Sapphire RX 7900 XT Pulse: RTX 4070 Ti crushed in rasterization

Ľubomír Samák

1 year ago

3DMark

While someone may find the RX 7900 XT graphics card to be the most attractive within its price range, another may see the RTX 4070 Ti in that position. And both are right, they might just disagree on priorities. Outside of ray-tracing graphics, the Radeon is even attacking the RTX 4080. Admittedly this is at the expense of, say, worse technological balance and also worse power efficiency, but if someone doesn’t appreciate GeForce’s strengths…

The segmentation of gaming graphics cards with AMD chips is quite a bit thinner than Nvidia’s. The latter has already four GeForce cards (RTX 4070, RTX 4070 Ti, RTX 4080 and RTX 4090) in the current generation (Ada Lovelace) compared to two Radeons (RX 7900 XT and RX 7900 XTX) with RDNA 3. Still, it can’t be said that Nvidia has a downright equivalent to any of the RX 7000 Radeons.

The emphasis of the companies (AMD and Nvidia) on different features is so pronounced that it raises quite a lot of emotions around what is more and less important for whom. Arguments about the higher attractiveness of one graphics card or another are somewhat based on different “worldviews”. Everyone prefers slightly different things, and the task of this article is to point out all the strengths as well as weaknesses of the Sapphire RX 7900 XT Pulse compared to the MSI RTX 4070 Ti Suprim X 12G and MSI RTX 4080 16GB Suprim X.

From the basic specs that apply to all Radeon RX 7900 XT graphics cards: the GPU is the Navi 31 XT, the same chip that the RX 7900 XTX has, but without 12 active compute units. With 84 CUs, the RT 7900 XT has thus 5376 shaders at its disposal. The memory is “only” GDDR6 type (GDDR6X, as in GeForce, is not used by AMD), but because of the wide memory bus, an extremely high bandwidth is still achieved, over 800 GB/s. The memory size is always 20 GB with the RX 7900 XT. Furthermore, DisplayPort 2.1 is supported, which compared to the GeForce RTX 4000 (with DP 1.4a) is a key thing for those interested in an 8K monitor with 165 Hz. Graphics cards with Ada Lovelace GPUs will also display 7680 × 4320 px resolution, but at a lower refresh rate (60 Hz). Traditionally, the full parameters can be found in the overview chart of the second chapter.

And now to the specifics of the Sapphire Pulse: One is the increased clock speed for the GPU boost, compared to the reference model (2394 MHz) by 56 MHz (2450 MHz). These are the official, guaranteed speeds, which tend to be exceeded in practice. Sometimes quite significantly, but we’ll discuss that later, during the evaluation.

Sapphire RX 7900 XT Pulse in detail

The “Pulse” series of graphics cards represents the lowest-cost non-reference cooler options in Sapphire’s lineup. The cooler is built on the conventional all-copper coldplate and heatpipe technology. Only the Nitro+ Vapor-X model is based on a vapor chamber.

Three 90 mm fans are used to cool the heatsink. However, the contacts to connect them are no longer in the frame, but each fan has a short cable that chains to the remaining fans. While this eliminates the extremely quick replacement of a faulty fan, at least you still don’t have to replace all three fans at the same time when one fails. Disconnecting the fan does require removing the housing, however. The model designation of the fans, by the way, is FD9015H12D. Might come in handy later to find them, at the moment search engines don’t index it much.

The backplate (as opposed to the plastic housing around the fans) is made of metal (aluminum) and Sapphire also uses it to absorb heat from the memory. Not only between the memory and the heatsink, but also behind the memory – on the back of the PCB – there are thermal pads. And full-sized, big ones.

There is then a grille at the back of the card to better dissipate heated air. It’s smaller than on GeForce graphics cards, but that’s because Radeons have significantly larger PCBs. In this case, the cooler only extends it by 35.5mm.

When looking at the fans in detail, it is worth pointing out the two elements that stiffen the blades. Firstly, the central ‘bend’ and secondly, the framing of the blades with a circular tunnel. In both cases, the manufacturer works to increase the strength of the blades and eliminate vibration at the tips that would occur without these modifications due to high flexibility. This would then, in addition to the more pronounced resonant frequencies (tonal peaks) of the sound, eventually lead to resonating the case and subsequently the cooler, which occurs with some simpler designs.

At 315mm, it’s not a downright short graphics card, but it’s not a long one either. Still, you won’t usually have compatibility complications in smaller cases (as with the more powerful GeForce RTX 4000s) even if you install a liquid cooler radiator (with fans) behind the front panel.

The Sapphire RX 7900 XT Pulse is also significantly slimmer compared to GeForce cards from similar performance class (RTX 4070 Ti and RTX 4080). With almost 55 mm, it does not extend above the fourth PCI Express slot, so you can comfortably, that is, with a reasonable margin for (de)mounting, use it for an additional card. In width (from the connector) the RX 7900 XT Pulse has 134 mm. It thus extends some 27 mm beyond the cover around the outputs, but this is usually again less than with the overall larger competing graphics cards with Nvidia GPUs.

The cooler heatsink itself is 20–25 mm in height (the fins above the GPU and VRAM are taller) at a weight of 658 grams. A network of six heatpipes is used for heat distribution.

Note: The article continues in the following chapters.


Parameters		Sapphire RX 7900 XT Pulse
		MSI RTX 4070 Ti Suprim X 12G
Architecture		RDNA 3
Die		Navi 31 XT (215-145000156)
Manufacturing node		5 nm + 6 nm
Die size		300 mm² + 5× 37 mm²
Transistor count
Compute units		84
Shaders/CUDA cores		5376
Base Clock		1720 MHz
Game Clock (AMD)		2075 MHz
Boost Clock		2450 MHz
RT units		84
AI/tensor cores		–
ROPs		192
TMUs		336
L2 Cache		6 MB
Infinity Cache		80 MB
Interface		PCIe 4.0 ×16
Multi-GPU interconnect		–
Memory		20 GB GDDR6
Memory clock (effective)		20.0 GHz
Memory bus		320 bit
Memory bandwidth		800.0 GB/s
Pixel fillrate		470.2 Gpx/s
Texture fillrate		822.9 Gtx/s
FLOPS (FP32)		52.0 TFLOPS
FLOPS (FP64)		1609.0 GFLOPS
FLOPS (FP16)		104.62 TFLOPS
AI/tensor TOPS (INT8)		–
AI/tensor FLOPS (FP16)		–
TDP		315 W
Power connectors		2× 8-pin
Card lenght		313 mm
Card slots used		54 mm
Shader Model		6.7
DirectX/Feature Level		DX 12 Ultimate (12_2)
OpenGL		4.6
Vulkan		1.3
OpenCL		2.2
CUDA		–
Video encoder engine		VCN 4.0
Encoding formats		HEVC, H.264, AV1
Encoding resolution		8K
Video decoder engine		VCN 4.0
Decoding formats		HEVC, H.264,VP9, AV1
Decoding resolution		8K
Max. Monitor resolution		7680 × 4320 px
HDMI		2× (2.1a)
DisplayPort		2× (2.1 UHBR)
USB-C		–
MSRP		899 EUR

/* Here you can add custom CSS for the current table */ /* Lean more about CSS: https://en.wikipedia.org/wiki/Cascading_Style_Sheets */ /* To prevent the use of styles to other tables use "#supsystic-table-2291" as a base selector for example: #supsystic-table-2291 { ... } #supsystic-table-2291 tbody { ... } #supsystic-table-2291 tbody tr { ... } */

Gaming tests

The largest sample of tests is from games. This is quite natural given that GeForce and Radeons, i.e. cards primarily intended for gaming use, will mostly be tested.

We chose the test games primarily to ensure the balance between the titles better optimized for the GPU of one manufacturer (AMD) or the other one (Nvidia). But we also took into account the popularity of the titles so that you could find your own results in the charts. Emphasis was also placed on genre diversity. Games such as RTS, FPS, TPS, car racing as well as a flight simulator, traditional RPG and sports games are represented by the most played football game. You can find a list of test games in the library of chapters (9–32), with each game having its own chapter, sometimes even two (chapters) for the best possible clarity, but this has its good reason, which we will share with you in the following text.

Before we start the gaming tests, each graphics card will pass the tests in 3D Mark to warm up to operating temperature. That’s good synthetics to start with.

We’re testing performance in games across three resolutions with an aspect ratio of 16:9 – FHD (1920 × 1080 px), QHD (2560 × 1440 px) and UHD (3840 × 2160 px) and always with the highest graphic settings, which can be set the same on all current GeForce and Radeon graphics cards. We turned off proprietary settings for the objectivity of the conclusions, and the settings with ray-tracing graphics are tested separately, as lower class GPUs do not support them. You will find their results in the complementary chapters. In addition to native ray-tracing, also after deploying Nvidia DLSS (2.0) and AMD FidelityFX CAS.

If the game has a built-in benchmark, we use that one (the only exception is Forza Horizon 4, where due to its instability – it used to crash here and there – we drive on our track), in other cases the measurements take place on the games’ own scenes. From those we capture the times of consecutive frames in tables (CSV) via OCAT, which FLAT interprets into intelligible fps speech. Both of these applications are from the workshop of colleagues from the gpureport.cz magazine. In addition to the average frame rate, we also write the minimum in the graphs. That contributes significantly to the overall gaming experience. For the highest possible accuracy, all measurements are repeated three times and the final results form their average value.

Computational tests

Testing the graphics card comprehensively, even in terms of computing power, is more difficult than drawing conclusions from the gaming environment. Just because such tests are usually associated with expensive software that you don’t just buy for the editorial office. On the other hand, we’ve found ways to bring the available computing performance to you. On the one hand, thanks to well-built benchmarks, on the other hand, there are also some freely available and at the same time relevant applications, and thirdly, we have invested something in the paid ones.

The tests begin with ComputeBench, which computes various simulations (including game graphics). Then we move on to the popular SPECviewperf benchmark (2020), which integrates partial operations from popular 2D and 3D applications, including 3Ds max and SolidWorks. Details on this test package can be found at spec.org. From the same team also comes SPECworkstation 3, where GPU acceleration is in the Caffe and Folding@Home tests. You can also find the results of the LuxMark 3.1 3D render in the graphs, and the remarkable GPGPU theoretical test also includes AIDA64 with FLOPS, IOPS and memory speed measurements.

For obvious reasons, 3D rendering makes the largest portion of the tests. This is also the case, for example, in the Blender practical tests (2.91). In addition to Cycles, we will also test the cards in Eevee and radeon ProRender renderers (let AMD have a related test, as most are optimized for Nvidia cards with proprietary CUDA and OptiX frameworks). Of course, an add-on for V-ray would also be interesting, but at the moment the editorial office can’t afford it, we may manage to get a “press” license in time, though, we’ll see. We want to expand application tests in the future. Definitely with some advanced AI testing (we haven’t come up with a reasonable way yet), including noise reduction (there would be some ideas already, but we haven’t incorporated those due to time constraints).

Graphics cards can also be tested well in photo editing. To get an idea of the performance in the popular Photoshop, we’re using a script in PugetBench, which simulates real work with various filters. Among them are those that use GPU acceleration. A comprehensive benchmark suggesting the performance of raster and vector graphics is then also used in alternative Affinity Photo. In Lightroom, there are remarkable color corrections (Enhance Details) of raw uncompressed photos. We apply these in batches to a 1 GB archive. All of these tasks can be accelerated by both GeForce and Radeon.

From another perspective, there are decryption tests in Hashcat with a selection of AES, MD5, NTLMv2, SHA1, SHA2-256/512 and WPA-EAPOL-PBKDF2 ciphers. Finally, in the OBS and XSplit broadcast applications, we measure how much the game performance will be reduced while recording. It is no longer provided by shaders, but by coders (AMD VCE and Nvidia Nvenc). These tests show how much spare performance each card has for typical online streaming.

There are, of course, more hardware acceleration options, typically for video editing and conversion. However, this is purely in the hands of encoders, which are always the same within one generation of cards from one manufacturer, so there is no point in testing them on every graphics card. It is different across generations and tests of this type will sooner or later appear. Just fine-tuning the metric is left, where the output will always have the same bitrate and pixel match. This is important for objective comparisons, because the encoder of one company/card may be faster in a particular profile with the same settings, but at the expense of the lower quality that another encoder has (but may not have, it’s just an example).

Update: As of November 18, 2022, we are testing all graphics cards only in Resizable BAR active mode. There are three reasons why we will not continue with measurements without ReBAR.

The main reason is that new motherboards starting with Intel Z790 and AMD X670(E) chipset models already have it enabled, which wasn’t the case before, and the PCIe settings required ReBAR to be enabled manually. So those who don’t turn it off will be running with ReBAR active, which is a good thing from a gaming perspective where it adds performance. This is perhaps to some extent because Intel graphics cards without ReBAR don’t seem to behave correctly, and there will probably be more and more graphics cards that count on it in the future. You already know the number two reason for ReBAR-only tests.

Finally, it is also true that testing all tests twice (with and without ReBAR) with triple repeatability is extremely time consuming. However, it is still true what we have argued many times – a platform with ReBAR is less stable when it comes to measurement results. Over time, some things may change in the debugging process (from driver to driver) and may not “make sense” when compared to each other. So when you see somewhere that in other tests a slower card outperforms a more powerful one in some particular case, remember these words.
The disadvantage of measurements with active ReBAR is, in short, that all comparison tests may not always be perfectly consistent. And it is possible that there will continue to be cases where ReBAR reduces performance rather than adding to it. These are things to be reckoned with when studying results. This applies not only to our tests, but to the tests of all the others who do not retest all the older models in comparison with every new graphics card tested.

Methodology: how we measure power draw

We have been tuning the method of measuring power draw for quite a long time and we will also be tuning it for some time. But we already have gimmicks that we can work with happily.

To get the exact value of the total power draw of the graphics card, it is necessary to map the internal power draw on the PCI Express slot and the external one on the additional power supply. For the analysis of the PCIe slot, it was necessary to construct an in-between card on which the power draw measurement takes place. Its basis is resistors calibrated to the exact value (0.1 Ω) and according to the amount of their voltage drop we can calculate the current. We then substitute it into the formula for the corresponding value of the output voltage ~ 12 V and ~ 3.3 V. The voltage drop is so low that it doesn’t make the VRM of the graphics card unstable and the output is still more than 12/3.3 V.

We measure power consumption on the card between the graphics card and the PCI Express slot. Rado Kopera took care of the design and implementation (thank you!)

We are also working on a similar device for external power supply. However, significantly higher currents are achieved there, longer cabling and more passages between connectors are necessary, which means that the voltage drop will have to be read on an even smaller resistance of 0.01 Ω, the current state (with 0.1 Ω) is unstable for now. Until we fine-tune it, we will use Prova 15 current clamp for cable measurements, which also measures with good accuracy, they just have a range of up to 30 A. But that is also enough for the OC version of the RTX 3090 Gaming X Trio. If a card is over the range, it is always possible to split the consumption measurement (first into one half and then into the other half of the 12 V conductors).

And why bother with such devices at all when Nvidia has a PCAT power draw analyzer? For complete control over the measurements. While our devices are transparent, the Nvidia’s tool uses the processor that can (but of course does not have to) affect the measurements. After testing the AMD graphics card on the Nvidia’s tool, we probably wouldn’t sleep well.

To read and record measurements, we use a properly calibrated multimeter Keysight U1231A, which exports samples to XLS. From it we obtain the average value and by substituting into the formula with the exact value of the subcircuit output voltages we obtain the data for the graphs.

We will analyze the line graphs with the waveforms for each part of the power supply separately. Although the 3.3 V value is usually negligible, it needs to be monitored. It is difficult to say what exactly this subcircuit powers, but usually the consumption on it is constant and when it changes only with regard to whether a static or dynamic image is rendered. We measure consumption in two sort of demanding games (F1 2020 and Shadow of the Tomb Raider) and one less demanding one (CS:GO) with the highest graphic details preset and UHD resolution (3840 × 2560 px). Then in 3D rendering in Blender using the Cycles renderer on the famous Classroom scene. However, in addition to high-load tests, it’s important to know your web browser consumption (which, in our case, is accelerated Google Chrome), where we also spend a lot of time watching videos or browsing the web. The usual average load of this type is represented by the FishIE Tank (HTML5) website with 20 fish and the web video in our power draw tests is represented by a sample with the VP9 codec, data rate of 17.4 mb/s and 60 fps. In contrast, we also test offline video consumption, in VLC player on a 45 HEVC sample (45.7 mb/s, 50 fps). Finally, we also record the power consumption of the graphics card on the desktop of idle Windows 10 with one or two active UHD@60 Hz monitors.

Noise measurement…

Noise, as well as other operating characteristics, which we will focus on, we’re measuring in the same modes as consumption, so that the individual values overlap nicely. In addition to the level of noise produced, we also record the frequency response of the sound, the course of the GPU clock speed and its temperature.

In this part of the methodology description, we will present something about the method of noise measurement. We use a Reed R8080 sound level meter, which we continuously calibrate with a calibrated Voltcraft SLC-100 digital sound level meter. A small addition to the sound level meter is a parabola-shaped collar, which has two functions. Increases the sensitivity to distinguish the sound produced even at very low speeds. It is thus possible to better compare even very quiet cards with the largest possible ratio difference. Otherwise (without this adjustment) it could simply happen that we measured the same noise level across several graphics cards, even though it would actually be a little different. This parabolic shield also makes sense because, from the outer convex side (from the back), it reflects all the parasitic sounds that everyone who really aims for accuracy of the measurements struggles with during the test. These are various cracks of the body or objects in the room during normal human activity.

To ensure the same conditions when measuring the noise level (and later also the sound), we use acoustic panels with a foam surface around the bench-wall. This is so that the sound is always reflected to the sound level meter sensor in the same way, regardless of the current situation of the objects in the test room. These panels are from three sides (top, right and left) and their purpose is to soundproof the space in which we measure the noise of graphics cards. Soundproofing means preventing different reflections of sound and oscillations of waves between flat walls. Don’t confuse it with sound-absorbing, we’ve had that solved well in the test lab for a long time.

During the measurements, the sound level meter sensor is always placed on a tripod at the same angle and at the same distance (35 cm) from the PCI Express slot in which the graphics card is installed. Of course, it’s always closer to the card itself, depending on its depth. The indicated reference point and the sensor angles are fixed. In addition to the “aerodynamic noise” of the coolers, we also measure the noise level of whining coils. Then we stop the fans for a moment. And for the sake of completeness, it should be added that during sound measurements, we also switch off the power supply fan as well as the CPU cooler fan. Thus, purely the graphics card is always measured without any distortion by other components.

…and the sound frequency response

From the same place, we also measure the frequency of the sound produced. One thing is the noise level (or sound pressure level in decibels) and the other thing is its frequency response.

According to the data on the noise level, you can quickly find out whether the graphics card is quieter or noisier, or where it is on the scale, but it is still a mix of different frequencies. Thus, it does not say whether the sound produced is more booming (with a lower frequency) or squeaking (with a high frequency). The same 35 dBA can be pleasant but also unpleasant for you under certain circumstances – it depends on each individual how they perceive different frequencies. For this reason, we will also measure the frequency response of the sound graphics card in addition to the noise level, via the TrueRTA application. The results will be interpreted in the form of a spectrograph with a resolution of 1/24 octave and for better comparison with other graphics cards we will include the dominant frequency of lower (20–200 Hz), medium (201–2,000 Hz) and higher (2,001–20,000 Hz) sound spectrum into standard bar graphs. For measurements, we’re using a calibrated miniDSP UMIK-1 microphone, which accurately copies the position of the sound level meter, but also has a collar, even with the same focal length.

At the end of this chapter, it should be noted that measurements of noise and frequency response of sound will be performed on most cards only in load tests, as out of load and at lower load (including video decoding) operation is usually passive with fans turned off. On the other hand, we must also be prepared for exceptions with active operation in idle or graphics cards with dual BIOS setup, from which the more powerful one never turns off the fans and they run at least at minimum speed. Finally, as with measuring the noise level in one of the tests, we also record the frequency response of whining coils. But don’t expect any dramatic differences here. It will usually be one frequency, and the goal is rather to detect any potential anomalies. The sound of the whining coils is of course variable, depending on the scene, but we always measure in the same scene (in CS:GO@1080p).

Methodology: temperature tests

We’re also bringing you temperature tests. You are at HWCooling after all. However, in order to make it sensible at all to monitor temperatures on critical components not only of the graphics card, but anything in the computer, it is important to simulate a real computer case environment with healthy air circulation. The overall behavior of the graphics card as such then follows from this. In many cases, an open bench-table is inappropriate and results can be distorted. Therefore, during all, not only heat tests, but also measurement of consumption or course of graphics core frequencies, we use a wind tunnel with equilibrium flow.

Two Noctua NF-S12A fans are at the inlet and the same number is on the exhaust.When testing various system cooling configurations, this proved to be the most effective solution. The fans are always set to 5 V and the speed corresponds to approx. 550 rpm. The stability of the inlet air is properly controlled during the tests, the temperature being between 21 and 21.3 °C at a humidity of ±40 %.

We read the temperature from the internal sensors via GPU-Z. This small, single-purpose application also allows you to record samples from sensors in a table. From the table, it is then easy to create line graphs with waveforms or the average value into bar graphs. We will not use the thermal camera very much here, as most graphics cards have a backplate, which makes it impossible to measure the PCB heating. The key for the heating graphs will be the temperature reading by internal sensors, according to which, after all, the GPU frequency control also takes place. It will always be the heating of the graphics core, and if the sensors are also on VRAM and VRM, we will extract these values into the article as well.

Test Setup

Patriot Blackout memory (4× 8 GB, 3600 MHz/CL18)

2× Patriot Viper VPN100 SSD (512 GB + 2 TB)

BeQuiet! Dark Power Pro 12 PSU w/ 1200 W


Test configuration
Processor	AMD Ryzen 9 5900X
CPU Cooler	Noctua NH-U14S@12 V s NT-H2
Motherboard	MSI MEG X570 Ace
Memory (RAM)	Patriot Blackout, 4× 8 GB, 3600 MHz/CL18
SSD	2× Patriot Viper VPN100 (512 GB + 2 TB)
PSU	BeQuiet! Dark Power Pro 12 (1200 W)

/* Here you can add custom CSS for the current table */ /* Lean more about CSS: https://en.wikipedia.org/wiki/Cascading_Style_Sheets */ /* To prevent the use of styles to other tables use "#supsystic-table-979" as a base selector for example: #supsystic-table-979 { ... } #supsystic-table-979 tbody { ... } #supsystic-table-979 tbody tr { ... } */

Note: At the time of testing, graphics drivers are AMD Adrenalin 23.4.1 and the OS Windows 10 Enterprise build is 19043.

3DMark

For the tests we’re using 3DMark Professional and the Night Raid (DirectX12) is suitable for comparing weaker GPUs, for more powerful ones there is Fire Strike (DirectX11) and Time Spy (DirectX12).

Age of Empires II: DE

Test platform benchmark, API DirectX 11; graphics setting preset Ultra; no extra settings.

Assassin’s Creed: Valhalla

Test platform benchmark; API DirectX 12; graphics setting preset Ultra High; no extra settings.

Battlefield V

Test platform custom scene (War stories/Under no flag); API DirectX 12, graphics setting preset Ultra; TAA high; no extra settings.

Battlefield V with DXR

Test platform custom scene (War stories/Under no flag); API DirectX 12, graphics setting preset Ultra; TAA high; extra settings DXR.

Note: The game also supports DLSS, but as it’s an older title and there are many tests, we will not address it in standard tests. However, measurements on request are possible if you ask for it.

Borderlands 3

Test platform benchmark; API DirectX 12, graphics setting preset Ultra; TAA; no extra settings.

Control

Test platform custom scene (Polaris chapter); API DirectX 11, graphics setting preset High; no extra settings.

Control with DXR

Test platform custom scene (Polaris chapter); API DirectX 12, graphics setting preset High; extra settings DXR.

DXR (native)

Counter-Strike: GO

Test platform benchmark (Dust 2 map flyover); API DirectX 9, graphics setting preset High; 4× MSAA; no extra settings.

Cyberpunk 2077

Test platform custom scene (Little China); API DirectX 12, graphics setting preset Ultra; no extra settings.

Cyberpunk 2077 with DXR

Test platform custom scene (Little China); API DirectX 12, graphics setting preset Ultra; extra settings Ray Tracing on (Ultra).

DXR

DOOM Eternal

Test platform custom scene; API Vulkan, graphics setting preset Ultra Nightmare; no extra settings.

F1 2020

Test platform benchmark (Australia, Clear/Dry, Cycle); API DirectX 12, graphics setting preset Ultra High; TAA; extra settings Skidmarks blending off*.

*on GeForce graphics cards, the Skidmarks blending option is disabled. This option is missing on AMD graphics cards. However, the overall quality of Skidmarks is otherwise set to High on both GeForce and AMD.

Note: The game also supports DLSS 2.0 and FidelityFX for upscaling and sharpening, but due to the relatively low hardware requirements in the native settings, we will not address them in standard tests. However, measurements on request are possible if you ask for it.

FIFA 21

Test platform custom scene (Autumn/Fall, Overcast, 9pm, Old Trafford); API DirectX 12, graphics setting preset Ultra; no extra settings.

Forza Horizon 4

Test platform custom scene; API DirectX 12, graphics setting preset Ultra; 2× MSAA; no extra settings.

Mafia: DE

Test platform custom scene (from the Salieri’s Bar parking lot to the elevated railway gate); API DirectX 11, graphics setting preset High; no extra settings.

Metro Exodus

Test platform benchmark; API DirectX 12, graphics setting preset Extreme; no extra settings.

Metro Exodus with DXR

Test platform benchmark; API DirectX 12, graphics setting preset Ultra; extra settings DXR.

DXR (native)

Microsoft Flight Simulator

Disclaimer: We do not use the results from this game to calculate average game performance. This is because updates often change the game’s performance and when they do, we start building the results database from scratch. To check the consistency of the MFS results, we run a test scene with the MSI RTX 3080 Gaming X Trio before testing each new graphics card.

Test platform custom scene (Paris-Charles de Gaulle, Air Traffic: AI, February 14, 9:00) autopilot:from 1000 until hitting the terrain; API DirectX 11, graphics setting preset Ultra; TAA; no extra settings.

Red Dead Redemption 2 (Vulkan)

Test platform custom scene; API Vulkan, graphics setting preset Favor Quality; no extra settings.

Why is the result missing for some graphics cards? At some point, both AMD and Nvidia stopped supporting Vulkan for Red Dead Redemtion 2. Because of this, it is not possible to test under this API, although it is still in the game’s settings. But don’t turn it on, you won’t be able to launch the game afterwards and you’ll need to manually override Vulkan to Dx12 in the system.xml file.

Red Dead Redemption 2 (Dx12)

Test platform custom scene; API DirectX 12, graphics setting preset Favor Quality; no extra settings.

Shadow of the Tomb Raider

Test platform custom scene; API DirectX 12, graphics setting preset Highest; TAA; no extra settings.

Shadow of the Tomb Raider with DXR

Test platform benchmark; API DirectX 12, graphics setting preset Highest; extra settings DXR.

Note: This game also supports DLSS and FidelityFX CAS, but since this is an older title and there are more than enough tests, we will not address this setting in standard tests. However, testing on request is possible if you ask for it.

Total War Saga: Troy

Test platform benchmark; API DirectX 11, graphics setting preset Ultra; 4× AA, no extra settings.

Wasteland 3

Test platform custom scene; API DirectX 11, graphics setting preset Ultra; no extra settings.

Overall gaming performance

We calculate the average performance so that each game has an equal weight on the result. You can find out exactly how we arrive at the result in this article.

Disclaimer: The charts below present averages from games with settings where only the graphics generated by rasterization appear. This selection was originally intended to allow the inclusion of graphics cards without ray-tracing support. However, it is also reasonable to consider speed with ray-tracing graphics as well. Therefore, prospectively, we will incorporate additional sets of charts into the tests, where one will also take into account graphics details with ray tracing and the other will be exclusively for gaming setups under DXR with RT.

Performance per euro

CompuBench 2.0 (OpenCL)

Test platform benchmark; API OpenCL; no extra settings.

Game Effects

Advanced Compute

High Quality Computer Generated Imagery and Rendering

Computer Vision

SPECviewperf 2020

Test platform benchmark; API OpenGL a DirectX; no extra settings.

SPECworkstation 3

FLOPS, IOPS and memory speed tests

Test platform benchmark; app version 6.32.5600; no extra settings.

LuxMark

Test platform benchmark; API OpenCL; no extra settings.

Blender@Cycles

Test platform render BMW and Classroom; renderer Cycles, 12 tiles; extra settings OpenCL for Radeon graphics cards and CUDA for GeForce. The way most people will use it. OpenCL with GeForce is always slow because path tracing doesn’t support GPU acceleration and is computed by the CPU. Nvidia OptiX is tested separately on supported cards (GeForce RTX) and we put the results in a separate chart.

Blender@Radeon ProRender

Test platform render BMW and Classroom; renderer Radeon ProRender, 1024 samples; no extra settings. Extra settings are OpenCL for Radeon graphics cards and CUDA for GeForce. Nvidia OptiX is tested separately on supported cards (GeForce RTX) and we put the results in a separate chart.

Blender@Eevee

Test platform animation render Ember Forest; renderer Eevee, 350 images; extra settings OpenCL.

Photo editing

Adobe Photoshop: Test platform PugetBench; no extra settings.

Affinity Photo: Test platform built-in benchmark; no extra settings.

Adobe Lightroom: Test platform: custom 1-gigabyte archive of 42 RAW photos (CR2) from a DSLR; no extra settings.

Broadcasting

OBS Studio and XSplit: Test platform F1 2020 game benchmark; extra nastavenia enabled encoders AMD VCE/Nvidia Nvenc (AVC/H.264), output resolution 2560 × 1440 px (60 fps), target bitrate 19,700 kbps.

Password cracking

Test platform Hashcat; no extra settings. You can easily try the tests yourself. Just download the binary and enter the cipher you are interested in using the numeric code on the command line.

Why are results missing for some GeForce graphics cards? Nvidia’s newer 5xx.xx gaming and studio drivers no longer support Hashcat. Although it can be run via the command line and a test for a specific cipher can be entered, but no action, calculation, comes after confirmation. We will investigate the reasons and possible solutions in more detail over time, but it probably can’t be done without a fix at the graphics driver level.

GPU clock speed

GPU temperatures

VRAM temperature

Note: If the measured value is missing for a selected graphics card, it means that it cannot be detected by the internal sensor.

Net graphics card power draw

Performance per watt

Analysis of 12 V branch power supply (higher load)

Analysis of 12 V branch power supply (lower load)

Analysis of 3.3 V branch power supply

Noise level

Frequency response of sound

Measurements are performed in the TrueRTA application, which records sound in a range of 240 frequencies in the recorded range of 20–20,000 Hz. For the possibility of comparison across articles, we export the dominant frequency from the low (20–200 Hz), medium (201–2,000 Hz) and high (2,001–20,000 Hz) range to standard bar graphs.

However, for an even more detailed analysis of the sound expression, it is important to perceive the overall shape of the graph and the intensity of all frequencies/tones. If you don’t understand something in the graphs or tables below, you’ll find the answers to all your questions in this article. This explains how to read the measured data below correctly.

The microphone we use to analyze the sound of coolers and coils


Graphics card		Dominant sound freq. and noise level in F1 2020@2160p	NF-F12 PWM		NF-A15 PWM
		Low range		Mid range		High range
		Frequency [Hz]	Noise level [dBu]	Frequency [Hz]	Noise level [dBu]	Frequency [Hz]	Noise level [dBu]
Sapphire RX 7900 XT Pulse, ReBAR on		100,8	-75,1	213,6	-70,9	6450,8	-82,4
MSI RTX 4070 Ti Suprim X 12G (G), ReBAR on		97,9	-77,0	1107,9	-74,1	5747,0	-81,8
MSI RTX 4080 16GB Suprim X (G), ReBAR on		71,3	-77,2	1076,3	-69,3	11830,8	-75,4
MSI RTX 3050 Ventus 2X 8G OC, ReBAR off		138,5	-78,8	1107,9	-78,6	2031,9	-84,8
MSI RTX 3050 Ventus 2X 8G OC, ReBAR on		123,4	-81,2	1107,9	-80,0	18245,6	-83,7
Sapphire RX 6650 XT Nitro+ (P), ReBAR on		50,4	-83,3	1107,9	-72,4	7240,8	-82,5
Sapphire RX 6650 XT Nitro+ (P), ReBAR off		184,9	-82,3	1107,9	-71,4	6834,4	-82,2
Sapphire RX 6600 XT Pulse, ReBAR on	Sapphire RX 6600 XT Pulse, ReBAR on	100,8	-71,8	1356,1	-72,7	6088,7	-80,9
Sapphire RX 6600 XT Pulse, ReBAR off	Sapphire RX 6600 XT Pulse, ReBAR off	100,8	-71,8	219,8	-74,5	6088,7	-81,0
Aorus RTX 3080 Xtreme 10G (OC), ReBAR on		50,4	-77,0	1076,3	-56,5	2031,9	-69,4
Aorus RTX 3080 Xtreme 10G (OC), ReBAR off		50,4	-75,9	1076,3	-56,7	2031,9	-69,6
Sapphire RX 6900 XT Toxic LE (P), ReBAR on	Sapphire RX 6900 XT Toxic LE (P), ReBAR on	138,5	-62,4	1107,9	-56,6	11166,8	-74,7
Sapphire RX 6900 XT Toxic LE (P), ReBAR off	Sapphire RX 6900 XT Toxic LE (P), ReBAR on	138,5	-61,9	1107,9	-55,6	5747,0	-74,7
Sapphire RX 6700 XT Nitro+ (P), ReBAR on		100,8	-73,2	1076,3	-71,2	7034,6	-76,5
Sapphire RX 6700 XT Nitro+ (P), ReBAR off		100,8	-75,2	1076,3	-73,5	7034,6	-76,5
MSI RTX 3060 Ti Gaming X Trio, ReBAR off		100,8	-70,6	1107,9	-82,8	7034,6	-83,7
Gigabyte RTX 3060 Eagle OC 12G, ReBAR off		100,8	-71,6	213,6	-64,3	2031,9	-74,2
MSI RTX 3090 Gaming X Trio, ReBAR off		100,8	-72,3	1076,3	-76,0	4561,4	-81,2
MSI RTX 3070 Gaming X Trio, ReBAR off		100,8	-73,9	1076,3	-79,7	6267,2	-85,1
AMD Radeon RX 6800, ReBAR on		100,8	-71,0	1076,3	-66,5	9665,3	-81,3
AMD Radeon RX 6800, ReBAR off		100,8	-71,8	1107,9	-67,4	2091,4	-75,3
TUF RTX 3080 O10G Gaming, ReBAR off		100,8	-76,0	1107,9	-77,9	7034,6	-74,4
AMD Radeon RX 6800 XT, ReBAR on		100,8	-71,6	1107,9	-74,7	10848,9	-76,3
AMD Radeon RX 6800 XT, ReBAR off		100,8	-73,0	1107,9	-74,7	10848,9	-76,5

/* Here you can add custom CSS for the current table */ /* Lean more about CSS: https://en.wikipedia.org/wiki/Cascading_Style_Sheets */ /* To prevent the use of styles to other tables use "#supsystic-table-2285" as a base selector for example: #supsystic-table-2285 { ... } #supsystic-table-2285 tbody { ... } #supsystic-table-2285 tbody tr { ... } */


Graphics card		Dominant sound freq. and noise level in SOTTR@2160p	NF-F12 PWM		NF-A15 PWM
		Low range		Mid range		High range
		Frequency [Hz]	Noise level [dBu]	Frequency [Hz]	Noise level [dBu]	Frequency [Hz]	Noise level [dBu]
Sapphire RX 7900 XT Pulse, ReBAR on		100,8	-74,4	213,6	-70,8	5915,4	-79,7
MSI RTX 4070 Ti Suprim X 12G (G), ReBAR on		100,8	-77,9	1076,3	-77,9	5583,4	-82,2
MSI RTX 4080 16GB Suprim X (G), ReBAR on		190,3	-78,4	1045,7	-74,4	11830,8	-75,3
MSI RTX 3050 Ventus 2X 8G OC, ReBAR off		138,5	-78,4	1140,4	-78,5	2031,9	-84,7
MSI RTX 3050 Ventus 2X 8G OC, ReBAR on		138,5	-78,3	1107,9	-78,4	2031,9	-84,6
Sapphire RX 6650 XT Nitro+ (P), ReBAR on		50,4	-80,1	1107,9	-76,5	6834,4	-84,3
Sapphire RX 6650 XT Nitro+ (P), ReBAR off		47,6	-84,3	1107,9	-75,9	6834,4	-82,8
Sapphire RX 6600 XT Pulse, ReBAR on		100,8	-68,8	1356,1	-75,7	6088,7	-82,6
Sapphire RX 6600 XT Pulse, ReBAR off		100,8	-69,5	1356,1	-74,8	5915,4	-83,1
Aorus RTX 3080 Xtreme 10G (OC), ReBAR on		44,9	-73,0	1045,7	-50,3	2031,9	-60,5
Aorus RTX 3080 Xtreme 10G (OC), ReBAR off		41,8	-72,6	1076,3	-51,4	2031,9	-60,7
Sapphire RX 6900 XT Toxic LE (P), ReBAR on	Sapphire RX 6900 XT Toxic LE (P), ReBAR on	138,5	-63,1	1140,4	-57,9	5747,0	-74,7
Sapphire RX 6900 XT Toxic LE (P), ReBAR off	Sapphire RX 6900 XT Toxic LE (P), ReBAR off	134,5	-61,7	1107,9	-58,6	5747,0	-74,2
Sapphire RX 6700 XT Nitro+ (P), ReBAR on		100,8	-73,9	1140,4	-75,4	5915,4	-77,2
Sapphire RX 6700 XT Nitro+ (P), ReBAR off		100,8	-75,1	1107,9	-75,2	5915,4	-76,5
MSI RTX 3060 Ti Gaming X Trio, ReBAR off		100,8	-70,8	1076,3	-83,6	7034,6	-81,9
Gigabyte RTX 3060 Eagle OC 12G, ReBAR off		100,8	-71,9	213,6	-64,5	2031,9	-73,8
MSI RTX 3090 Gaming X Trio, ReBAR off		106,8	-74,5	213,6	-71,3	4561,4	-79,3
MSI RTX 3070 Gaming X Trio, ReBAR off		100,8	-73,0	213,6	-72,3	6267,2	-84,9
AMD Radeon RX 6800, ReBAR on		100,8	-71,8	1140,4	-66,1	9948,5	-81,3
AMD Radeon RX 6800, ReBAR off		100,8	-71,6	1140,4	-67,8	9665,3	-80,6
TUF RTX 3080 O10G Gaming, ReBAR off		100,8	-75,4	1076,3	-72,3	7240,8	-74,2
AMD Radeon RX 6800 XT, ReBAR on		100,8	-73,2	1107,9	-73,9	10848,9	-76,3
AMD Radeon RX 6800 XT, ReBAR off		100,8	-73,2	1107,9	-75,3	10848,9	-75,4

/* Here you can add custom CSS for the current table */ /* Lean more about CSS: https://en.wikipedia.org/wiki/Cascading_Style_Sheets */ /* To prevent the use of styles to other tables use "#supsystic-table-2286" as a base selector for example: #supsystic-table-2286 { ... } #supsystic-table-2286 tbody { ... } #supsystic-table-2286 tbody tr { ... } */


Graphics card		Dominant sound freq. and noise level in CS:GO@2160p	NF-F12 PWM		NF-A15 PWM
		Low range		Mid range		High range
		Frequency [Hz]	Noise level [dBu]	Frequency [Hz]	Noise level [-dBu]	Frequency [Hz]	Noise level [-dBu]
Sapphire RX 7900 XT Pulse, ReBAR on		100,8	73,7	213,6	-71,4	5747,0	-76,3
MSI RTX 4070 Ti Suprim X 12G (G), ReBAR on		100,8	-77,9	207,5	-81,6	4305,4	-83,5
MSI RTX 4080 16GB Suprim X (G), ReBAR on		100,8	-79,0	1076,3	-72,2	11830,8	-76,4
MSI RTX 3050 Ventus 2X 8G OC, ReBAR off		138,5	-79,8	1107,9	-77,6	2031,9	-83,4
MSI RTX 3050 Ventus 2X 8G OC, ReBAR on		123,4	-81,0	1107,9	-77,8	2031,9	-83,6
Sapphire RX 6650 XT Nitro+ (P), ReBAR on		50,4	-79,6	1107,9	-74,2	7240,8	-80,8
Sapphire RX 6650 XT Nitro+ (P), ReBAR off		49,0	-84,3	1107,9	-80,0	6834,4	-80,2
Sapphire RX 6600 XT Pulse, ReBAR on		100,8	-68,7	1356,1	-74,7	6088,7	-80,8
Sapphire RX 6600 XT Pulse, ReBAR off		100,8	-69,3	1356,1	-75,1	6088,7	-79,2
Aorus RTX 3080 Xtreme 10G (OC), ReBAR on		47,6	-67,1	1045,7	-49,6	2031,9	-60,1
Aorus RTX 3080 Xtreme 10G (OC), ReBAR off		47,6	-70,3	1140,4	-50,8	2031,9	-60,2
Sapphire RX 6900 XT Toxic LE (P), ReBAR on	Sapphire RX 6900 XT Toxic LE (P), ReBAR on	138,5	-64,1	1107,9	-60,1	8610,8	-70,9
Sapphire RX 6900 XT Toxic LE (P), ReBAR off	Sapphire RX 6900 XT Toxic LE (P), ReBAR off	134,5	-71,6	1107,9	-66,4	8365,6	-72,1
Sapphire RX 6700 XT Nitro+ (P), ReBAR on		100,8	-72,6	1173,8	-74,9	5915,4	-74,6
Sapphire RX 6700 XT Nitro+ (P), ReBAR off		100,8	-75,0	1107,9	-73,8	5747,0	-74,2
MSI RTX 3060 Ti Gaming X Trio, ReBAR off		100,8	-71,4	1107,9	-83,1	6267,2	-82,5
Gigabyte RTX 3060 Eagle OC 12G, ReBAR off		100,8	-72,6	213,6	-64,8	2031,9	-73,8
MSI RTX 3090 Gaming X Trio, ReBAR off		106,8	-75,7	213,6	-73,4	4695,1	-77,6
MSI RTX 3070 Gaming X Trio, ReBAR off		106,8	-75,7	213,6	-73,4	6267,2	-82,7
AMD Radeon RX 6800, ReBAR on		100,8	-71,2	1107,9	-66,2	9948,5	-77,4
AMD Radeon RX 6800, ReBAR off		100,8	-71,1	1076,3	-77,3	9665,3	-77,7
TUF RTX 3080 O10G Gaming, ReBAR off		100,8	-74,2	1076,3	-70,9	7240,8	-74,4
AMD Radeon RX 6800 XT, ReBAR on		100,8	-73,0	1107,9	-74,3	7671,3	-72,4
AMD Radeon RX 6800 XT, ReBAR off		100,8	-72,3	1107,9	-73,7	10848,9	-72,5

/* Here you can add custom CSS for the current table */ /* Lean more about CSS: https://en.wikipedia.org/wiki/Cascading_Style_Sheets */ /* To prevent the use of styles to other tables use "#supsystic-table-2287" as a base selector for example: #supsystic-table-2287 { ... } #supsystic-table-2287 tbody { ... } #supsystic-table-2287 tbody tr { ... } */


Graphics card	Dominant sound freq. and noise level in Blender (Cycles), Classroom	NF-F12 PWM		NF-A15 PWM
	Low range		Mid range		High range
	Frequency [Hz]	Noise level [dBu]	Frequency [Hz]	Noise level [dBu]	Frequency [Hz]	Noise level [dBu]
Sapphire RX 7900 XT Pulse, ReBAR on	100,8	-73,8	207,5	-74,1	7034,6	-85,2
MSI RTX 4070 Ti Suprim X 12G (G), ReBAR on	100,8	-77,8	207,5	-82,3	5747,0	-89,5
MSI RTX 4080 16GB Suprim X (G), ReBAR on	100,8	-80,3	1140,4	-86,7	11830,8	-89,4
MSI RTX 3050 Ventus 2X 8G OC, ReBAR off	123,4	-79,3	213,6	-81,4	18245,6	-85,5
MSI RTX 3050 Ventus 2X 8G OC, ReBAR on	123,4	-79,5	213,6	-81,5	18245,6	-85,6
Sapphire RX 6650 XT Nitro+ (P), ReBAR on	50,4	-77,9	1107,9	-83,5	7240,8	-87,3
Sapphire RX 6650 XT Nitro+ (P), ReBAR off	50,4	-79,1	1107,9	-83,9	7240,8	-87,5
Sapphire RX 6600 XT Pulse, ReBAR on	100,8	-70,1	1356,1	-73,4	5583,4	-86,1
Sapphire RX 6600 XT Pulse, ReBAR off	100,8	-69,8	1356,1	-73,7	5915,4	-86,0
Asus GT 1030 SL 2G BRK, ReBAR off	50,397	-71,7	1107,9	-94,9	19330,5	-90,5
Aorus RTX 3080 Xtreme 10G (OC), ReBAR on	50,4	-76,4	1107,9	-57,9	2031,9	-69,7
Aorus RTX 3080 Xtreme 10G (OC), ReBAR off	50,4	-78,7	1076,3	-60,9	5424,5	-74,0
Sapphire RX 6900 XT Toxic LE (P), ReBAR on	116,5	-65,0	1107,9	-68,5	5120,0	-77,3
Sapphire RX 6900 XT Toxic LE (P), ReBAR off	116,5	-65,1	1107,9	-68,4	5120,0	-77,1
Sapphire RX 6700 XT Nitro+ (P), ReBAR on	100,8	-72,6	1173,8	-86,6	5915,4	-82,4
Sapphire RX 6700 XT Nitro+ (P), ReBAR off	100,8	-75,8	1076,3	-87,2	5915,4	-82,1
MSI RTX 3060 Ti Gaming X Trio, ReBAR off	100,8	-70,4	987,0	-89,5	6450,8	-89,0
Gigabyte RTX 3060 Eagle OC 12G, ReBAR off	100,8	-72,6	213,6	-70,0	2031,9	-79,1
MSI RTX 3090 Gaming X Trio, ReBAR off	100,8	-71,2	1076,3	-85,3	5915,4	-92,0
MSI RTX 3070 Gaming X Trio, ReBAR off	100,8	-71,2	1076,3	-85,3	18245,6	-90,8
AMD Radeon RX 6800, ReBAR on	100,8	-71,9	987,0	-89,2	7452,9	-88,3
AMD Radeon RX 6800, ReBAR off	100,8	-71,1	987,0	-89,0	7452,9	-88,2
TUF RTX 3080 O10G Gaming, ReBAR off	106,8	-81,5	1660,0	-80,6	6834,4	-78,0
AMD Radeon RX 6800 XT, ReBAR on	97,9	-79,8	1208,2	-89,6	7671,3	-85,2
AMD Radeon RX 6800 XT, ReBAR off	100,8	-73,0	1243,6	-95,2	7671,3	-85,0

/* Here you can add custom CSS for the current table */ /* Lean more about CSS: https://en.wikipedia.org/wiki/Cascading_Style_Sheets */ /* To prevent the use of styles to other tables use "#supsystic-table-2288" as a base selector for example: #supsystic-table-2288 { ... } #supsystic-table-2288 tbody { ... } #supsystic-table-2288 tbody tr { ... } */


Graphics card		Dominant sound freq. and noise level in CS:GO@1080p (coils only*)	NF-F12 PWM		NF-A15 PWM
		Low range		Mid range		High range
		Frequency [Hz]	Noise level [dBu]	Frequency [Hz]	Noise level [dBu]	Frequency [Hz]	Noise level [dBu]
Sapphire RX 7900 XT Pulse, ReBAR on		100,8	-73,2	1076,3	-81,5	5747,0	-76,3
MSI RTX 4070 Ti Suprim X 12G (G), ReBAR on		100,8	-78,0	987,0	-78,6	5583,4	-84,3
MSI RTX 4080 16GB Suprim X (G), ReBAR on		100,8	-80,5	1140,4	-71,8	11830,8	-74,9
MSI RTX 3050 Ventus 2X 8G OC, ReBAR off		50,4	-83,6	1317,5	-83,0	7896,1	-83,9
MSI RTX 3050 Ventus 2X 8G OC, ReBAR on		50,4	-77,7	1317,5	-87,3	10848,9	-84,5
Sapphire RX 6650 XT Nitro+ (P), ReBAR on		50,4	-81,8	1045,7	-84,2	2091,4	-77,7
Sapphire RX 6650 XT Nitro+ (P), ReBAR off		50,4	-83,3	1974,0	-90,0	7034,6	-82,4
Sapphire RX 6600 XT Pulse, ReBAR on		100,8	-72,0	1107,9	-83,7	2215,8	-79,6
Sapphire RX 6600 XT Pulse, ReBAR off		100,8	-68,4	1917,8	-88,7	6450,8	-81,4
Asus GT 1030 SL 2G BRK, ReBAR off		50,4	-71,1	1107,9	-91,7	12534,3	-89,8
Aorus RTX 3080 Xtreme 10G (OC), ReBAR on		50,4	-80,6	1660,0	-80,3	7896,1	-80,2
Aorus RTX 3080 Xtreme 10G (OC), ReBAR off		50,4	-78,8	1660,0	-82,6	7671,3	-80,4
Sapphire RX 6900 XT Toxic LE (P), ReBAR on	Sapphire RX 6900 XT Toxic LE (P), ReBAR on	100,8	-74,9	739,4	-67,9	5915,4	-78,5
Sapphire RX 6900 XT Toxic LE (P), ReBAR off	Sapphire RX 6900 XT Toxic LE (P), ReBAR off	50,4	-81,4	739,4	-70,2	8610,8	-73,6
Sapphire RX 6700 XT Nitro+ (P), ReBAR on		100,8	-74,6	987,0	-84,8	5747,0	-69,6
Sapphire RX 6700 XT Nitro+ (P), ReBAR off		100,8	-74,7	1395,9	-88,4	5747,0	-70,3
MSI RTX 3060 Ti Gaming X Trio, ReBAR off		100,8	-73,0	1974,0	-88,1	6267,2	-83,6
Gigabyte RTX 3060 Eagle OC 12G, ReBAR off		100,8	-73,6	1974,0	-90,2	6088,7	-83,1
MSI RTX 3090 Gaming X Trio, ReBAR off		50,4	-76,1	987,0	-84,8	5915,4	-83,3
MSI RTX 3070 Gaming X Trio, ReBAR off		100,8	-74,7	1317,5	-81,4	6088,7	-84,6
AMD Radeon RX 6800, ReBAR on		100,8	-71,8	987,0	-87,7	7452,9	-80,4
AMD Radeon RX 6800, ReBAR off		100,8	-72,0	1660,0	-90,4	8863,1	-84,5
TUF RTX 3080 O10G Gaming, ReBAR off		100,8	-75,6	1140,4	-81,7	9948,5	-78,7
AMD Radeon RX 6800 XT, ReBAR on		100,8	-73,6	1660,0	-79,8	7452,9	-74,0
AMD Radeon RX 6800 XT, ReBAR off		100,8	-73,3	1660,0	-83,3	7452,9	-76,4

/* Here you can add custom CSS for the current table */ /* Lean more about CSS: https://en.wikipedia.org/wiki/Cascading_Style_Sheets */ /* To prevent the use of styles to other tables use "#supsystic-table-2289" as a base selector for example: #supsystic-table-2289 { ... } #supsystic-table-2289 tbody { ... } #supsystic-table-2289 tbody tr { ... } */

*The Sapphire RX 6900 XT Toxic LE is the only tested graphics card whose spectral analysis also includes the sound of the water pump.

If we had only two variables – rasterization speed for games and price, the Radeon RX 7900 XT has no competition in its price range. The latter only appears, speed-wise, in the significantly more expensive GeForce RTX 4080, for which Nvidia bet on different customer requirements. But we’ll get to where Radeon loses out significantly to GeForce later. Aside from settings applying ray-tracing units, the RX 7900 XT beats the RTX 4070 Ti, a rival graphics card, by a class. The difference in average gaming performance in favour of the Radeon (RX 7900 XT) is 15–19%. The lower number (15%) refers to UHD/2160p resolution, which these graphics cards are usually fine for. Unless you require RT settings and a high-speed monitor. The RTX 4070 Ti is the “weaker” option even in titles that are better optimized on GeForce (including Control).

Of the 18 games tested, the only one where the RTX 4070 Ti has the edge over the RX 7900 XT is DOOM Eternal. And that’s only in high resolution of 2160p. In Assassin’s Creed: Valhalla and Borderlands 3, the RX 7900 XT’s edge is as much as 26–27%, in CS:GO as much as 40 %. It’s not usually this abysmal, and in most games the RTX 4070 Ti’s deficit is under 10% (F1 2020, Forza Horizon 4, Mafia: DE, Metro Exodus, Shadow of the Tomb Raider and so on), but it often gravitates more towards comparisons with the RTX 4080. Even that one lags behind in UHD in games like Red Dead Redemption 2, CS:GO or FIFA, but is usually faster on average, as illustrated by the overall gaming performance charts. But notice one thing in them, namely that while in 2160p the RX 7900 XT loses to the RTX 4080 by 7%, in QHD/1440p the results are already even and In FHD/1080p, Radeon breaks into the lead.

How is it possible that the RTX 4080 is faster in high resolutions where only the GPU is the limit, but in FullHD the order flips and suddenly the RX 7900 XT is better despite the fact that the CPU limit should knock both cards to the same level at most? The graphics drivers of the Nvidia Ampere and Ada Lovelace architectures have more overhead according to various observations. That is, they consume slightly more CPU cycles for each frame in the drivers than AMD graphics card drivers with RDNA 2 and 3 architectures. If the game is CPU-limited , then there may be a situation where Radeon can achieve higher fps than GeForce because the number of CPU cycles is finite and an AMD driver with this allocation can handle more frames than an Nvidia driver. And even if the Nvidia GPUs themselves need less CPU time than the AMD GPUs (which shows up in 2160p resolution), this doesn’t show up in a CPU bottleneck situation – it’s how many frames the CPU can handle per second that’s critical. So much for a small digression on why, across all games, the RX 7900 XT is the faster card in FHD than the RTX 4080, although in UHD this is no longer the case.

For owners of high-speed monitors, the Radeon RX 7900 XT may be a more attractive option than the RTX 4080. Of course still assuming that gaming performance with ray-tracing graphics is not taken into account. The latter is significantly weaker with the RX 7900 XT even compared to the RTX 4070 Ti. Only in Shadow of the Tomb Raider are the results balanced, with a relatively small amount of RT workload. Where the ray tracing application is more complex, the RX 7900 XT is already roughly 25–35% slower compared to the RTX 4070 Ti. This is the case in Control, in Cyberpunk 2077 or in Battlefield V. In the original Metro Exodus (so it doesn’t apply to the Enhanced Edition with extremely hard RT) the difference is smaller, but that’s for a similar reason to Shadow of the Tomb Raider – again, there’s not that much ray-tracing.

So if you want ray tracing in your games, the RX 7900 XT’s attractiveness slowly diminishes, even more so when you’re also interested in DLSS 3 for Frame Generation, which only the GeForce RTX 4000s exclusively support. Then there’s also the GPU acceleration of tensor cores in the application environment in Nvidia’s favour. The RX 7900 XT also often lags behind shader acceleration in 3D rendering, in Blender under OpenCL it takes longer to do everything than the RTX 4070 Ti with CUDA. But again, impressive results are in Autodesk applications, where the RX 7900 XT dominates over the GeForce (including the RTX 4080) in most cases.

Although there are scenarios where the RX 7900 XT offers a more attractive price/performance ratio compared to GeForce (RTX 4070 Ti and RTX 4080), the overall Radeon picture is made worse by weaker power efficiency. Not significantly, but the difference of some 15% is there at the limit of maximum performance. The RX 7900 XT Pulse draws up to 325W, which is 5–10% more than the MSI RTX 4080 16GB Suprim X. On the other hand, Sapphire spikes GPU clock speeds quite high. In games it reaches at least 2500 MHz, in CS:GO even up to 2623 MHz. So there is some room for tuning (undervolting and underclocking) to increase efficiency. Due to the relatively high power draw, the RX 7900 XT in the Sapphire Pulse version is somewhere in the middle in the noise measurement charts, also when it comes to the coils. These are noisier, but less so than the RTX 4080 16GB Suprim X.

The Sapphire RX 7900 XT Pulse is a very solid card with no obvious flaws, and moreover, among such powerful graphics cards, has a nice price/performance ratio, without applying RT, more attractive than the RTX 4070 Ti and RTX 4080 have. If you don’t rely on ray tracing and can survive without DLSS 3, the RX 7900 XT Pulse is a great choice. Especially for lower resolutions in combination with monitors with very high image refresh rates.

English translation and edit by Jozef Dudáš


Sapphire RX 7900 XT Pulse
+ Very high performance (also suitable for 2160p/4K gaming)
+ Without ray-tracing graphics settings, it significantly outperforms the RTX 4070 Ti...
+ ... and at lower resolutions it is even faster than the RTX 4080
+ Attractive price/performance ratio for a high-end card
+ Solid cooler. It is not noisy even at this high power draw...
+ ... and yet the temperature is average – GPU temperatures are kept to 65 °C and memory temperatures do not exceed 85 °C
+ In all axes smaller size compared to most RTX 4070 Ti and RTX 4080
+ ... and thus better compatibility with both cases and surrounding PCIe expansion cards
+ Support for AV1 encoding
+ DisplayPort 2.1 support, which GeForces don't have
- Weaker ray-tracing graphics performance. Does not even reach the level of GeForce from the Ampere generation (RTX 3000)
- Slightly worse power efficiency than RTX 4070 Ti and RTX 4080
Odporúčaná koncová cena: 899 EUR

/* Here you can add custom CSS for the current table */ /* Lean more about CSS: https://en.wikipedia.org/wiki/Cascading_Style_Sheets */ /* To prevent the use of styles to other tables use "#supsystic-table-2292" as a base selector for example: #supsystic-table-2292 { ... } #supsystic-table-2292 tbody { ... } #supsystic-table-2292 tbody tr { ... } */

For cooperation in providing the tested hardware, we would like to give special thanks to the Datacomp e-shop

Continue: Age of Empires II: DE