AMD Radeon RX 5700 XT and RX 5700 Graphics Card Review – Bringing Competition Back

The day many have been waiting for has finally arrived. AMD has finally launched both their Ryzen 3000 series CPUs, and the Radeon RX 5700 series GPUs. In this report we’ll be focusing on the latter as we take a look at AMD’s long-awaited Navi-based RX 5700-series of graphics cards.

With the recent announcement of NVIDIA’s refreshed SUPER series, AMD has been driven to lower prices from the original announcement of the 5700-series, which is good news for gamers as a whole. This also means that the overall targeting has been changed as well. Originally, the RX 5700 XT was planned to compete with the RTX 2070, but it will now compete directly in pricing with the slightly cut-down RTX 2060 SUPER instead which should make for some interesting results.

Radeon RX 5700 XT and RX 5700 Specifications

Model	RX 5700 XT	RX 5700	RX Vega 64
Compute Units	40	36	64
Stream Processors	2560	2304	4096
ROPs	64	64	64
TMUs	160	144	256
Base Clock (MHz)	1605	1465	1200
Game Clock (MHz)	1755	1625	N/A
Boost Clock (MHz)	1905	1725	1536
Peak Compute (TFLOPs)	9.75	7.9	10.5
Memory Type	14 Gbps GDDR6	14 Gbps GDDR6	HBM2
Memory Bus	256-bit	256-bit	2048-bit
Memory Bandwidth	448 GB/s	448 GB/s	483.8 GB/s
VRAM	8GB	8GB	8GB
TBP/TDP	225W (TBP)	180W (TBP)	295W (TDP)
Manufacturing Process	TSMC 7nm	TSMC 7nm	14nm
Architecture	RDNA	RDNA	GCN 5.0
GPU	Navi 10	Navi 10	Vega 10
Die Size	251 mm²	251 mm²	495 mm²
Launch Price	$399 ($449 for Anniversary Edition)	$349	$499
Launch Date	07/07/2019	07/07/2019	08/14/2017

Comparing to the previous generation Vega, we can see that the Navi 10 GPU is much smaller overall, in fact it’s nearly 50% the total die size. This is mostly attributed to the smaller 7nm process node, but also due to the ~37% reduction in Compute Units from 64 to 40 total. That said, despite that we have a 24% increase in ‘Boost’ clock speed, which results in a Peak Compute that is only 7% lower overall.

Comparing the flagship RX 5700 XT to the cut-down RX 5700, we see a 10% reduction in Compute Units and Texture Mapping Units, while the ROPs (Raster Operations Pipeline) remain the same. There’s also a 9.5% reduction in ‘Boost’ clock speed. Thankfully, memory remains unchanged as we see the same 8GB of 14Gbps GDDR6 on a 256-bit bus. Taking all the cuts together results in a Peak Compute that is actually 18% lower. We’ll find out soon enough how that translates into actual gaming performance.

A note on clock speeds: AMD has told us that the ‘Boost’ clock is neither guaranteed nor set in VBIOs, but is the maximum clock speed one should expect during lighter works. Likewise, the ‘Game’ clock is also not guaranteed or set in VBIOs, but is the typical clock speed a user can expect during actual gaming workloads.

RDNA Architecture Overview

With the RX 5700-series, AMD is introducing what they call ‘RDNA’ an all-new architecture built to replace the aging GCN, which has been in use in some variation since the introduction of the HD 7970 in 2012. That said, don’t expect GCN to go away, as the company will still use the GCN-based Vega graphics in a number of enterprise focused products. Still, in terms of gaming, the company says it is all in on RDNA going forward.

Taking a look at the ‘Navi 10’ GPU, we can see that it features two Shader Engines consisting of 20 Compute each, for a total of 40. It also features a streamlined Graphics Engine, which consists of 4 ‘enhanced’ Asynchronous Compute Engines, along with a centralized Geometry Processor with 4 Primitive Units. That last one is interesting, as the Primitive Shader, a feature originally touted by Vega has also made a return with Navi, and it is now enabled and fully exposed. This was a rather controversial feature for Vega as it was never actually turned on in any products.

There’s also a new multilevel cache hierarchy designed to lower latency, increase bandwidth and reduce power. This includes a new 128KB L1 cache, with two located in each Sahder Engine for a total of 512KB. In addition, there’s also 4MB of L2 cache, which features 24% reduced latency compared to GCN.

Digging into the RDNA Compute Unit itself, we can see that quite a bit has changed compared to GCN.

The SIMD has been redesigned, with an increase in width from SIMD 16 to SIMD 32. In addition, the wavefront width has also been halved from 64 threads wide, to a native 32 threads, which means the number of wavefront threads will match the number of SIMDs. All of this allows for a more efficient use of clock cycles, and is said to improve single-threaded performance.

Of course, Navi features more than just architectural improvements, as there is also a fairly major improvement in manufacturing process with the switch from Globalfoundries’ 14nm finFET to TSMC’s 7nm node. This change, along with all the others allows for a claimed 1.5x performance per watt compared to the Vega 64. This translates into 14% more performance and 23% lower power draw. In addition, the company claims a massive 2.3x increase in performance per area with the reduction in die size from 495 mm² to 251 mm².

Radeon RX 5700 series is also the first AMD graphics card series to feature GDDR6 memory, as well as the first graphics card to feature PCI-Express 4.0 support. Other notable hardware improvements include the Radeon Media Engine, which allows for a wider range of media decoding and encoding support.

	VP9	H.264	H.265
Decode	4K90 8K24	1080p600 4K60	1080p:360 4K90 8K24
Encode	N/A	1080p360 4K90	1080p360 4K90

There’s also the new Radeon Display Engine; which features support for DisplayPort 1.4 Display Stream Compression or DSC: This compresses the image out to the monitor to reduce the necessary bandwidth for 4K HDR at 144Hz, or 8K HDR at 60Hz.

You also get HDMI 2.0b, unfortunately HDMI 2.1 and its native Variable Refresh Rate support are absent here. However, FreeSync is still supported over HDMI using AMD’s proprietary implementation. This isn’t a huge deal so long as it plays nice with TV’s that support FreeSync, and it should.

Radeon Software Features

Along with the Radeon RX 5700 series, AMD is also introducing a number of new software features to be included in the Radeon Adrenalin 2019 driver. These features include Radeon Image Sharpening (RIS) and Radeon Anti-Lag, we’ll go over what they are and how they work below.

Radeon Image Sharpening

RIS is a solution for a few different problems. First, it can help to restore image sharpness that is lost due to other post-processing effects such as TAA or FXAA. It also works as a competitor to DLSS, to improve image quality while upscaling to 4K from lower resolutions such as 1440p or 1800p.

RIS is based on an algorithm that modulates the sharpening strength based on the contrast of the image. This allows it to sharpen interior object details, while leaving the higher contrast areas such as edges mostly untouched. This allows the feature to be combined with any other anti-aliasing effect without producing the ‘halo’ effect that is commonly displayed with other sharpening methods.

We won’t be testing this feature in the scope of this report, but you should check back for more from us on the subject.

Radeon Anti-Lag

Radeon Anti-Lag is designed to reduce the perceivable amount of input lag or the ‘click-to-response’ time from when you click a button and see the results on screen. Anti-Lag isn’t magic, and it works mostly in GPU-bound scenarios by reducing the latency in the interactions between the game and the graphics driver.

We did test this for ourselves in a couple of titles, and we’ll go over the results later.

AMD FidelityFX

Finally, we have FidelityFX which is a Contrast Adaptive Sharpening (CAS) technique which can be integrated into directly into games. It is very similar to Radeon Image Sharpening, but can provide even greater detail with minimal artifacting.

This feature is mostly GPU-agnostic, but it is claimed to run the best on Vega and Navi-based GPUs thanks to their support for Rapid Packed Math. It does require the game to support it, but as of writing it is already available in Rage 2 and World War Z.

Again, we’ll be covering this in greater detail in a future report.

Next, we’ll go over the physical overview of the RX 5700 XT and RX 5700 XT, testing methods, power, thermals, game performance, and our conclusion.

Radeon RX 5700 XT and RX 5700 Overview

Both the Radeon 5700-series cards feature a dual-slot, blower-style cooling solution. However, after that the similarities in design pretty much end. The RX 5700 XT features a unique, machined aluminum shroud with a mostly dark gray color with red accents. Along the side, sits an LED-backlit Radeon logo, which glows red when powered on. Along that same side also sits a red accented stripe which bends along the inward curve towards the back. At the back, there’s some ventilation, which allows the fan to pull in cool air from your case’s front intake. This is an interesting design for a blower-style cooler, and we’ll see how effective it is later.

There’s also a premium, all black metal backplate that features the same red accented stripes and Radeon logo toward the back. At this point, the “dent” has probably reached peak-memeifcation, but I honestly find the overall design to be quite nice and sophisticated. It is definitely one of the nicest Radeon reference designs to date.

The RX 5700 also features an all-aluminum shroud, but its much more subtle, and looks very similar to past Radeon-reference designs. It lacks any special curves, or stripes, and while it does feature a red Radeon logo on the side, it is not LED-backlit. Some may actually prefer this cleaner design, and I wouldn’t blame them for it. However, those clean looks are betrayed by the fact that the 5700 lacks a backplate, which is really a shame for a card of this price point.

Both cards measure about 266mm long (10.5-inches). They feature a 7-phase all-digital VRM system that is powered by a both an 8-pin and a 6-pin PCIe power connector. There’s also 3x DisplayPort 1.4 and a single HDMI 2.0b for connectivity.

Graphics Card Test Setup and Methodology

All game benchmarks are taken from a 3-run average either in-game or using the title’s built-in benchmark utility. Performance is recorded using OCAT: the open-source frontend for PresentMon, which itself is an open-source capture and analytics tool which supports modern APIs such as DirectX 12 and Vulkan.

For our graphics card reviews, we use a 2nd generation Ryzen-based system with some parts supplied by AMD and DEEPCOOL. Thanks to all who sponsor components in our test systems, as without them it would be very costly to source the components we need.

	Test Bench
CPU	AMD Ryzen 7 2700 @ 4.1GHz
Motherboard	ASUS ROG Crosshair VI Hero X370
RAM	16GB (2x8GB) Corsair Vengeance RGB Pro DDR4-3200 CL16
GPU	N/A
Storage	500GB Samsung 850 EVO SSD (OS), 480GB Patriot Burst SSD
Cooler	DeepCool Castle 240 RGB all-in-one liquid cooler.
Case	Phanteks P350X
PSU	DeepCool Quanta DQ1250

Radeon RX 5700 XT and RX 5700 Power and Thermals

Power Draw

Temperatures

Radeon RX 5700 XT and RX 5700 Performance

3DMark Fire Strike and Time Spy

Fire Strike

Time Spy

Assassin’s Creed: Odyssey

We test this title using the game’s built-in benchmark utility.

1440p

2160p

Radeon Anti-Lag Performance Testing

While we lack the necessary equipment to measure input lag for the full chain from mouse to display, we can at least measure the driver latency, which is what Anti-Lag aims to improve. AMD recently updated the open-source performance capturing application OCAT to include ‘Average Estimated Driver Lag (ms)’ in the output. We’ll average that over three runs in order to get our final result.

Taking a look at the above results, we can see that the feature does indeed improve driver lag, shaving off about 4ms on average between the two titles we tested. That said, this does not come without a cost as we’ll see when we look at performance impact below.

Looking at the above results, we can see that with Anti-Lag turned on, we saw a 3% reduction in performance in Rainbow: Six Siege, which isn’t too bad considering the latency reduction we received. That said, we saw a 7% decrease in performance in Battlefield V, which is too much for me to want to enable it in that title.

We’ll have to investigate further with more titles in order to come to any real conclusion, but the results here are interesting nonetheless.

Conclusion

After a very interesting week and a boatload of benchmarking, I think I can finally come to a decisive conclusion regarding AMD’s Radeon RX 5700 series. When it was first announced, I have to admit I was a little underwhelmed, mostly due to the original pricing, which I felt was a bit too high to get me excited. After all, these new Radeon cards do not feature any sort of hybrid Ray Tracing support, and therefore cannot compete directly with NVIDIA’s RTX lineup, at least purely from a feature standpoint. That said, AMD’s recent decision to cut prices ahead of launch has changed my view. Now, whether that decision was a reaction to NVIDIA’s release of ‘SUPER’, or it was planned from the start– I don’t know, and frankly, it doesn’t matter. The bottom line is, you’re getting more performance per dollar than you would have before, and that’s definitely something to be happy about.

Now, as I said in the introduction: The RX 5700 XT and RX 5700 are priced to compete with the GeForce RTX 2060 SUPER and RTX 2060. Unfortunately, we do not have an RTX 2060 SUPER for comparison, but we do have an RTX 2070 and that should fill that role just fine as others have found the RTX 2060 SUPER to be just 1-2% slower than the 2070, overall. So, with that mind, let’s take a look at some performance summary charts!

Looking at the above chart we see that there’s a couple of losses, a few draws, a few small wins and some huge victories. Overall, the RX 5700 XT is 10% faster than the RTX 2070 on average, meaning it should be even faster than the RTX 2060 SUPER. That said, the score is helped a lot by a rather huge win in World War Z, so removing that does give us a difference of just ~5% on average, which is still a fairly decent win.

The RX 5700 fares even better against the RTX 2060, as it manages to outpace it in every title tested and wins by 16% overall. Again, this is helped a lot by a huge win in World War Z, but if we remove that it’s still 11% faster overall, which is a pretty decisive result.

Now, performance isn’t everything, and no doubt many of you will be wondering about power consumption. Well, as we saw in our powerd draw tests, the RX 5700 XT drew ~6% more power than the RTX 2070, which makes it roughly even, if not a little better in terms of performance-per-watt. Meanwhile, the RX 5700 drew just about 4% more power compared to the RTX 2060 while delivering 16% more performance, making it the clear winner in terms of efficiency. So, I’d say that the 7nm node has paid off for AMD, but I will admit that only just beating NVIDIA on a much smaller node isn’t really all that impressive. Still, a win is a win.

So, which one should you buy? Well, NVIDIA has two key value additions when compared to the RX 5700 series. For one, you have RTX or hybrid ray tracing, which may or may not interest you and for two you have what I feel is a better game bundle in the inclusion of Control and Wolfenstein: Young Blood, whereas the RX 5700 series includes 3-months of the PC version of Xbox Game Pass, which is nice, but really only boils down to a $15 value. So, those do factor into the decision and they make a recommendation either way a bit more difficult.

That said, it’s still pretty clear to me that if you’re looking for the best possible performance at the $399 and $349 price points, the AMD Radeon RX 5700 XT and RX 5700 are the clear choice, especially the latter as it leaves the RTX 2060 in the dust. I’m also giving them both our first ever Gold Adored award for bringing back fierce competition to a segment of the market that hasn’t seen any in quite a while.

Liked it? Take a second to support Donny Stanley on Patreon!