AMD Radeon RX 9070 XT Review (2025): Performance, Benchmarks, and Gaming Tests

Over the past few months, AMD has made waves with its new Radeon RX 9000 series. Powered by the RDNA 4 architecture, this lineup features two gaming-focused GPUs: the Radeon RX 9070 XT and the Radeon RX 9070. Each card comes with 16 GB of VRAM and is specifically engineered for high-performance gaming at 1440p and 4K, allowing gamers to enjoy even the most demanding titles with ease.

In this article, Tech2Geek takes a closer look at the XFX MERC RGB Radeon RX 9070 XT Black, a custom variant featuring a triple-fan cooling system and a striking illuminated light bar. We tested the card extensively at 1440p and 2160p across modern and classic titles, measuring performance in both rasterization and ray tracing. Beyond gaming, we also evaluated its capabilities in content creation and AI workloads, while analyzing power consumption, cooling efficiency, and overall stability.

Does AMD finally have a graphics card capable of shaking up the market? The company certainly hopes so. Promises are plenty, ranging from competitive pricing to a notable leap in performance—especially in ray tracing—along with the arrival of FSR 4. The Radeon RX 9070 XT launches with a suggested retail price of $699 in the U.S.

The RDNA 4 Graphics Architecture

According to AMD, RDNA 4 represents the culmination of a long, multi-generational evolution. The architecture is designed not only to meet today’s gaming demands but also to prepare for future titles over the next several years. With the Radeon RX 9070 series, AMD is setting its sights on the highly competitive segment of GPUs priced under $700, aiming to deliver the best balance between performance and value.

RDNA 4 promises performance improvements across several key areas. To achieve this, its Compute Units (CUs) have been optimized, while the Ray Tracing accelerators are now in their third generation, and the AI accelerators have reached their second generation. The GPU also features an enhanced multimedia engine, with boost clock speeds pushing closer to 3 GHz.

Compute Units

RDNA 4 aims to improve both general workloads (rasterization) and compute performance. While new rendering techniques are gradually being adopted, these fundamental workloads remain at the core of most real-time rendering experiences.

The refinements made to the RDNA 4 Compute Units enhance the memory subsystem and scalar units while introducing dynamic register allocation. Combined with significantly higher operating frequencies compared to RDNA 3, these improvements boost performance per CU. As a result, the Radeon RX 9070 series can deliver performance on par with the Radeon RX 7900, despite having a lower overall number of Compute Units.

Third-generation Ray Tracing accelerators

This is arguably the area that has evolved the most with this generation, as AMD seeks to make up ground against Nvidia’s solutions. The new 3rd-generation Ray Tracing (RT) accelerators promise up to 2x the ray tracing performance of RDNA 3.

One of the most demanding aspects of ray tracing workloads is the construction of the BVH (Bounding Volume Hierarchy) data structure. Its purpose is to accelerate rendering by reducing the number of collision checks required between rays and objects in a scene.

When a GPU runs a ray tracing algorithm, it must determine which objects are intersected by a ray of light. Testing every single object individually would be computationally prohibitive—especially in complex environments with millions of triangles. The BVH addresses this by organizing objects into a hierarchy of bounding volumes, which allows the GPU to quickly discard large sections of the scene without having to test each triangle.

Here’s how it works: when a ray is cast into the scene, it traverses the BVH tree step by step. If the ray does not intersect a bounding box, all objects within that box are skipped. If it does, the algorithm descends further into smaller bounding volumes until it reaches the actual geometry.

AMD’s GPUs feature dedicated Ray Accelerators designed to optimize BVH construction and traversal, ensuring faster and more efficient real-time rendering. With RDNA 4, AMD is introducing an advanced method called Oriented Bounding Boxes (OBB). This reduces BVH size and complexity, cutting VRAM requirements while boosting ray tracing efficiency.

Additional improvements include:

• A second ray intersection engine within the RT accelerators, which doubles throughput for Ray/Box and Ray/Triangle intersection tests.
• A dedicated ray transformation block, improving performance when rays pass through the deeper levels of the BVH tree.

2nd-Generation AI Accelerators

Alongside ray tracing, AMD has upgraded its AI hardware with 2nd-generation AI Accelerators, designed to handle modern AI workloads more efficiently. This evolution combines architectural refinements, expanded instruction support, and enhanced data handling.

Key upgrades include:

• Additional math pipelines to accelerate AI-specific calculations.
• Support for new data types, such as FP8, improving performance in machine learning and inference.
• Structured sparsity, an optimization that speeds up computations by skipping over non-essential values in neural networks, preserving accuracy while reducing processing time.

Together, these advancements significantly improve both ray tracing and AI-driven rendering techniques, making RDNA 4 a more future-ready architecture for next-generation games and creative workloads.

It reduces the number of calculations by eliminating certain multiplications, which allows operations to run faster while easing memory pressure. Storing lighter matrices also saves memory bandwidth and, more broadly, ensures better inference performance. For example, in tasks such as image recognition or natural language processing, this makes it possible to handle more queries per second.

Multimedia Engine

Multimedia Engine

When it comes to the media engine, RDNA 4 introduces improvements focused on encoding quality for both recording and live streaming, across all major codecs including H.264, HEVC, and AV1.

Decoding support includes:

• VP9: 4K @ 210 fps / 8K @ 48 fps (4:2:0, 8/10-bit)
• H.264: 4K @ 330 fps (4:2:0, 8-bit)
• H.265 (HEVC): 4K @ 210 fps / 8K @ 48 fps (4:2:0, 8/10-bit)
• AV1: 4K @ 240 fps / 8K @ 60 fps (4:2:0, 8/10/12-bit)

Encoding support includes:

• H.264: 4K @ 180 fps (4:2:0, 8-bit)
• H.265 (HEVC): 4K @ 180 fps / 8K @ 48 fps (4:2:0, 8/10-bit)
• AV1: 4K @ 240 fps / 8K @ 60 fps (4:2:0, 8/10-bit)

These upgrades make RDNA 4 particularly attractive for streamers, video creators, and content professionals, providing smoother high-resolution workflows and more efficient hardware-accelerated media processing.

FSR 4

FidelityFX Super Resolution 4 (FSR 4) is AMD’s latest upscaling technology, now powered by machine learning (ML). Unlike previous versions that relied primarily on traditional spatial and temporal algorithms, FSR 4 leverages an advanced AI model to convert lower-resolution frames into high-resolution outputs.

This new approach delivers:

• Sharper details and improved image reconstruction.
• Better handling of fine textures and edges.
• Optimized performance for demanding titles by reducing GPU workload.

With FSR 4, gamers can enjoy higher frame rates and smoother gameplay at 1440p and 4K, without compromising visual quality.

The process begins with low-resolution inputs combined with critical scene data such as depth, color, and motion vectors. These inputs allow the algorithm to better interpret object positioning, lighting, and movement within the scene. Using this data, FSR 4 applies AI-driven upscaling, accelerated by the AI accelerators built into RDNA 4. Thanks to this approach, FSR 4 is capable of running in real time—even at high refresh rates—making it particularly well-suited for competitive gaming and immersive experiences.

FSR 4 promises high-definition upscaling with sharper details, improved clarity, and fewer visual artifacts. It can also be paired with advanced frame generation technologies that insert additional frames to deliver smoother gameplay, while still aiming to maintain high visual quality.

As you might expect, FSR 4 requires an RDNA 4 GPU equipped with dedicated AI accelerators, which limits its availability to the Radeon RX 9000 series and games that support FSR 3.1 or newer. Despite this limitation, the technology represents a significant step forward, bringing AMD closer to Nvidia’s DLSS, which also leverages AI to enhance real-time rendering.

DLSS 4.0: Compatible Games (2025)

• The Alters
• Bellwright
• Call of Duty: Black Ops 6 and 7*
• Creatures of Ava
• Dragonkin: The Banished
• Enotria: The Last Song
• FragPunk
• Funko Fusion
• God of War: Ragnarok
• Horizon Zero Dawn Remastered
• Horizon Forbidden West
• Hunt: Showdown 1896
• Incursion
• Red River Kristala
• Marvel Rivals
• Marvel’s Spider-Man 2
• Marvel’s Spider-Man Remastered
• Marvel’s Spider-Man: Miles Morales
• MechWarrior 5: Clans
• Monster Hunter Wilds
• Nightingale
• No More Room in Hell 2
• PANICORE
• Predator: Hunting Grounds
• Ratchet & Clank: Rift Apart
• Remnant 2
• Smite 2
• The Axis Unseen
• The Last of Us: Part I
• The Last of Us: Part II Remastered
• Until Dawn
• Warhammer 40,000: Space Marine 2
• Kingdom Come: Deliverance II
• Dynasty Warriors: Origins
• Civilization VII
• Half-Life 2 RTX
• Lost Soul Aside
• Stellar Blade
• Tides of Annihilation
• Mecha BREAK
• Phantom Blade Zero
• GTA V Enhanced
• Senua’s Saga: Hellblade II
• Cyberpunk 2077

New titles announced with DLSS 4.0 (2026)

• Resident Evil Requiem
• PRAGMATA
• Borderlands 4
• Vampire: The Masquerade – Bloodlines 2
• Indiana Jones and the Great Circle
• Directive 8020
• Neverness To Everness (NTE)

For this launch, the flagship of the RDNA 4 generation is the Radeon RX 9070 XT. This graphics card is powered by a GPU manufactured on the 4 nm process, with a die size of 357 mm² housing an impressive 53.9 billion transistors.

Inside, the chip integrates:

• 64 Compute Units (CUs)
• 64 Ray Accelerators
• 128 AI Accelerators
• 4,096 Stream Processors
• 64 ROPs (Render Output Units)

This combination of high transistor density and advanced architecture makes the RX 9070 XT the showcase product for AMD’s RDNA 4 lineup, targeting gamers who demand both performance and next-gen features.

The Game Clock is set at 2,400 MHz, while the Boost Clock nearly reaches the 3 GHz mark at 2,970 MHz. In terms of compute performance, the card delivers 48.7 TFLOPs in single precision and 97.3 TFLOPs in half precision. For AI workloads, the GPU reaches 779 TOPS (INT8) and up to 1,557 TOPS (INT4).

The GPU also benefits from 64 MB of 3rd-generation Infinity Cache and comes equipped with 16 GB of GDDR6 memory, running at 20 Gbps on a 256-bit bus, for an effective memory bandwidth of 640 GB/s.

On the display side, the Radeon RX 9070 XT supports both DisplayPort 2.1a (UHBR13.5) and HDMI 2.1b.

This gaming graphics card uses the PCIe 5.0 x16 interface and has a Total Board Power (TBP) of 304W, with AMD recommending a 750W power supply for stable operation.

For this review, we used a custom model from XFX, the MERC RGB Radeon RX 9070 XT Black. This card is massive in size and requires caution when installing it in cases with limited internal space.

Its large dimensions are explained by the inclusion of a triple-fan cooling system, which extends well beyond the PCB.

This design creates an open airflow path, allowing the air pushed by the rearmost fan to pass from top to bottom through the heatsink fins and multiple heat pipes, which are responsible for transferring heat to the dissipation zones.

The design is fairly straightforward, featuring an all-black shroud with a plastic top cover and an aluminum backplate. The backplate serves a dual purpose: it reinforces the card’s structural rigidity while also contributing to cooling.

As its name suggests, the card features a distinctive customizable aRGB light bar running along its entire side. A bundled 3-pin cable allows connection to an aRGB header on a motherboard, controller, or hub. Additionally, the PCB includes a small switch that lets users switch to a backup VBIOS in case the primary one becomes corrupted or damaged.

On the technical side, this is a 3.5-slot card with a length of 36 cm, so extra caution is required when installing it in smaller cases.

Power delivery relies on three 8-pin PCIe connectors, positioned centrally along the side. This placement isn’t the most convenient for cable management, but since the connectors are recessed into the shroud, there’s no risk of interference with the side panel.

The GPU is clocked at 1,870 MHz base, with a Game Clock of 2,670 MHz, and can reach up to 3,100 MHz in Boost mode—surpassing the 3 GHz mark. However, this peak frequency is only achievable in specific scenarios and for short bursts. On the memory side, there is no factory overclock, with the GDDR6 VRAM set at 20 Gbps.

Test Protocol

To evaluate the AMD Radeon RX 9070 XT, we used a combination of gaming benchmarks, synthetic tests, and GPGPU applications. All games were tested at maximum graphics settings across multiple resolutions to push the card to its limits.

Test Platform

• Motherboard: Asus ROG Maximus Z690 HERO
• Processor: Intel Core i9-12900K
• Memory: G.Skill Trident Z5 RGB, 2 × 16 GB DDR5-6000 CL28
• Storage: Kioxia Exceria Pro 2 TB SSD
• Power Supply: NZXT C1500 Platinum
• Cooling: DeepCool LS720 AIO liquid cooler

Games Tested

• Metro Exodus
• Far Cry 6
• Shadow of the Tomb Raider
• Horizon Zero Dawn
• Watch Dogs: Legion
• Cyberpunk 2077
• Star Wars Outlaws

Benchmarks

• 3DMark Fire Strike (Performance & Extreme)
• Blender (Monster, JunkShop, and Classroom scenes)
  • Supports NVIDIA OptiX acceleration for GeForce and Quadro RTX cards.
  • For AMD Radeon RX 6000/7000 series, OpenCL API was used.
• LuxMark 3.1
  • Leverages the OpenCL API to perform 3D rendering workloads.
  • Can run on GPU, CPU, or both, providing insights into performance scaling.
  • Produces scores in Samples/sec and Rays/sec.
• FAHBench
  • Runs molecular dynamics simulations to measure GPU performance for biophysics research.
• UL Procyon AI Benchmark
  • Focused on AI text generation, tested with PHI 3.5, Mistral-7B, Llama 3-1, and Llama-2 models.

Additional Testing

We also measured:

• Power consumption
• Thermal performance
• Noise levels
• Frequency stability

Cooling & Thermal Performance

Thermal monitoring was conducted using GPU-Z. To stress the card, we ran a 10-minute full-load torture test with the GPU operating at maximum capacity. All thermal tests were performed in open-air conditions to remove any influence from case airflow and to focus solely on the cooler’s capabilities.

At low load, the card operates in silent mode: all three fans remain completely off. In this state, the GPU idles at 41 °C with an ambient room temperature of 20 °C.

When pushed to 100% load, the Radeon RX 9070 XT stabilizes at just 52 °C, which is an excellent result for a high-end GPU. Memory modules, however, run hotter, peaking at 80 °C.

Fan behavior is dynamic and well-managed. Initially, all three fans spin up to 39% of their maximum speed, before gradually settling at around 28% once the cooling system reaches equilibrium.

Noise Levels

“When idle, Fanless mode makes the card completely silent, which is perfect. After 10 minutes under full load, our sound level meter registers 42 dBA, making this card quieter than the vast majority of other models we’ve tested. Under heavy load, a faint airflow can be heard, but it quickly fades into the background. However, be cautious: this balance can change in a poorly ventilated case, as the card’s cooling system expels hot air inside the chassis and then reuses that same air for cooling.

Frequency stability

Here’s a look at GPU and memory frequency tracking. There’s absolutely nothing to criticize—both the GPU and VRAM remain perfectly stable over time. With GPU frequency under tight control, performance is consistently maintained. During this test, the GPU averaged 2601 MHz, while the GDDR6 memory ran at 2505 MHz.

Power consumption

Depending on the charge level, the consumption is between 11 and 340 Watts. This profile reveals a Radeon RX 9070 XT Black RGB MERC that is a bit load-hungry knowing that AMD is counting on 304 Watts. However, we are very far from the gargantuan demand of the RTX 5090 with its 602 Watts! We find the power profile of the Radeon RX 7900 XTX and a slightly more demanding solution than the GeForce RTX 4080 SUPER. It should normally be a little less greedy.

MERC RGB Radeon RX 9070 XT Black, les performances

You will find performance reports across three scenarios: Rasterization, Ray Tracing (Ultra mode), and Ray Tracing (Ultra mode) combined with FSR in Performance mode. In addition, we also tested FSR 4.0 in one title.

For this review, we use the GeForce RTX 5090 Master as our reference point. The Radeon RX 9070 XT is a high-performance solution, and to fully showcase its capabilities—and highlight the differences from its predecessors—it must be tested in high resolutions. That’s why our benchmarks focus on 1440p and 4K (2160p), with graphics settings pushed to their maximum.

Rasterization Performance

In rasterization, the Radeon RX 9070 XT delivers 76% of the GeForce RTX 5090’s performance at 1440p and 61.6% at 2160p. This positions it on par with the GeForce RTX 4080 Super, which reaches 76.9% and 62.2% of the RTX 5090’s performance, and close to the Radeon RX 7900 XTX (74.3% and 62.7%). It also outpaces the RTX 4070 Ti Super, which achieves 68.9% at 1440p and 54.5% at 2160p. The RTX 5080, however, maintains the lead with a 6% advantage at 1440p and 10% at 2160p.

Framerates reveal a card perfectly suited for 1440p gaming with all settings maxed out. Regardless of the title—whether older or more recent—the framerate consistently stays above 60 FPS. In fact, 7 out of 8 games surpass the 90 FPS mark, making for an excellent balance of smoothness and visual quality.

At 2160p, older games pose no challenge, with framerates still holding above 90 FPS at maximum settings. However, the situation becomes more demanding with newer titles such as Star Wars Outlaws, Kingdom Come: Deliverance II, and Cyberpunk 2077. In these cases, gameplay remains smooth and responsive, but the 60 FPS threshold is not always reached, or only just surpassed.

3DMark Fire Strike

3DMark’s Fire Strike benchmark, tested in 1080p (Default), 1440p (Extreme), and 2160p (Ultra), confirms our overall ranking—although the performance gap with the RTX 5090 is more pronounced here. Using the RTX 5090 as a baseline, the Radeon RX 9070 XT Black RGB MERC achieves 75%, 57%, and 55% of its performance at 1080p, 1440p, and 2160p respectively. These results place it ahead of the RTX 4080 Super, but still behind its bigger sibling, the Radeon RX 7900 XTX.

Ray Tracing Performance

We then move to game benchmarks with Ray Tracing set to Ultra. All graphics settings are pushed to their maximum, and neither DLSS nor FSR upscaling technologies are enabled at this stage. Once again, the RTX 5090 serves as our reference point.

With Ray Tracing enabled, the rankings shift slightly but without any major surprises—highlighting the progress AMD has made in this area. The Radeon RX 9070 XT delivers 69% of the RTX 5090’s performance at 1440p and 61% at 2160p. At both resolutions, it trails the RTX 4080 Super by around 7% at 1440p and just 1% at 4K. However, its results place it ahead of the RX 9700 XTX, AMD’s previous flagship gaming GPU. Meanwhile, the RTX 4070 Ti Super and RTX 5070 lag significantly behind.

Ray Tracing is demanding on resources, leading to a noticeable drop in framerates—struggling to surpass 60 FPS at 1440p. As a reminder, our tests are conducted with all settings maxed out and Ray Tracing set to Ultra. At 4K, the challenge becomes even greater, particularly in modern titles like Star Wars Outlaws, where performance fails to reach the 30 FPS mark.

For those aiming to maintain optimal visual quality with Ray Tracing on Ultra at 4K, the practical solution is to enable an upscaling technology such as FSR 3, combined with frame generation.

Finally, 3DMark’s Speed Way benchmark at both 1440p and 2160p further confirms this performance hierarchy.

The 3DMark FSR 2 Feature Test provides an estimate of the performance ‘boost’ offered by this upscaling technology.

The performance gains are substantial—+221% at 1440p and +275% at 4K.

FSR 3 Performance

That said, these results come from a theoretical benchmark. To provide a more practical perspective, we extended the tests to two demanding titles: Cyberpunk 2077 and Star Wars Outlaws. Gameplay was evaluated at both 1440p and 2160p, with Ray Tracing enabled in each case.

In Cyberpunk 2077, the performance gains are remarkable—+395% at 1440p and +469% at 4K. The results are equally impressive in Star Wars Outlaws, with +420% at 1440p and +484% at 4K. Such dramatic boosts push framerates beyond 220 FPS at 1440p and over 120 FPS at 4K, delivering a truly transformative gaming experience

FSR 4.0

We extended the study to Kingdom Come: Deliverance II with the activation of FSR 4.0. The gains are striking: framerates jump from 88 FPS to 212 FPS at 1440p, and from 52 FPS to 130 FPS at 4K.

Performance in Creation and AI

While the Radeon RX 9070 XT is primarily designed for gamers, its RDNA 4 architecture also delivers strong potential in other workloads. Many creative and AI-driven applications can harness its GPU for hardware acceleration, resulting in significant time savings and improved productivity.

However, some of our applications do not yet fully recognize the card—Blender 4.3, for example. This is most likely a driver-related issue, and we will update our findings once proper testing is possible. In the meantime, here are some of the results we were able to collect:

LuxMark v3.1

FAHBench

IA Text Generation

Using the Procyon AI Test Generation benchmark, we evaluated performance in the context of AI workloads, particularly with several local LLM models. Taking the RTX 5090 as our reference, the Radeon RX 9070 XT reaches 33% of its performance. In this benchmark, NVIDIA solutions clearly dominate, especially with the RTX 50 series generation.

Here are the detailed results:

Conclusion

With the Radeon RX 9070 XT, AMD delivers a well-balanced graphics card capable of competing in both 1440p and 4K gaming. In terms of rasterization performance, it can match or even surpass the RTX 4080 Super FE, offering smooth gameplay at 1440p and solid 4K performance. However, the most demanding titles may require some adjustments to maintain high framerates.

In Ray Tracing, AMD’s progress is undeniable, though NVIDIA still maintains a lead with its RTX 50 series. Nevertheless, the RX 9070 XT comes close to the 4K performance of the RTX 4080 Super, and the integration of FSR 3 and 4 significantly enhances the experience, enabling impressive framerates, especially when frame generation is enabled.

While the RX 9070 XT is primarily aimed at gamers, its creative and AI capabilities remain limited. Some applications, such as Blender 4.3, do not yet fully support the card—a point to monitor, though it does not diminish the card’s overall potential.

Overall, the Radeon RX 9070 XT positions itself as a very strong alternative to NVIDIA’s offerings, including the RTX 5070 Ti, with a more attractive performance-to-price ratio and reasonable power consumption. It is particularly compelling for demanding gamers seeking a powerful GPU without overspending on MSRP or electricity. This is especially notable compared to the RTX 5070 Ti, which carries a recommended price above €880 in France.

For AMD’s custom XFX variant, the MERC RGB Radeon RX 9070 XT Black, the cooling system is efficient and provides pleasantly quiet operation even under full load. On the downside, power consumption is higher than expected.

In short, the Radeon RX 9070 XT is a formidable option in its class, provided availability and pricing are maintained—conditions under which NVIDIA could face serious competition.

Did you enjoy this article? Feel free to share it on social media and subscribe to our newsletter so you never miss a post!

And if you'd like to go a step further in supporting us, you can treat us to a virtual coffee ☕️. Thank you for your support ❤️!