standard gaming testbed from Intel to AMD yet, because Intel platforms do maintain an advantage in this specific space. But in terms of investigating the idea of an all-AMD system at a reasonable CPU and GPU price point, this is a good time to compare on Ryzen Zen 3. After considering the options, we decided to test with a 3700X. It’s not AMD’s absolute fastest gaming CPU, but it matches well against a $279 graphics card.

We’re comparing three AIB cards today rather than reference designs. Our RX 5600 XT is a Sapphire Pulse with an updated VBIOS for factory overclocking as previously discussed. It’s going up against the EVGA RTX 2060 KO and the GTX 1660 Super SC Ultra.

Here’s the Sapphire Pulse. The card has a single eight-pin power connector and feels well-built in the hand. It’s also very quiet relative to the 1660 Super or the 2060 KO. I don’t have a decibel meter to formally test, but the 5600 XT completely misses the “Audible enough to be annoying, but not so audible as to call it <em>loud</em>” trap that GPUs sometimes fall into.

The RTX 2060 KO. (06G-P4-2066-KR). Provided by EVGA, it shares a common cooler design with the 1660 Super SC Ultra.

Finally, the GeForce GTX 1660 Super SC Ultra (06G-P4-1068-KR). Again, common cooler design. Neither of these cards are currently in-stock on EVGA’s website, but hopefully that’ll change in the very near future.

Meet the RX 5700 RX 5600 XT?

The GPU AMD is launching today is not, strictly speaking, the GPU it intended to launch. The original version of the Radeon 5600 XT had a game clock of 1375MHz and a boost clock of 1560MHz, with a 150W TDP. The new version of the GPU has a game clock of 1615MHz and a boost clock of 1750MHz. That works out to an increase of 1.17x to game clock and 1.12x to boost. At the same time, AMD boosted the GDDR6 speed on the card from 12Gbps to 14Gbps, increasing memory bandwidth from 288GB/s to 336GB/s.

AMD took this step to counter Nvidia’s sudden price cut on the RTX 2060 last week. The launch price on the RX 5600 XT is set at $279, which made it a very attractive card next to the RTX 2060 at $350. Then Nvidia cut the price of the RTX 2060 to $299, wrecking AMD’s price/performance curve. AMD has responded by sharply increasing the clock speed on the RTX 5600 XT, for reasons that will become clear.

But bumping up the RTX 5600 XT’s clock has consequences of its own. As a result, the 5600 XT is virtually on top of the 5700, as shown in the chart below:

AMD hasn’t given us the base clock on the 5600 XT, but the game and boost clocks match or even slightly exceed the 5700’s. The sole advantage remaining to the higher-end card is its larger 8GB memory pool and wider 256-bit memory bus. There’s been some concern about the longevity of 6GB solutions considering we are on the cusp of another console generation, but the Nvidia solutions at this price point are equipped similarly.

On the other hand, AMD has previously gone after Nvidia for this kind of thing with the GTX 1060. When AMD wanted to pump the value proposition of the Polaris family (RX 400, RX 500), we heard a great deal about the long-term benefits of an 8GB frame buffer as opposed to a 6GB buffer. Now that AMD needs to create some meaningful product distinctions, however, we’re back down to 6GB from both companies.

Given how close the Radeon RX 5600 XT is to the Radeon RX 5700, you might think AMD would be preparing to kill off the 5700. According to AMD, you’d be wrong. Here’s AMD’s official word:

The Radeon RX 5600 XT is not replacing the Radeon RX 5700 graphics card. The two products address different market needs. The Radeon RX 5600 XT was specifically designed and optimized to deliver the best 1080p gaming experiences, while the Radeon RX 5700 – with its additional memory and memory bandwidth – was specifically designed for 1440p gaming.

The Radeon RX 5700 continues to enjoy significant demand from gamers, and we will continue to offer the card based on market demand.

On paper, the RX 5600 XT and RX 5700 should perform identically. We’ll see if they actually do in practice.

Test Setup

We tested on an MSI X570 Godlike motherboard with the 7C34v17 UEFI, dated from November 2019, with AGESA 1.0.0.4. Windows 1909 and AMD’s Core Boost driver were both used, with all Windows 10 updates and patches installed. A 1TB Corsair MP600 SSD provided storage. We tested the Nvidia 441.87 driver and a reference driver from AMD. Public drivers for the 5600 XT will be available by the time you read this.

Using AMD over Intel offered an additional opportunity to change GPU settings without disrupting previous data sets. I’ve relaxed the AA settings I typically test with in Rise of the Tomb Raider, Shadow of the Tomb Raider, and Deus Ex: Mankind Divided. Metro Last Light Redux is still tested with SSAA enabled.

The one downside to focusing on AMD instead of Intel for this testing is that we’re limited in terms of the number of comparative GPU results we can offer. This review contains results from four cards: GTX 1660 Super and RTX 2060 from Nvidia (both provided by EVGA in the form of the GTX 1660 Super SC Ultra and the RTX 2060 KO), the Radeon RX 5600 XT (Sapphire Pulse), and the Radeon RX 5700 (AMD reference).

Warhammer II was tested in DX11 mode on our Nvidia cards and in DX12 mode on AMD GPUs. While we normally test in the same API, Warhammer II shows very strong preferences for DX12 on AMD cards and DX11 on Nvidia. We recommend you use the respective APIs for each GPU to maximize performance.

Data is being uploaded for the review as we speak, so if you’re here just after 10 AM, give me a minute to finish uploading.

Ashes of the Singularity: Escalation

Ashes of the Singularity kicks things off, with a healthy win for the RX 5700 and a tie between the Radeon RX 5600 XT and the RTX 2060. The 1660 SC Ultra lags behind here, though its lower price makes this more acceptable than it would be otherwise.

Deus Ex: Mankind Divided

I typically test DXMD with 4x MSAA enabled and I’ve often remarked on how heavy a hit it imposed. I had no idea that turning it off would produce results like this on AMD GPUs, however:

Both of AMD’s GPUs win this test easily. The RTX 2060 and GTX 1660 take their worse loss of the review here.

Hitman

Hitman is another good showing for the 5600 XT. The RTX 2060 and 5600 XT keep pace with each other, while the 1660 Super is well behind. Up until now, the 5600 XT and RTX 2060 have matched each other — but that’s about to change.

Metro Last Light Redux

Even midrange cards are now powerful enough to render 1080p with SSAA enabled (effectively 4K) and to keep the frame rate above 60fps. The 5600 XT lands halfway between the 1660 Super and the EVGA RTX 2060 KO.

Shadow of War

In Shadow of War, the EVGA RTX 2060 KO outperforms the RX 5600 XT, which in turn outperforms the 1660 Super, though not by an enormous margin in either case. The gap between the 5600 XT and the 1660 Super actually shrinks a bit at 1440p compared with 1080p, which is the opposite of what we’d typically expect. The RX 5700 wins the test overall.

Rise of the Tomb Raider

In RotTR, the 5600 XT lags the RTX 2060 by 8 percent and 6 percent at 1080p and 1440p, while the RX 5700 still takes the overall benchmark.

Shadow of the Tomb Raider

SotTR shows roughly the same pattern as RotTR, with the 5600 XT edged out by the RTX 2060.

Despite AMD’s comments about how the RX 5600 XT is aimed at 1080p gaming, it’s really only fair to note that this GPU handles 1440p just fine, even if it’s a bit slower than the RTX 2060.

Warhammer II

AMD takes a hammering in Warhammer II, though not an unexpected one. If DXMD is our most AMD-friendly benchmark, Warhammer II is one of the most Nvidia-friendly test. Here, the RX 5600 XT is only a bit ahead of the EVGA GTX 1660 Super. It takes the RX 5700 to surpass the RTX 2060.

Far Cry 5

After the recent drubbing AMD has been taking, it’s nice to see a near-parity result again. The Sapphire Pulse RTX 5600 XT is running 3-4 percent slower than the RTX 2060, which is just on the outer edge of what we consider a tie (or margin of error).

Assassin’s Creed: Origins

ACO is almost as Nvidia-friendly as Warhammer II, so we’re not surprised by what we see here. The 1660 Super actually claims a win off the Sapphire Pulse RX 5600 XT — the first and only win it’ll see in this review — while the EVGA RTX 2060 is ahead by just shy of 1.2x. In this case, the RX 5700 brought very little additional to performance to bear over the RX 5600 XT, implying that GPU clock and core count, rather than memory bandwidth, may be the limiting factor for this game.

Metro Exodus

In Metro Exodus, the 5600 XT takes one last opportunity to snag a victory over the Radeon RTX 2060, and does so. Metro Exodus is a strong title for AMD overall, and while the GeForce cards turn in strong performance, they don’t quite match AMD.

The Sapphire Pulse takes the lead over the RTX 2060 and 1660 Super, while the Radeon RX 5700 breaks away from the rest of the pack.

Power Consumption

We tested load power using Metro Last Light Redux in 1080p with SSAA enabled. AMD has shown much better power efficiency with Navi as compared with GCN, and the results here are similar to what we’d expect:

The 5600 XT and RX 5700 both show slightly higher power efficiency than the GTX 1660 / RTX 2060, but the gap is small. At least some of AMD’s intended power savings on the 5600 XT went out the window when it shipped new high-performance UEFIs to increase overall performance against the RTX 2060, which undoubtedly impacts these results. EVGA’s RTX 2060 KO uses more power than the Sapphire Pulse. Both the 5700 and 5600 XT are more efficient (in terms of perf/W) than the Nvidia cards they compare against here, but only by a small amount.

Comparative Results

At the end of all the benchmarking, the 5600 XT winds up behind the RTX 2060, despite the clock improvements. The average frame rate on the RTX 2060 is roughly 5 percent faster than the 5600 XT based on the geometric mean of their respective test scores (85.0 for the Sapphire, 88.5 for the EVGA RTX 2060). The RTX 5600 XT, which appears quite close to the RX 5700 on paper, falls rather behind it in practical testing, which turns in a geomean score of 95fps. The 5700 is roughly 1.12x faster than the 5600 XT in 1080p and 1.14x faster in 1440p. It costs 25 percent more — I suspect the original price banding here was meant to put a linear price/performance curve between the 5600 XT and the 5700 — but jacking up the 5600 XT’s clock to match the EVGA RTX 2060 KO made the comparison less friendly to the 5700. Either way, AMD has a justification for keeping the RX 5600 XT in the mixture.

The data here cumulatively suggests that the RX 5600 XT is bandwidth-limited, not clock limited. It generally clocks faster than the Radeon RX 5700, but the 5700 is definitely the faster card between the two. Geometric mean was used instead of arithmetic mean to reduce the impact of large outliers.

Conclusion

When AMD launched Navi back in July 2019, it offered moderately faster performance than the GeForce cards it went up against but lacked features like ray tracing. Nvidia just gave the baseline GeForce RTX 2060 a 15 percent price cut, which improves the overall value proposition. AMD’s clear original goal was to hit the spot between RTX 2060 and GTX 1660 Super with a value proposition that would appeal to potential buyers of either card.

Given their respective present positions, the RTX 2060 is about 5 percent faster than the RX 5600 XT and roughly 7 percent more expensive. That’s not the worst position to be in, but it’s a weaker one than the 5700 and 5700 XT occupied at launch, when both AMD cards offered uplifts against their Nvidia counterparts as well as cheaper prices.

Here’s how I’m thinking about the 5600 XT overall. Objectively, it’s a great card. It’s performance-competitive with the RTX 2060, with small advantages in terms of price and power consumption. Nvidia’s Ampere is expected to launch later this year, but upper-midrange cards like the RTX 2060 often refresh several months behind their high-end counterparts. If I had to guess, I’d guess that Nvidia and AMD intend to slug it out in the upper tiers of the market this year, at least if they both get to pick.

The RTX 2060’s price cut make ray tracing intrinsically more attractive, but I wouldn’t necessarily buy solely on this feature. Take a look at the games that currently feature RT and the list of titles expected to do so in the near future, and weigh your decision accordingly. Now that we know that both consoles will feature at least some level of ray tracing, as will Navi 20, we can assume that the capability will start to be adopted more widely. Practically, however, that could still mean widespread use is 2-3 years away — and that may mean that the RTX 2060 won’t have the oomph to use ray tracing effectively by the time the feature becomes useful.

Now Read:

Tuesday, 21 January 2020

AMD Radeon RX 5600 XT Review: Testing With Third-Gen Ryzen