A bunch of AMD Radeon 8000 specs have just apparently been leaked online, potentially giving us a bigger picture of what’s to come with the forthcoming AMD RDNA 4 graphics card launch. All four of the new cards appear to be based on the Navi44 and Navi48 chips that have already appeared in listings elsewhere, but it looks like we now also have some information about the VRAM speeds, cache, and memory interfaces.
AMD is expected to release its new lineup of RDNA 4 GPUs at CES in January 2025, with rumors pointing to the company focusing on mid-range performance with this series of GPUs, leaving the ultimate best graphics card title to the Nvidia RTX 5090.
This latest rumor comes from regular tech leaker All The Watts!! who, as usual, has provided a cryptic series of numbers in a post on X (formerly Twitter). It looks as though six products are detailed, and All The Watts!! has helpfully provided some emoji clues to the last two. One has an owl and an angel, which we assume means Strix Halo CPU with its mighty integrated GPU, and the other has a US flag and a valley, which we assume refers to AMD’s rumored Sonoma Valley mobile chip, which is expected to be less powerful than Strix Halo.
The top four products, however, look very much like they might refer to new RDNA 4 graphics cards, so let’s take a look at them. The first set of numbers in each entry is either a 44 or a 48, which we can only assume refers to Navi44 and Navi48. If this is true, we’re looking at the specs for four GPUs based on Navi48, and one based on Navi44. Next up is a three-digit number, which is either 128, 192, or 256. These look very much like the figures for the width of the memory interface on each GPU, with a 256-bit bus providing more bandwidth than a 128-bit bus.
The third set of numbers is trickier. They each have two digits, which are either 64, 48, or 32. This could potentially refer to the amount of L3 cache in each GPU in megabytes. However, it could also refer to the number of compute units enabled in each GPU. None of the numbers are 56, though, and we’ve already seen a Navi48 GPU with 56 compute units appear in the Geekbench results browser. This makes us think these figures are more likely to refer to the amount of L3 cache in the GPU.
Finally, the fourth set of numbers each have two digits, which are either 18, 19, or 20, and these figures look very much like they’re referring to the effective speed of the VRAM in GHz (or, more accurately, Gbps), as also shown in a recent post by Kepler_L2. What does this all mean? Well, if we’ve correctly identified these specs, and AMD sticks to its previous naming conventions, we can start to paint a picture of how these new cards might look.
With all this in mind, these are the estimated AMD Radeon 8000 specs:
| 8800 XT | 8800 | 8700 XT | 8600 XT | |
| GPU | Navi48 | Navi48 | Navi48 | Navi44 |
| Memory interface | 256-bit | 256-bit | 192-bit | 128-bit |
| L3 cache | 64MB | 64MB | 48MB | 32MB |
| VRAM speed | 20Gbps | 18Gbps | 19Gbps | 18Gbps |
There’s not much difference between the specs of the top two GPUs here, but AMD could differentiate these two cards further by disabling a few of the compute units on the GPU – this could make the difference between an XT with 60 CUs and non-XT GPU with 56 CUs, which might be called the Radeon RX 8800 and 8800 XT, for example. You could then have a Radeon RX 8700 XT with perhaps 48 CUs, and an 8600 XT with 32 CUs. This is all pure speculation, though.
Indeed, bear in mind that absolutely none of this has come officially from AMD, and we’re still very much at the rumor and speculation stage. We’ll have to wait until AMD officially lifts the lid on its new GPUs before we know all the details for sure. In the meantime, check out our AMD Zen 5 guide, to see the details of AMD’s new CPU lineup.
