Radeon RX 9070 XT Review
We independently test the products and technologies that we recommend.

1. Features of RDNA 4 Architecture

The new RDNA 4 architecture brings more key changes than we saw in previous updates. AMD and TMSC engineers have abandoned the chiplet layout using separate dies for graphics and memory controllers, as was the case in RDNA 3. Instead, we have a monolithic die called Navi 48, which, in addition to streaming processors, now includes separate tensor accelerators for AI tasks and separate RT-accelerators, similar to NVIDIA. The die area is almost 30% smaller compared to the flagship solutions of the previous generation Radeon.
The RDNA graphics architecture includes an improved memory subsystem, a dual computing unit, a new generation AI Accelerator for matrix computation, and an RT-accelerator. Thanks to architectural and logical improvements, performance gains in some cases were almost 100% compared to the previous architecture. The fact is that ray tracing and acceleration for neural networks are very sensitive to the graphics memory subsystem, so AMD made significant changes to the VRAM management system.
Along with the new graphics cards, Radeon also introduced the fourth version of the AMD FidelityFX Super Resolution complex with an improved image scaling system and a refined frame generator. In terms of image quality, FSR 4 is getting closer and closer to NVIDIA DLSS, but it still cannot generate multiple frames at a time like DLSS 4. According to Radeon representatives, the detail clarity even in "Performance" mode is comparable to "Quality" mode in FSR 3.1 and is almost indistinguishable from native rendering. The performance boost from upscaling can range from 30 to 100% depending on the chosen mode. The only downside is that new tensor cores from RDNA 4 are now required for processing, so FSR 4 cannot be launched on video cards from NVIDIA, Intel, and older Radeon models.
Additionally, a new media processor was added to Navi 48, which on average should increase rendering speed by a quarter when using H.264 codecs. Thus, Radeon attempts to close the gap that separates it from NVIDIA when it comes to comprehensive resource-intensive work.
2. Review of the Asus Radeon RX 9070 XT OC Prime 16GB Graphics Card

Telemart.ua | 37 999 ₴ | To Store |
Comtrading.ua | 36 199 ₴ | To Store |
Artline.ua | 36 999 ₴ | To Store |
MTA.ua | 40 480 ₴ | To Store |
Moyo.ua | 36 199 ₴ | To Store |
The technical specifications of the Radeon RX 9070 XT in the base version without overclocking are as follows: Navi 48 XTX chip (4096 stream processors, 128 tensor cores, 64 RT-accelerators), working frequency 2970 MHz, 16 GB of GDDR6 memory (256-bit bus), 64 MB of Infinity Cache. The new tensor cores will attract the attention of AI specialists: for example, in Stable Diffusion, they promise almost a twofold increase in performance compared to the predecessor — the Radeon RX 7800 XT. However, NVIDIA retains its leadership in this segment, as evidenced by their revenues from AI developments and data centers, which are almost seven times higher than revenues from the gaming market.
It seems AMD analyzed the market, accepted the reality, and decided to focus on the mid-price segment. Radeon RX 9070 XT is not positioned as a flagship and does not claim to compete with the RTX 4090/5090. It's a balanced and optimized graphics card for 2K gaming (and with some compromises for 4K), capable of pushing the ever-increasing cost of pre-top-level GeForce models (RTX 3070/4070/5070) out of the market. In terms of performance, it is comparable to the GeForce RTX 4070 SUPER, but slightly inferior to the Radeon RX 7900 XTX — AMD's 2022 flagship.
The Radeon RX 9070 XT modification from the ASUS Prime series is a classic gaming graphics card with optimal features, affordable pricing, and a minimalist design without unnecessary lighting and decorative elements. It is equipped with a massive heatsink with three axial fans, occupies 2.5 slots (length — 312 mm), consumes about 300 W (a 750 W power supply is recommended), and has a mode switch (quiet/performance).
3. Performance Tests in Games

The testing was conducted on major AAA games of recent years, including classics like Cyberpunk 2077 and Alan Wake 2, as well as newer, more demanding projects on Unreal Engine 5, such as S.T.A.L.K.E.R. 2 and Black Myth: Wukong. The main goal of the test was to measure the pure performance of the model, excluding factors like hardware ray tracing, frame generation, upscaling, and other secondary factors that can significantly affect the final picture.
GeForce RTX 4070 Ti | GeForce RTX 5070 Ti | Radeon RX 9070 XT | |
---|---|---|---|
Black Myth: Wukong | 74 | 87 | 84 |
Starfield | 90 | 104 | 106 |
Cyberpunk 2077 | 94 | 129 | 132 |
Dragon’s Dogma 2 | 98 | 122 | 119 |
Alan Wake 2 | 82 | 94 | 101 |
Elden Ring | 137 | 162 | 153 |
Monster Hunter Wilds | 73 | 85 | 80 |
S.T.A.L.K.E.R. 2: Heart of Chornobyl | 74 | 89 | 93 |
Let’s clarify upfront that this is not the most balanced option for gaming at 1080p, as it turns out to be too expensive and powerful. But for 1440p, this is exactly what the doctor ordered. The card delivers an average of over 80 frames per second in each tested game, including extremely demanding Black Myth: Wukong and S.T.A.L.K.E.R. 2: Heart of Chernobyl. Considering that the tests were conducted at maximum graphics settings with no upscaling or frame generation, it becomes the optimal option for the "1440p + 120 Hz" setup.
GeForce RTX 4070 Ti | GeForce RTX 5070 Ti | Radeon RX 9070 XT | |
---|---|---|---|
Black Myth: Wukong | 41 | 46 | 46 |
Starfield | 57 | 62 | 67 |
Cyberpunk 2077 | 44 | 48 | 46 |
Dragon’s Dogma 2 | 58 | 63 | 66 |
Alan Wake 2 | 47 | 55 | 59 |
Elden Ring | 78 | 86 | 79 |
Monster Hunter Wilds | 48 | 50 | 48 |
S.T.A.L.K.E.R. 2: Heart of Chornobyl | 42 | 47 | 51 |
Meanwhile, the Radeon RX 9070 XT performs decently at 4K, especially in less demanding games of the previous generation, yet its prospects aren’t the brightest. We didn’t have time to include the test data of the recently released The Elder Scrolls IV: Oblivion Remastered and Clair Obscur: Expedition 33, both built on the Unreal Engine 5, but they effectively bring the Radeon RX 9070 XT to its knees in 4K, delivering 20 to 40 frames per second with noticeable stutters and slowdowns.
In general, Radeon presented exactly what it promised — a robust model at the cusp of the mid-to-high price segment, capable of competing on par with the new GeForce RTX 5070 Ti. In native rendering, the “green” model surpasses the review hero by only 2 – 5%, with ray tracing increasing the gap slightly but not significantly (up to 10 – 15%), however, the RX 9070 XT theoretically wins due to its lower price ($599 vs. $749 MSRP).
4. A Few Words About FSR 4

As we mentioned in the RDNA 4 block, AMD presented a new version of the upscaling technology in FSR 4, which uses machine learning for the first time. According to most reviewers, the scaling quality in the new version has made an impressive leap, and the image with FRS and DLSS upscalers is now hardly distinguishable. Especially in 4K, where differences in detail require a magnifying glass to find. Unlike FSR 3.1, the upscaler handles moving objects, water rendering, and reflective surfaces much better. Ghosting during sharp movements is almost absent, hair doesn’t flicker in the dark, and anti-aliasing often looks better than TAA.
As for frame generation, there are no significant improvements. Like the previous version, the Fluid Motions Frame technology generates one intermediate frame for one rendered, thereby making the gameplay smoother and more enjoyable. However, graphics cards on RDNA 4 still cannot generate several artificial frames at once, like the new NVIDIA RTX 50 cards based on Blackwell architecture.
5. Conclusion

The Radeon RX 9070 XT is one of the most impressive graphics card releases since the pandemic. At an official price of just over $600, you get a modern mid-class card with moderate power consumption and a level of performance on par with past years' flagships. In traditional rasterization (without ray tracing), it feels comfortable at 1440p resolution and is quite suitable for 4K, especially if you add a pinch of FSR 4. However, it still lags behind NVIDIA graphics cards in terms of hardware ray tracing and is less optimized for machine learning, neural network work, and generative AI due to ecosystem limitations and less advanced AI accelerators. Perhaps the main criticism in this case is the exclusive FSR 4 support and the inflated launch prices.
Articles, reviews, useful tips
All materials




