Radeon RX 5700 XT and Radeon RX 5700 review

July 2019 is, in effect, a very hot month. Not only in the meteorological sense of the term, but also for the amount of new hardware products presented, especially considering that some, such as the new graphics cards AMD Navi Radeon RX 5700 and Radeon RX 5700 XT (subject of this review) and the Ryzen 3000 processors, are among the most anticipated news of the year.
The two newcomers not only raise the level of performance of the models bearing the same name, but are sold at a lower price than the previous price lists, while the maintenance of the RTX 2060 in the catalog, at a reduced price, lowers the entry barrier for those who want to buy an RTX board.

This move by Nvidia has a double reading: on the one hand it strengthens its position and advantage in the world of hardware able to manage Ray Tracing in real-time, and on the other hand, it makes a preventive move to counter the arrival of the RX 5700 AMD. After all, AMD’s sights, with the RX 5700 and RX 5700 XT, included the RTX 2060 and RTX 2070, and somehow, changing the target at the last moment by presenting the SUPERs, Nvidia tries to dodge the bullet. We will see during this review whether Nvidia’s move has taken effect, or not.

Radeon RX 5700

It’s been about two years since the offer of AMD graphics cards is substantially devoid of really interesting news. We could mention the Radeon RX 590, presented at the end of 2018, which has slightly improved the performance of the RX 580, and the Radeon VII of the past months, plus a taste of what AMD was working on, rather than a card that can really annoy the GeForce RTX. These are the reasons why the RX 5700 and RX 5700 XT are so eagerly awaited by fans.
The main novelties of the new proposals, codenamed Navi, are a new architecture called Radeon DNA – RDNA – which is exclusively dedicated to gaming, and the adoption of the 7 nm production process, of which Radeon VII gave us a taste. However, there are also several additional features, which we will present in this article, but that we will deepen in the coming weeks in dedicated articles.

As you may have understood from the first lines of this article, RX 5700 and RX 5700 XT are not high-end cards. Originally targeting the GeForce RTX 2060 and RTX 2070, it places the newcomers in the mid and mid-high end of the market, while the lower end is presided over by the RX 570 / 580 / 590, still able to offer more than worthy performance at Full HD resolution, at a very affordable price.

Radeon RX 570 XT

The first two Navi cards, therefore, aim at 1440p resolution, the maximum of detail and quality for the XT version, and something less for the RX 5700 “smooth.” By extension, the latter we can also consider it a presumably interesting choice for Full HD at high frame rates.
The highest end of the market still remains the prerogative of the RTX 2080 / 2080 S and 2080 TI, and of course, there will be no clash in the field of Ray Tracing, as AMD has not yet extracted from the cylinder its proposal for the new (in the field of gaming) rendering technique.

Before getting to the heart of the technical specifications of the two products and the tests, let’s talk briefly about the highlights of the Radeon DNA – RDNA architecture.

RDNA Architecture, Radeon Display Engine, and Media Engine

We have dedicated an article to RDNA architecture, so here we will be more concise – if you want to go deeper then go to that article. To develop RDNA, AMD has stripped the well-known Graphics Core Next (GCN) architecture, which has been the backbone of its solutions for years, of its superfluity.

This immediately gives us an indication of the potential behavior of new cards in video games: we know that there are titles that behave better with Nvidia and others that are more comfortable with AMD. Although AMD speaks of “a new architecture,” I remain of the basics, of the similarities with the GCN, and consequently, we can expect to find the same behavior of the “old” AMD cards with different game engines.

With RDNA, AMD’s goal is to have a fast, efficient, feature-rich and above all scalable architecture, so that it can cover multiple market sectors, from the cloud to smartphones, from a few watts up to hundreds.

We must remember that AMD, thanks to its dualism as a manufacturer of CPUs and GPUs, has signed agreements with major companies in the industry, and its hardware will be the basis of the new Xbox, Playstation 5, Stadia and even some mobile devices (agreement with Samsung). Scalability is essential to adapt the hardware to such different sectors and needs, and as for the new Ryzen 3000 processors, also in this case a key element to ensure all this is the production process to 7 nanometers of TSMC.

With RDNA, AMD’s goal is to have a fast, efficient, feature-rich and above all scalable architecture, so that it can cover multiple market sectors, from the cloud to smartphones, from a few watts up to hundreds.

With RDNA, AMD’s goal is to have a fast, efficient, feature-rich and above all scalable architecture, so that it can cover multiple market sectors, from the cloud to smartphones, from a few watts up to hundreds.

We must remember that AMD, thanks to its dualism as a manufacturer of CPUs and GPUs, has signed agreements with major companies in the industry, and its hardware will be the basis of the new Xbox, Playstation 5, Stadia and even some mobile devices (agreement with Samsung). Scalability is essential to adapt the hardware to such different sectors and needs, and as for the new Ryzen 3000 processors, also in this case a key element to ensure all this is the production process to 7 nanometers of TSMC.

The first novelty of the architecture concerns the Compute Unit, totally redesigned so that it manages the information in a more efficient and fast way. AMD’s goal was to improve single-thread performance first and foremost. The company doubled the speed of instruction management by integrating twice as many scalar units and schedulers. The new architecture can also issue a single instruction per clock cycle and work on two operating modes to make the most of ALU.

Two Compute Units can also work together to form a single element – capacity called “resource pooling” – so as to reduce execution latency (the GPU is used more), improve single-thread performance and manage the cache more efficiently.

RDNA has a new multi-level cache hierarchy that is also designed to contain latency, increase bandwidth, and control power consumption. The new project foresees the introduction of the L1 cache (512KB in total) and a doubling of the bandwidth load as far as the L0 cache up to the ALU. Sunnyvale has improved the Delta Color Compression (DCC) algorithm to use the available bandwidth more efficiently.

AMD has also streamlined the graphics pipeline to achieve higher performance per clock and frequency. By working on clocking gating to reduce power consumption (one of Vega’s problems) and reducing logic levels to ensure higher frequencies, engineers were able to achieve a performance increase of about 10% over GCN.

All these changes should result in an improvement of about 50% in performance per watt compared to the Vega 10 (RX Vega 64/56), a 25% increase in performance per clock, and up to 2.3 times more performance per area.

Radeon Display Engine and Radeon Media Engine

The new Radeon RX 5000 boards come with a new Radeon Display Engine that provides support for the latest standards and resolutions: we speak of HDMI 2.0b and DisplayPort 1.4 with Stream Compression Display 1.2a that allows you to manage with a single cable 4K at 240 Hz, 4K HDR at 120 Hz and 8K HDR at 60 Hz.

Stream Compression 1.2a display allows 4K monitors to work at refresh rates over 120 Hz without the need for chromatic subsampling. Until now, 3840×2160 and 144 Hz operation required a sacrifice in image quality to stay within DisplayPort bandwidth. Alternatively, 120 Hz with 8-bit color depth could be used. And if you found yourself using an HDR monitor, you had to switch to 98 Hz for a 10-bit color depth.

Now, with the benefit of almost lossless compression, AMD allows you to operate 4K monitors up to 120 Hz or 8K displays up to 60 Hz via a single cable and without worrying about compromise. The only thing to remember is that the monitor must also support DSC for it to work. Asus is in the front row with a 43-inch DisplayHDR1000 at 3840×2160 and refresh rate at 144 Hz.

AMD has also improved the Radeon Media Engine, with 40% more performance than the previous generation. This allows the card to support those who stream their gameplay and want to do so efficiently: the Navi 10 GPU can handle decoding via VP9 in 4K90 or 8K24.

As far as decoding in H.264 is concerned, 1080p600 or 4K150 are mentioned, while for AMD encoding 1080p360 or 4K90 are mentioned. On the H.265 front, AMD indicates 1080p360, 4K90 or 8K24 decoding, while on the encoding front it indicates 1080p360 or 4K60.

Comparison of technical specifications

The Radeon RX 5700 and RX 5700 XT have two new features compared to other cards currently on the market: GDDR6 memory and support for PCI Express 4.0. The GDDR6 can work at much higher speeds than the GDDR5. Remember that AMD has never adopted GDDR5X, using GDDR5 or HBM/HBM2.

However, the latter is expensive and despite the high bandwidth provided, have not given a clear advantage over Nvidia’s competitors. So AMD decided to use the GDDR6, the right compromise in all respects.

PCI Express 4.0 offers twice the bandwidth of PCIe 3.0, but cannot be considered a real added value, at least not today.

We will certainly test for further testing, but at the moment we are quite sure that in real video games PCIe 4.0 does not offer any advantage, because the current cards, even the most powerful models, struggle to saturate a PCIe 3.0 16x connection (often the loads do not exceed half, ie, do not saturate a PCIe 3.0 8x). Using specific tests, such as the new test integrated in the 3DMark, you can see a difference, but we are talking about a test made with the sole purpose of showing this difference, not a real game.

Both Radeons adopt a GPU called Navi 10, produced at 7 nanometers, which occupies an area of 251 mm2 (10.3 billion transistors), much smaller than the 495 mm2 of the Vega 10 GPU on board the RX Vega 64 (14 nanometers).

The Radeon RX 5700 XT is equipped with 40 Compute Units, for a total of 2560 stream processors (64 x 40) and up to 9.75 TFlops computing power with FP32 operations. There are also 8 GB of GDDR6 memory and GPU frequencies described as follows: 1605 MHz base, 1755 MHz in Game and 1905 MHz in Boost. The TBP (typical board power) of this card is 225 watts. The Game frequency, or Game Clock, is nothing more than the frequency at which the video card should work in the presence of a typical workload, such as that of video games, where the clock range may vary from one scene to another.

The smaller model Radeon RX 5700 has 36 active Compute Units, for a total of 2304 stream processors. As a result, the board provides FP32 computing power of 7.95 TFLOPs. Again we find 8 GB of GDDR6 memory, while the operating frequencies are lower: 1465 MHz base, 1625 MHz in Game and 1725 MHz in Boost. The TBP is 180 watts.

On both the GDDR6 memory is set to 14 Gbps, while the bus is 256 bits, for a bandwidth of 448 GB/s, higher than that of the RX Vega 56 (410 GB/s), but lower than the 483 GB/s of the RX Vega 64. Other common features are the 64 ROPs, while the texture drives are 160 on the RX 5700 XT and 144 on the RX 5700.

On the front of the video outputs, note the three DisplayPorts 1.4 with DSC (Display Stream Compression 1.2a) and the HDMI 2.0b output. If anyone was wondering, there is no USB C: at the moment AMD does not seem to believe in the takeoff of virtual reality, even if technically VirtualLink is supported by the GPU. As a result of these video outputs, the cards can handle 4K screens at 240 Hz, 4K HDR screens at 120 Hz, or 8K HDR screens at 60 Hz.

As far as the cooling system of the reference proposals is concerned, AMD has once again relied on a “blower” solution, with a single fan and a cover that expels the hot air produced entirely from the rear slots. The cover and backplate are made of aluminium alloy. The cover has an optimized design to ensure good airflow and low noise, although the small recess you see on the 5700 XT is only aesthetic in character.

The RX 5700 XT is equipped with a 7-step digital power supply system (6 on the RX 5700), which AMD claims to be “overclocking ready”. The boards are powered by two power connectors, one 8-pin and one 6-pin.

AMD has also created a third video card, a golden edition of the Radeon RX 5700 XT in homage to the 50 years of AMD, which is distinguished by higher working frequencies than the classic model: we speak of 1680 MHz base, 1830 MHz for the Game Clock and 1980 MHz in Boost.

Let’s take a look at the technical specifications of the two RX 5700s and their direct competitors, the GeForce RTX 2060 and 2070 boards. Net of architectural differences, which can result in more or less efficient handling of specific instructions, the RX 5700, with its 36 CU and 2304 stream processors, as well as a proportionally higher number of texture units, ROPs, etc., is on paper ahead of the RTX 2060.

This disparity has been partially recovered by the RTX 2060 SUPER, which increases the technical specifications in all respects. The same is true for the higher, albeit slightly higher, operating frequencies on the RX 5700, which can make a difference where the architecture is equally efficient.

The same situation is repeated when comparing the RX 5700 XT and the RTX 2070. The RTX 2070 offers more limited technical features, which are then practically evened out by the RTX 2070 SUPER.

The values of energy consumption are in favour of Nvidia, especially considering the production process at 7 nm, which should give AMD an advantage also in terms of energy efficiency.

If we were to look only at these features, we would say that the RX 5700 and 5700 XT will be faster than the RTX 2060 and RTX 2070 “base”, while in a direct comparison with the SUPER, the RX 5700 can still have its say, while the RX 5700 XT will have to struggle to do better than the RTX 2070 SUPER.

The only difference, then, will be the efficiency of the RDNA architecture, as well as the drivers, which will play a key role while we should not underestimate the games themselves, which will be fed into a new architecture, although the general logic, and those defined by the GCN of the previous AMD cards, have not been distorted.

Radeon Anti-Lag, Radeon Image Sharpening and Radeon Chill

Among the new software functions, we mentioned at the beginning of the article is Radeon Anti-Lag (RAL). It is a function that reduces the input lag, i.e. the time that passes between the click you make on the mouse, the pressure of a key on the keyboard or on a controller, and the time when the action is represented on the screen. This is a very important parameter in multiplayer, where a split second can mark your victory or defeat.

According to AMD by going to work on GPU and CPU rendering cycles and introducing “waiting times” into scenarios where you are limited by the GPU, you can synchronize the work of the CPU and GPU and reduce the lag time by up to 31% on average, for example from 56 to 44 milliseconds. That’s what’s behind Anti-Lag.

Nvidia claims to have developed over the years a similar system, called “maximum pre-rendered frames”, which controls the size of the flip queue. It can be managed from the Control Panel of the video card. “The higher the number set, the higher the lag between the input and the displayed frame, while the shorter the tail length, the lower the lag. However, this increases the work on the CPU”, explains Nvidia.

The other important news is Radeon Image Sharpening (RIS). Many games use post-processing anti-aliasing forms such as TAA and FXAA to smooth out the scale on the edges of objects. These AA methods offer a good mix of performance and quality but have the disadvantage of blurring the entire scene.

To make matters worse, many players choose to play at low resolutions on high-resolution screens to maintain a high frame rate. RIS tries to restore image sharpness: it’s based on a technique called Contrast Adaptive Sharpening (CAS), plus optional upscaling.

The result is that with RIS enabled some elements of the scene appear sharper and more visible. According to AMD the performance impact is in the order of 1% and RIS is supported by thousands of games as it does not require integration in the title. It works on titles in DX9, DX12 and Vulkan. Developers can implement CAS directly during development using the FidelityFX toolkit.

AMD has also renewed Radeon Chill. This function dynamically adjusts the frame rate according to the movements detected during gameplay, so as not to produce more fps than necessary and thus reduce power consumption and temperature of the GPU.

With the new drivers, the function becomes somehow “screen conscious”, and sets the frame rate limits based on the refresh rate of the monitor – whether it’s fixed or variable. This way, according to AMD, Radeon Chill saves 2.5 times more power than in the past.

As mentioned at the beginning of the article, we will devote to the individual technologies of the insights that we will publish in the coming weeks, so stay tuned.


The new RXs have been expected for months, and now that we have tried them, we feel a taste between hope and missed an opportunity. The Navi architecture has clear potential, the RX 5700 offers performance comparable to the RTX 2060, while the RX 5700 XT equals the RTX 2070 in performance. Nvidia, however, played ahead, presenting the RTX 2060 SUPER and RTX 2070 SUPER, almost surprisingly, a couple of weeks before this launch.

As a result, Nvidia’s newcomers offer on average better performance than the two new Radeons: if we consider the 1440p resolution, the most interesting for these cards, the RTX 2060 SUPER is on average faster (4%) than the RX 5700, despite the fact that there are games where the Radeon shows muscles, but also titles where the Nvidia card is superior. The RX 5700 XT is worse off, with the RTX 2070 SUPER with an average advantage of 10%, but even in this case there are games where the architecture Navi proves very efficient, and cases where there is no match.

In short, from the performance point of view AMD cannot say that it has taken the scepter from Nvidia, at least in the range of belonging of these cards, but the difference in performance is not what can be considered a resounding defeat. Especially considering the fact that these new Radeons have pushed Nvidia to make the SUPER models, and to revise downwards all the prices of RTX cards. In short, there have been a couple of well-settled shots by AMD, which as a result have made the market more competitive and economical, which will make all users happy. Incidentally, if Nvidia had not presented the SUPER models, and had not lowered prices, the conclusion of this test would have been much clearer and in favor of AMD.

To further simplify the demand, we could also focus solely on the lack of Ray Tracing and DLSS, since as we have seen the disparity in performance, although objectively exists, is not such as to fall back the cards in different performance bands, in addition to the fact that AMD has accustomed us to see an increase in the performance of its cards in the months following the launch, thanks to optimized drivers, in addition to possible patches of the same games that today are found to be managed by a new architecture.

As for the RX 5700, for us, the answer is no. Although it is the board that has shown the most solid performance throughout our suite of benchmarks, always doing as well as an RTX 2060 and beating, and sometimes doing better than the RTX 2060 SUPER, the price difference is too small to do without the technologies now offered by the RTX line. Even if you think that at the moment the Ray Tracing is still too little important or supported, although we have already had the opportunity to think about how much has changed in a year, the few tens of euros that distance the two cards do not justify the renunciation.

For the RX 5700 XT, the situation is a bit different, as the price difference of about 20% starts to be important. And that’s where other factors come in, especially “how much more does that hundred dollars weigh to you? If the answer is “not much”, then it makes sense to opt for the Nvidia solution, which is more expensive but has more to offer. Especially since the RTX 2070 SUPER almost always performs better than the RX 5700 XT in terms of performance, more than the RTX 2060 SUPER does compare to the RX 5700. The same can be said if you are buying a new PC, and you are ready to spend over a thousand euros: how much do those 100 euros affect the final expenditure? Certainly less than the purchase of a single video card, and it makes sense to consider an RTX card.

If you have a fixed budget and are interested in generic performance with all games, and Ray Tracing is just something more you can still do without – because today you can still do without it. In this case, AMD offers an economic advantage; whether it’s 30 or 100 dollars; however, it does offer it. If we were to consider the performance per watt ratio, Nvidia is superior to AMD, but considering the performance per euro ratio, the RX 5700 boards are definitely better – always net of Ray Tracing and other additional technologies.

These are the facts from which we can draw conclusions today. From this moment on, it will be necessary to follow the arrival of the cards on the market, and the offers of the various partners carefully. AMD may become even more aggressive, with every euro less that it decides to remove from its cards, more attractive will become for all those attentive to spending and who want to give the maximum value to each euro. But beware, because we still have to witness the fall in prices of the RTX 2060, and we can not rule out that in the coming weeks the RTX 2070 and RTX 2080, in phase-out, could be subject to major price cuts.

More Similar Posts

Trending Posts