Introduction

Sapphire NITRO+ Radeon RX 7900 XTX is the company’s flagship air-cooled graphics card based on Team Red’s fastest next-generation GPU. The NITRO+ brand in its latest generation incorporates a more flamboyant, colorful, and industrial product design, not too unlike the ASUS ROG STRIX. Large ARGB LED diffusers accentuate a premium cooler shroud, with three of Sapphire’s latest-generation fans in a cooling solution Sapphire calls the Pantheon+. Each of the three fans can be easily disconnected and individually pulled out of the cooler, the TriXX software can detect faults in each fan, letting you request RMAs easily, in case of issues. The cooler features a new composite material for the heatpipes. Other gamer-friendly features include dual-BIOS, a reinforcement brace, ARGB headers to let you sync the rest of your rig’s lighting with the card (controllable through the TriXX software), and a display output setup that includes two each of HDMI 2.1b and DisplayPort 2.1 connectors.

The new AMD Radeon RX 7900 XTX is the top dog graphics card based on AMD’s new RDNA3 graphics architecture. The company has attempted to keep Moore’s Law alive in spirit by designing the “Navi 31” silicon powering this graphics card to be the first gaming GPU to incorporate chiplets (a multi-chip module with multiple logic dies that functionally make up a single chip). AMD identified specific components of the GPU that don’t benefit from the latest foundry node as much as the pure logic components, such as stream processors; and disintegrated them from the GPU die. The “Navi 31” GPU features six Memory Cache Dies (MCDs), each built on the slightly older 6 nm process. Inside each MCD you’ll find a 64-bit GDDR6 memory interface, the memory controller backing it, and a 16 MB segment of the GPU’s 96 MB Infinity Cache. Six of these make up the GPU’s 384-bit GDDR6 memory interface. The rest of the GPU with the bulk of the pixel-crunching machinery, is nucleated into a large die called the Graphics Compute Die (GCD), built on the 5 nm EUV process. The move helps save costs by maximizing its 5 nm wafer allocation with TSMC, and the company made sure to transfer some of these savings over to customers, with the Radeon RX 7900 series being 20-50% cheaper than the high-end NVIDIA GeForce RTX 40-series “Ada” GPUs launched to date, which are built on monolithic 4 nm dies.

The new RDNA3 graphics architecture sees the introduction of a new-generation Compute Unit (CU) with dual instruction issue-rate stream processors, support for new math formats, AI acceleration, and 2nd generation Ray Accelerators. There are 96 CUs spread across six Shader Engines, which make up 6,144 stream processors, 96 Ray Accelerators, 384 TMUs, and a whopping 192 ROPs. The RX 7900 XTX maxes out the “Navi 31” silicon enabling all 96 CU available, and the GPU’s full 384-bit GDDR6 memory interface. It now enjoys 24 GB of GDDR6 memory ticking at 20 Gbps, which works out to an impressive 960 GB/s memory bandwidth (87% higher than that of the RX 6900 XT), allowing AMD to slightly reduce the size of the Infinity Cache memory.

With RDNA3, AMD is hoping to repeat the 50% performance/Watt generational gain that springboarded it back to the high-end PC graphics segment with RDNA2. The 20% higher CU count, each with 17.5% higher IPC over RDNA2 CUs, coupled with higher engine clocks, and a faster memory sub-system, together help AMD achieve a 53% generational performance/Watt uplift that’s needed to remain competitive with NVIDIA. The RX 7900 XTX has a starting MSRP of USD $999, which is 20% cheaper than the NVIDIA RTX 4080, and 60% less than that of the RTX 4090 flagship, while targeting the same market—4K Ultra HD maxed-out gaming with ray tracing.

The Sapphire Radeon RX 7900 XTX NITRO+ OC ups the game with not just the various hard-product feature-additions, but also a factory-overclock, with the Game Clock set up to 2510 MHz compared to 2455 MHz reference. As we’ve seen on other non-reference cards, the RX 7900 XTX truly comes to life with a stronger power delivery system. The NITRO+ draws power from three 8-pin PCIe power connectors, which combined with the PCIe slot power, add up to 525 W. With its primary BIOS, the card has a power-limit of 420 W (compared to 350 W reference), which should enable higher boost frequency residency. The Tri-X Pantheon cooling solution, offers a low noise output. Official pricing hasn’t been released yet, but we’re expecting a $1100 price point, which makes the card $100 more expensive than the AMD reference design MSRP.