The new 7nm Radeon VII GPU was an unexpected announcement at AMD CEO Dr Lisa Su's CES 2019 keynote last week, and very few actual specifications and details were disclosed at the event. More information has come trickling out over the past few days, some of which is interesting and some that might dampen the appeal of this new top-end gaming hardware. The Radeon VII will be the world's first 7nm gaming GPU when it launches on February 7. It will supersede the current top-end Radeon RX Vega 64, at least until AMD launches its Navi generation of GPUs, expected later this year.
First and most surprinsgly, TweakTown cites unspecified industry contacts who seem to suggest that the entire production run of AMD Radeon VII cards will be less than 5,000 units, and that there will be no custom cards from the usual partner brands. AMD has announced retail sales from its own website, but partners should be able to ship the same reference design units as well. A small production run would fit this GPU's niche status, with the timeline of Navi-based GPUs launching in the second half of this year.
The AMD Radeon VII is aimed at gamers who want smooth frame rates at 4K, or the emerging ultra-wide 3440x1440 resolution, or very high frame rates at full-HD. It will also be aimed at those who want to stream gameplay video online while playing, or those who want to create games and other creative content but don't need a pro workstation-grade GPU.
AMD claims a performance uplift of 29 percent on average compared to the Radeon RX Vega 64. While it has not published any comparisons with Nvidia's performance numbers, the Radeon VII can be expected to trade blows with the GeForce GTX 1080 and GeForce RTX 2080. AMD will not be able to top Nvidia's flagship GeForce RTX 2080 Ti with this product.
As for specifications, AMD has only shared a few details. The Radeon VII GPU is codenamed Vega 20, and has 3,840 stream processors and 240 texture units with 64 ROPs (render output units). Radeon VII graphics cards using AMD's reference design will feature 16GB of HBM2 memory on a 4096-bit bus for up to 1TBps of memory bandwidth.
The AMD Radeon VII is said to be practically identical to the 7nm Radeon Instinct MI50 accelerator which was launched in November 2018, and uses the same Vega 20 die and memory configuration. However, double-precision floating point operations will not be supported, most likely in order to differentiate the Radeon VII as a gaming GPU rather than a professional accelerator. The consumer card will also not support PCIe 4.0. On the other hand, the Radeon VII runs slightly faster at a base clock of 1450MHz and a boost clock of 1800MHz.
The TDP rating of the Radeon VII is the same 300W as the Radeon Instinct MI50, which is considerably higher than Nvidia's current offerings. The use of a triple-fan reference cooler does suggest that this GPU will run hot, which could limit its appeal.
Nvidia CEO Jen-Hsun Huang publicly dissed the Radeon VII as “lousy” and “nothing new” at CES shortly after its introduction, as quoted by PC World. He suggested that it will barely keep up with an Nvidia GeForce RTX 2080 GPU even though it's a 7nm GPU with HBM2 memory. Calling AMD's return to high-end GPU performance “underwhelming”, he cited the new GPU's lack of AI and ray tracing capabilities, which the recently launched GeForce RTX series deliver.
In response, AMD's Lisa Su only remarked that Huang hasn't seen the Radeon VII in action yet, and that game developers are not yet very involved in ray tracing. According to AMD CTO Mark Papermaster, ray tracing will be important in the future but there's a long development cycle to go through first.