An update recently released for Linux contains details that hint at what AMD’s upcoming RX Vega graphics cards have to offer, according to HotHardware.

The Linux direct rendering manager (DRM) update revealed that a Vega GPU will come equipped with 64 next-generation compute units resulting in 4,096 stream processors. For comparison, the newly released AMD Radeon RX 580 has 36 compute units and 2304 stream processors. If we’re talking about teraflops (TFLOPS), the Radeon RX Vega would offer 12.5 TFLOPS, which is a substantial jump from the RX 580′s 6.2 TFLOPS.

Another significant change with Vega is its memory structure; it will be using second generation high-bandwidth memory (HBM2) instead of the GDDR5 memory of Nvidia’s Pascal cards and the RX 400 and 500 series from AMD. HBM2 achieves higher bandwidth, uses less power, and is physically compact since the actual dies are stacked on top of each other. This technology was also used in a few of AMD’s Fiji graphics cards: the R9 Fury X, R9 Fury, and R9 Nano.

In addition, there were leaked results from what seemed like a Vega-based GPU running the 3DMark benchmark Time Spy 1.0. According to Guru3D, the benchmark results gave the unreleased Vega GPU a 5721 score. It is unknown where this particular card is in the pecking order within the Vega line, but it looks to compete with the mid-high end GTX 1070. For context, the Nvidia GTX 1070 scored 5699, the GTX 1080 scored 7152, and AMD’s RX 580 scored 4388.

For more on gaming performance with the latest graphics cards, check out our reviews of the AMD Radeon RX 580, RX 570, and Nvidia’s latest in the GTX 1080 Ti.