Introduction

AMD Radeon RX 7900 GRE (Golden Rabbit Edition) isn't new, its first availability traces back to July 2023. The card was available almost exclusively in China and parts of Asia, but AMD has now decided to give it a wider launch, including in Europe. Why, and why now? NVIDIA recently stirred the pot with the January introduction of the GeForce RTX 4070 Super and the RTX 4070 Ti Super, which have sent prices of the RTX 4070 and RTX 4070 Ti down, creating a tightly-packed playing field between the $400 to $600 price range.

When it launched at $500, the Radeon RX 7800 XT beat the RTX 4070 in rasterization, reducing the RTX 4070 real-world pricing to around $560, which was pushed further down with the launch of the RTX 4070 Super at $600. This creates problems for the RX 7800 XT, sending it to below the $500-mark, causing a rather wide price-performance gap to emerge between the RX 7800 XT and the RX 7900 XT, which finds itself embattled at $700, squaring off against the $800 RTX 4070 Ti Super. To address this, AMD is now giving the RX 7900 GRE a wider market launch in western markets. For AMD, all this really takes is to scale up production and provide global shipping.



The Radeon RX 7900 GRE is a very interesting graphics card. It's based on the latest RDNA 3 graphics architecture, but has the same 5,120 stream processors and 16 GB of GDDR6 memory across a 256-bit memory bus as the previous-gen RX 6900 XT flagship. AMD took an interesting approach at the silicon level to create this desktop SKU. While the Navi 32 silicon was maxed out to create the RX 7800 XT, AMD decided against using the original Navi 31 to create the RX 7900 GRE, probably because it doesn't have enough imperfect large Navi 31 chips to create a SKU with fewer shaders and MCD chiplets than even the RX 7900 XT. Instead, AMD tapped into the physically smaller variant of Navi 31 that the company designed for mobile RX 7900 series SKUs.

The new Navi 31 mobile silicon has a package size resembling that of Navi 32, but with the larger Navi 31 chiplet GPU on top—this includes a full-sized 5 nm graphics compute die (GCD), surrounded by four functional 6 nm memory cache dies (MCDs). Physically there are six, but two are either disabled, or are structural dummies. The smaller PCB substrate has fewer pins, which include physically fewer memory IO pins to drive a memory bus no wider than 256-bit; and fewer electrical pins, which should give AMD's product managers limits to how many compute units (CU) they enable on the GCD, and what power budget they can give the card.

AMD went about carving out the RX 7900 GRE from Navi 31, by enabling 80 out of 96 CUs, which results in 5,120 stream processors, 160 AI accelerators, 80 Ray accelerators, and 320 TMUs. Out of 192 ROPs present on the silicon, 160 are enabled. Each of the four available MCDs has a 16 MB segment of the 64 MB Infinity Cache memory; and a 64-bit portion of the 256-bit memory bus that drives 16 GB of GDDR6 memory. AMD chose the same 18 Gbps GDDR6 memory it used on the RX 6950 XT; while the GPU runs at a game clock of 1880 MHz, with 2245 MHz boost. The board power limit is set to 260 W, which is lower than the 300 W for the RX 7900 XT, and 355 W for the RX 7900 XTX.

The RX 7900 GRE is well spaced apart from the RX 7900 XT and the RX 7800 XT in specs. It has 80 CU compared to 84 CU of the RX 7900 XT, which may not seem like a big gap, until you consider that the RX 7900 XT has a wider 320-bit memory bus pulling 20 GB of 20 Gbps memory for a 38% higher memory bandwidth, besides the chip's full 192 ROPs. The RX 7800 XT may have a similar memory configuration to the RX 7900 GRE, but has 25% fewer shaders, with just 60 CU, and 96 ROPs.

Driving the Radeon RX 7900 GRE is the RDNA 3 graphics architecture. Designed to take advantage of the new 5 nm foundry process, RDNA 3 debuts new generation dual-issue instruction rate compute units with a 17% IPC gain over RDNA 2, the new AI accelerator, which prepares matrix math for the stream processors, offering significant speedups for AI DNN building and training; support for new math formats and instructions; and the 2nd generation Ray accelerator, with a 50% increase in ray intersection performance. RDNA 3 also introduces MDIA (multi draw indirect accelerator); a hardware component that provides a 2.3x speedup for specific DirectX 12 draw commands, and reduces CPU API and driver overhead, the new AMD Radiance display engine that supports DisplayPort 2.1 UHBR13.5 and HDMI 2.1b, and dual hardware encoders for AV1 and HEVC.

We have with us the Sapphire Radeon RX 7900 GRE Pulse, which we are treating as our de facto reference design review. There are made-by-AMD (MBA) reference design RX 7900 GRE cards out there, but those are restricted to the OEM channel, we neither could find them in retail, nor did AMD send us one. The Sapphire Pulse comes with a small factory overclock over the AMD baseline specs and features a well-balanced combination of a low-noise cooling solution, and a price that matches the $550 AMD MSRP.

Short 6-Minute Video Summarizing the RX 7900 GRE Pulse Review Results

Our goal with the videos is to create short summaries, not go into all the details and test results, which can be found in this written review.