Introduction

The ASRock Radeon RX 7900 XT Phantom Gaming White is a slick new custom-design of AMD's second fastest GPU this generation. The card is a style variant of the RX 7900 XT Phantom Gaming, but based on a white theme. There appears to be a growing demand for this color scheme globally, across all market segments. ASRock brought out the Steel Legend series to some of its lower-end Radeon GPUs, and for the high-end RX 7900, it has the Phantom Gaming White. The styling sees white surfaces across the cooler shroud, backplate, and the fan impellers. The cooler is much larger than the PCB underneath, which is barely visible, and so ASRock left it untouched and black, carried over from the original RX 7900 XT Phantom Gaming.



The Radeon RX 7900 XT sits firmly in the enthusiast segment of GPUs. It is designed to offer maxed out AAA gameplay at 4K, including with ray tracing, and has the silicon heft to pull it off. It's based on the same "Navi 31" chiplet-based GPU as the flagship RX 7900 XTX, and the latest RDNA 3 graphics architecture. RDNA 3 introduces a new generation dual instruction issue-rate compute unit with support for new math formats, and a 17% IPC improvement over RDNA2. The new dual compute unit also features AI Accelerators, which are fixed-function components that prepare matrix math workloads for the SIMD units. The new 2nd generation Ray Accelerators offer a 50% increase in ray intersection performance over those of RDNA2. All these are backed by generational increases in the GPU clocks thanks to the new 5 nm process, larger video memory, and a wider memory bus along with faster memory.

The "Navi 31" is the first chiplet-based GPU. A chiplet-based device is different from a multi-chip module (MCM). In an MCM, chips that can otherwise exist independently on their own packages, are made to share a common package, to conserve PCB footprint. In a chiplet-based device, something that should be a monolithic chip, is disaggregated into separate pieces of silicon, called chiplets, so some of them could be built on older foundry nodes (to reduce manufacturing costs), and made to connect to each other using a high-performance interconnect that make the chiplets work as if they're parts of a whole chip. In case of the "Navi 31," all the logic heavy components that can benefit from the switch to 5 nm are clumped into a large central chiplet called the graphics compute die (GCD), which is flanked on two sides by all the memory-heavy components, in the form of six memory cache dies (MCDs), built on the 6 nm node. Each MCD has a 16 MB segment of the chip's 96 MB Infinity Cache, and a 64-bit portion of its 384-bit GDDR6 memory interface. These are then interconnected by Infinity Fanout Links, a high-performance interconnect physical layer co-developed by AMD and TSMC; and encased in structural silicon that levels the chiplets to make uniform contact with the cooling solution.

AMD carved the RX 7900 XT out of the "Navi 31" by enabling 84 out of 96 compute units physically present on the GCD, and enabling 5 out of 6 MCDs. This results in 5,376 stream processors, 168 AI accelerators, 84 Ray accelerators, 336 TMUs, and the chip's full 192 ROP count. The memory bus width is reduced to 320-bit, and made to drive 20 GB of GDDR6 memory running at 20 Gbps, resulting in a hefty 800 GB/s of memory bandwidth. With one of the six MCDs disabled, the Infinity Cache size is reduced to 80 MB. ASRock has given the RX 7900 XT Phantom Gaming White factory overclocked speeds of 2075 MHz Game clock, and 2450 MHz boost, compared to 2025/2394 MHz AMD reference clocks. The ASRock Radeon RX 7900 XT Phantom Gaming White is priced at $800, a reasonable $20 increase over the most affordable RX 7900 XT models, and the default Phantom Gaming.

Packaging

The Card

ASRock's card looks exactly like the regular Phantom Gaming, just in white. The new color theme looks mighty good, thanks to the inclusion of some colored highlights to add more surface detail.


Dimensions of the card are 33.0 x 14.0 cm, and it weighs 1456 g.


Installation requires three slots in your system. The card's width is 58 mm.


Display connectivity includes three standard DisplayPort 2.1 ports and one HDMI 2.1a.

AMD has upgraded their encode/decode setup. It now comes with two independent hardware units that can encode and decode two streams of video in parallel, or one stream at double the FPS rate. There's support for VP9, H.264, H.265 and AV1 decode, and encoding is supported for H.264, H.265 and AV1.


Just like other RX 7900 XT models, the Phantom Gaming White requires two 8-pin power inputs. This configuration is good for up to 375 W of power input.

Teardown

Six heatpipes provide good cooling for the GPU's heat output. The main heatsink also provides cooling for the memory chips and VRM circuitry.


The backplate is made of metal and protects the card against damage during installation and handling.

High-resolution PCB Pictures

These pictures are for the convenience of volt modders and people who would like to see all the finer details on the PCB. Feel free to link back to us and use these in your articles, videos or forum posts.


High-resolution versions are also available (front, back).

Circuit Board (PCB) Analysis

GPU voltage is a 14-phase design, managed by a Monolithic Power Systems MP2857 controller.


Monolithic MP87997 DrMOS components are used for GPU voltage; they are rated for 70 A of current each.


Memory voltage is a three-phase design, managed by a Monolithic Power Systems MP2856 controller.


For memory, Monolithic MP87997 DrMOS with a 70 A rating are used again.


The GDDR6 memory chips are made by Samsung and carry the model number K4ZAF325BC-SC20. They are specified to run at 2500 MHz (20 Gbps effective).


AMD's Navi 31 graphics processor is the world's first GPU that uses a chiplet architecture. Note the large die in the center, called "GCD," graphics compute die, which houses the compute units, it is surrounded by six smaller "MCD," memory cache dies, that contain one memory controller interface and one slice of cache each. While they look similar, the MCDs are not HBM chips. The MCDs are fabricated on a 6 nm process at TSMC Taiwan with a die size of 36.6 mm² each, the GCD is fabricated using TSMC's 5 nanometer node, with a die size of 300 mm². The combined transistor count of the GPU is 57.7 billion.