AMD's RDNA 4-equipped line of RX 8000 graphics cards may have been spotted during tests, but don't believe the numbers just yet.

AMD's RDNA 4 Architecture, rumoured to have been built on a smaller process and to be more energy efficient, could finally make its debut with the RX8000 GPU series. We've only just seen a glimpse of what it could look like.

Benchleaks, a X account (via Tom's Hardware), pointed out that a series of Geekbench benchmark tests had been uploaded to the benchmarking website. This suggests AMD's upcoming cards. The device names of the registered graphics cards tell us this.

RDNA 3 cards are labelled with the gfx11XX format, while RDNA 3.5 cards are labelled with the gfx115X. The next major line is RDNA 4, which uses the Gfx120X. Gfx1201 appears in five different tests. Four of them use the Intel Core i7-12700F processor and one uses the i7-13700K variant. These are desktop Intel CPUs, so we're testing a desktop card.

Before looking at the numbers and score, it's important to know that RDNA 4 will launch sometime next year (probably around CES 2025). Take everything with a pinch of salt. The OpenCL score reported of a low 30,000 in the five tests may have been exploratory, e.g. to compare early builds or to test different types and machines. It's not indicative of the final result.

This is a sign that the next generation of chips are being actively developed. And, since RDNA 4 may be included in the PS5 Pro it could put a time limit on the architecture.

Kepler_L2, a leaker who has a good track record in the industry, is the main source of information not found in Geekbench's score. This isn't their first time reporting about this series of cards. They hinted at boosted performance for ray tracing earlier this year. The leaker also posted a cryptic message that suggested three cards would have the following stats.

• Memory speed of 20 Gbps on a 256-bit bus
• Memory speed of 18 Gbps with 256-bit bus
• Memory speed of 19 Gbps on 192-bit bus

The GPUs currently tested have a memory capacity up to 16 GB.

These numbers are not very meaningful unless you can test them. The GTX 1070 of 2016 had a 256 bit bus bandwidth, but it used an old memory type instead of the GDDR6 that was expected for the 8000 series. Raw numbers can give you a good idea of its capabilities, but there are many important details that we don't have.

RDNA 4 will, however, be a step behind Nvidia’s next offering. This is because it is reported that the new product will use GDDR7. RDNA 4 GPUs, as reported last year, are not expected compete with high-end Nvidia equipment, but rather to be a mid-range offering. Even if Geekbench scores aren't good, those stats at least look healthy.