AMD Radeon RX 9070 Series Is Now Official, Available March 6 – Starting At $549

After nearly two months of waiting since the announcement back in CES 2025, AMD is now fully ready to launch its new Radeon RX 9070 series GPUs, with two new models, new architecture, and new suite of technologies to complement the new offerings. Here’s the overview of what’s new, and what’s changed.

RDNA 4 Architecture

At its core, the new RDNA 4 architecture features several key upgrades: improved RT Accelerators, updated encode/decode engine, more powerful AI capabilities, 3rd Gen Infinity Cache, and improvements on the Radiance Display Engine. The GPU continues to use GDDR6 memory instead of jumping on the GDDR7 like NVIDIA’s RTX 50 series, and one notable change this time around is the reversal to monolithic chip – a departure from RDNA 3’s multi-chip module (MCM) design that ultimately found limited success.

GPUs: RX 9070 XT & RX 9070

So, the products: as revealed during CES, we have the Radeon RX 9070 XT serving as the lineup’s new flagship, followed by second model that is Radeon RX 9070. AMD positions both cards in between RX 7900 XT and RX 7900 GRE based on its targeted resolution and specs, although exactly where it falls under will depend on the performance – which will be revealed when reviews go live in due time.

The Radeon RX 9070 XT will pack 64 compute units and RT Accelerators, along with 128 AI Accelerators responsible for matrix and AI compute. Boost clock is rated at 2.97GHz at 304W TBP, though the latter may be a variable as AIBs will likely push for extra power as they see fit (though MSRP models should still stick to this figure). As for the RX 9070, it comes with 56 CUs, 56 RT Accelerators and 112 AI Accelerators, with a boost clock of 2.52GHz in a 220W TBP.

Both cards will feature 16GB of GDDR6 memory, runs on PCIe 5.0 x16 standard, and continue to use the PCIe 8-pin power for full compatibility with existing PSUs (on top of arguably better safety with how the 12V-2×6 connector is doing lately over at Team Green). They also retain DisplayPort 2.1a and HDMI 2.1b support, capable of outputting 8K resolution or 4K in high refresh rates.

AMD claims that the RX 9070 will be 38% faster than RX 6800 XT, or 26% faster than RTX 3080, both in 4K Max settings – though both cards are more than 4 years old by now. Comparing to a more contemporary example, the new card is 21% faster in 4K Ultra compared to the RX 7900 GRE, with similar improvements in 1440p. As for the stronger card, it’s claimed to be 51% faster than RDNA 2’s flagship GPU, or 26% faster than NVIDIA’s flagship in the same generation; comparing to the RX 7900 GRE again shows a 42% and 38% uplift in 4K and 1440p gaming benchmarks, respectively.

FSR 4 & HYPR-RX

Newly introduced alongside the RDNA 4 GPUs is the latest major revision of FidelityFX Super Resolution, or FSR 4 in short. For this version, AMD added machine learning-based upscaling to complement temporal-based upscaling algorithm, and this comes with a limitation that only RX 9070 series GPUs will support this version. That being said, all FSR 4 provides is upscaling quality enhancements, so there should be no difference in terms of performance uplift.

Still, the ML-based upscaling does bring significant quality improvements – as the image above demonstrates, FSR 4 at Performance Mode is significantly better at handling thin objects like the poles above the tower, with quality close to the native 4K resolution.

In principle, the new ML-based upscaling works the same as NVIDIA’s DLSS Super Resolution: data is trained on a dedicated server courtesy of AMD’s datacenter Instinct GPU accelerators, which eventually becomes the upscaling algorithm that runs on Radeon GPUs, with the upscaling process done on the GPU’s AI components. So far, AMD says over 30 games support FSR 4 at launch, with more than 75 available in 2025.

AMD’s HYPR-RX suite, which consists of several performance-boosting technologies that can be enabled via a single click, is also getting a minor update this time around. Among those include the updated AMD Fluid Motion Frames (AFMF) version 2.1, with more to come later this year; the chipmaker says the suite will now support and boost over 1000 games from the get-go with driver-level technologies.

Apex Legends is the latest addition to AMD’s Anti-Lag 2 latency reduction technology, which already includes Counter-Strike 2, Dota 2, and Ghost of Tsushima; the chipmaker says the new version works on the game engine level instead of driver-level compared to its predecessor (Anti-Lag/Anti-Lag+). This, again, mirrors the principle of NVIDIA’s Reflex implementation, which is widely supported in many competitive and esports titles by now.

AI Features

On the AI side of things, AMD Software now includes a dedicated AI section, along with Install Manager to quickly manage updates, and a local chatbot similar to NVIDIA’s ChatRTX, with image generation capabilities on top of the usual chatbot stuff, like asking for GPU status, or enabling/disabling features via text prompts.

The more interesting addition is none of the above, but rather a way for AMD to easily pick up graphical errors with the help of AI to accelerate the diagnostic process. Here’s how the AMD Image Inspector works: say, a driver update has inadvertently caused some memory-related errors in some games – when graphical artifacts appears on the display, the feature will instantly detect the anomaly and prompts the user to send a bug report to AMD for further investigation.

The chipmaker stressed that Image Inspector is an opt-in feature, and users will have to explicitly allow the feature to send any images to AMD should a bug surfaces. On another note, AMD is also using AI to help improve its driver refinement process, so that should hopefully translate to less teething issues during GPU launches, or just generally better stability for the vast majority of AMD GPUs currently in circulation.

Others

First off, the Radeon Image Sharpening tool has been upgraded to work system-wide – besides games, RIS can now upscale videos or “across the entire desktop”, all of which can be enable via a single click. In case you missed it, RIS is an algorithm-based sharpening tool that improves image quality in games, though as of this update, it’ll work in browsers like Google Chrome, Microsoft Edge, and Mozilla Firefox; it also works on VLC Media Player, along with several Microsoft Office apps.

AMD has also introduced a small update on the RX 9070’s Media Engine: encoding quality and efficiency is now generally improved across H.264, HEVC (H.265) and AV1, maxing out at 8K 75FPS for HEVC and AV1 in particular. More importantly, the GPUs won’t be hard-limited on the number of concurrent streams, which is something that NVIDIA has been artificially limiting for its consumer cards. Additionally, the new cards will also feature lower idle power consumption in dual-display configurations via enhanced power optimization for FreeSync displays.

Availability

AMD will be exclusively selling the new RX 9070 series through nine AIB partners at launch, with no first-party “MBA” (Made by AMD) models planned for this lineup. Both models will be widely available on March 6, 2025 at the price of $599 for RX 9070 XT, and $549 for RX 9070 – given the supposed two-month delay from the original launch, this should already provide ample time for board partners to stock up and meet the demand that NVIDIA’s new offerings currently fail to fulfill.

Pokdepinion: Let’s hope this is a good one – the GPU market really needs something good right now.