Introduction

The new RDNA3 graphics architecture releases today, and leading from the top are the new Radeon RX 7900 XT (in this review), and the RX 7900 XTX flagship, which we’ve also reviewed today. Both these graphics cards are designed to take the fight to NVIDIA’s high-end: the GeForce RTX 40-series “Ada,” but at highly competitive prices. The RX 7900 XT from this review targets a slightly lower price-point than the RX 7900 XTX flagship, while being designed for the exact same class of gaming—4K Ultra HD maxed out with ray tracing. The new RDNA3 architecture promises to repeat the 50% leap in performance/Watt that made it possible for AMD to return to the high-end graphics segment with their RX 6000 series RDNA2.

The company still believes that Moore’s Law exists, and that it requires efficiently utilizing the various foundry nodes to optimize cost. This is basically what AMD does in the CPU space, something it debuted with the Ryzen 3000 series and 2nd Gen EPYC processors. The CPU cores are built on the latest node, while the memory controller isn’t. In the same way, the stream processors and other number crunching machinery of the Shader Engines are built on a centralized Graphics Compute Die (GCD) that’s fabricated on the latest 5 nm EUV foundry node; while the Infinity Cache memory, memory controllers, and GDDR6 PHY, are pushed to multiple Memory Cache Dies (MCDs), built on the slightly older 6 nm node. The “Navi 31” silicon on which the RX 7900 series is based, features six of these, and hence has a 384-bit wide memory interface. Five of these are enabled on the RX 7900 XT, hence it ends up with a 320-bit memory interface. Each MCD has a 16 MB piece of the GPU’s 96 MB Infinity Cache, the RX 7900 XT gets 80 MB of it.

AMD carved the Radeon RX 7900 XT in this review from the “Navi 31” GPU, by enabling 84 out of 96 RDNA3 compute units physically present on the silicon, and 5 out of 6 MCDs. This results in hardware specs of 5,376 (out of 6,144) stream processors, 336 (out of 384) TMUs, the chip’s full 192 ROP count, and a 320-bit GDDR6 memory interface, which runs 20 GB of 20 Gbps memory, resulting in an impressive 800 GB/s memory bandwidth.

The RDNA3 graphics architecture introduced dual issue-rate compute units, with a high degree of optimization in the way idle SIMD resources are utilized, support for newer math formats, and a new AI accelerator that retasks the SIMD resources for matrix math functions. Together, these optimizations produce a 17% IPC uplift over the RDNA2 CU. There are 96 CUs on the silicon, which work out to 6,144 stream processors. The architecture also sees an increase in engine clocks, and a decoupling of the shader clock speeds to those of the GPU’s Front End, which operates at a 10-15% higher frequency. The most striking aspect of the RDNA3 architecture is that the typical board power of these GPUs is well contained, with the RX 7900 XTX rated at just 350 W, and the RX 7900 XT at 315 W—both of which can be fed by just two 8-pin PCIe power connectors, and cooled by solutions much smaller than those found on the competing NVIDIA GeForce RTX 4080 or RTX 4090 “Ada.”

AMD is transferring the costs saved with its chiplet architecture over to customers, by aggressively pricing the Radeon RX 7900 XT at $900. Its bigger sibling, the RX 7900 XTX, goes for just $1,000. Compared to these, the NVIDIA offerings are quite expensive, with the RTX 4080 priced at $1,200 and the flagship RTX 4090 at $1,600. What’s more, the RX 7900 XT has a typical power of just 315 W, and makes do with two conventional 8-pin PCIe power connectors, and the reference board design by AMD is as compact as the RTX 3080 Founders Edition, making it friendly with even some SFF cases. In this review, we compare the RX 7900 XT with its segment rivals, as well as a small but growing selection of graphics cards, on our swanky new 13900K-based VGA test-bench.