Introduction

The TUF Gaming Radeon RX 7900 XTX OC is the flagship custom-design RX 7900 XTX graphics card by ASUS, in the apparent-absence of an ROG Strix product. The card features the mammoth 4-slot TUF Gaming cooling solution that the company debuted with its GeForce RTX 4090 TUF Gaming graphics card, and a PCB with increased power delivery from a third 8-pin power connector, to support the factory overclocked speeds. The Radeon RX 7900 XTX is the new flagship from Team Red, with which the company plans to continue competing with NVIDIA in the high-end graphics card segment, and aggressively with pricing. This is thanks to a unique design choice by AMD that keeps Moore’s Law alive in spirit if not in letter.

The new Radeon RX 7900 XTX debuts the RDNA3 graphics architecture, with which the company’s is introducing the world’s first chiplet-based gaming GPU (a multi-chip module with multiple logic dies). AMD has taken a similar approach with its new GPU as it did with Ryzen processors a few years ago—those components that benefit the most from a switch to a new foundry process are built on them, while specific components that don’t need the latest node, are disintegrated and built on separate dies, on a slightly older node. In this case, the Shader Engines, media accelerators, and display controllers, are aggregated into a centralized die called the Graphics Compute Die (GCD) that’s built on the latest 5 nm process, while the Infinity Cache memory, memory controllers, and a 64-bit wide GDDR6 memory PHY, are separated into six Memory Cache Dies (MCDs), built on 6 nm. There are six of these on the “Navi 31” GPU powering the RX 7900 XTX, and so it enjoys a 384-bit memory interface. Future Navi 3x series GPUs may feature fewer MCDs to achieve narrower memory interfaces. This way, AMD gets to maximize its foundry allocation of both the 5 nm and 6 nm nodes.

The new RDNA3 graphics architecture introduces a new compute unit with dual instruction issue-rate SIMD components, support for new math formats, AI acceleration, and 2nd generation Ray Accelerators that improve ray tracing performance over the previous generation. AMD claims a 17.5% IPC uplift over RDNA2 compute units, which when combined with generationally increased clock-speeds, and a 20% increase in CU counts over “Navi 21,” form the bedrock of the performance uplift over the previous generation, with the company claiming to have retained the streak of 50% performance/Watt uplift with each RDNA generation. AMD has also rebalanced its memory system to rely less on the Infinity Cache memory. Each of the six MCDs has a 16 MB segment of the GPU’s 96 MB Infinity Cache, which seems less than the 128 MB of the “Navi 21,” until you find out that AMD has increased the memory bus width by 50% (256-bit to 384-bit), and is using even faster 20 Gbps memory (compared to 16 Gbps of the RX 6900 XT), which yields a massive 87% memory bandwidth increase.

The most striking aspect of the Radeon RX 7900 series is power, with the top RX 7900 XTX only featuring a 350 W typical board power at reference clocks, compared to the NVIDIA flagship with its 450 W power that can be increased to 600 W or beyond. The reference RX 7900 XTX only features two convenient 8-pin PCIe power connectors, while non-reference cards such as the ASUS TUF Gaming OC in this review, come with three 8-pin connectors for a total of 525 W power delivery capability.

ASUS hasn’t announced any official clock frequencies yet, the pricing is also unknown. Throughout this review, we’ve assumed a price point of $1100 and will update everything once official pricing is out.