Date: Thursday , February 14, 2019
On February 7th, 2019 AMD launched the Radeon VII video card with an MSRP of $699.99. This is the first 7nm GPU gaming video card with 16GB of HBM2 memory at 1TB/sec (Terabyte per second) of memory bandwidth. These are key features for AMD, as the industry moves toward the superior manufacturing processes below 14nm and 12nm with smaller dies and with higher capacities of VRAM and memory bandwidth.
The AMD Radeon VII is an unexpected video card from AMD. No one saw this video card coming, it was theorized that AMD would most likely wait until its next generation NAVI GPU architecture to bring new gaming GPUs to the desktop. However, AMD surprised everyone at CES by announcing the AMD Radeon VII based on a slightly evolved Vega architecture, but with a die shrink and tons of VRAM and memory bandwidth with higher clock speeds.
This is in fact not a new architecture, it is an evolved AMD Radeon RX Vega 64/56 architecture. Where Vega 64’s GPU is codenamed "Vega 10" GPU, this is "Vega 20." Right now, this appears to just be a one-off, that is there are no other "Vega 20" products planned, the Radeon VII is it.
AMD is marketing this video card as a gaming video card, but it is also clearly being marketed as a "content creators" video card as well. In fact, the official press deck (and our briefing) brings up all the benefits for content creators before it even talks about gaming benefits, so take that for what it’s worth. We were heavily inundated on how this video card can help professionals and content creators. Though we are going to focus on the gaming performance.
Here are the key differences with Radeon VII versus Radeon Vega 64. The RVII costs $699 direct from AMD and is aimed right squarely at NVIDIA GeForce RTX 2080 performance. This means it should be slightly faster than a GeForce GTX 1080 Ti. Up until now AMD has had nothing to provide gamers looking for that level of performance. The previous Radeon Vega 64 was aligned with GeForce GTX 1080 performance, they were neck and neck. AMD will now supposedly have something that competes with NVIDIA’s high-end, and allow gamers to enjoy 4K gaming with AMD hardware. Previously the Radeon Vega 64 was not fast enough to provide a true 4K gaming experience.
The Radeon VII is based on a 7nm manufacturing process, the first 7nm gaming GPU. With this die shrink AMD is able to crank up the clock speed on the Radeon VII far beyond the Radeon Vega 64’s capabilities. In addition, thanks to the die shrinking space saved AMD is able to put a total of 16GB of HBM2 memory on board with a much wider bus width than found on the Vega 64.
AMD has also done some actual low-level chip tweaks to the Vega architecture for the Radeon VII. This includes what AMD is calling enhanced Async Compute performance. Through optimizations in adding additional accumulators and reducing latency combined with increasing frequency and bandwidth Async Compute performance should be better. Reducing latency, increasing clock speeds and memory bandwidth seemed to be the key changes. There are actually more transistors in Radeon VII (13.2B) versus Vega 64 (12.5B.)
In terms of specifications there are 60 Compute Units (versus 64 on Vega 64) and 3840 Stream Processors versus 4096 on Vega 64. There are still 64 ROPs, same as Vega 64. From a pure compute unit standpoint, there are less on the new Radeon VII. The big difference is that the RVII clock speeds are ramped up, along with memory bandwidth as well.
AMD Radeon VII is introducing a new level of GPU Clock speed reporting. You will have the "Base GPU Clock" and the "Boost GPU Clock" and now there is a new "Peak Engine Clock" being introduced. The "Base" clock will be 1400MHz. The "Boost" clock will be 1750MHz and the "Peak" clock will be 1800MHz. All this means is that in certain workloads the GPU may be able to achieve a "Peak" clock speed of 1800MHz. We are told this is in more "Content Creator/Professional" application scenarios. Whereas for gaming, the "Boost" clock of 1750MHz will be the common frequency.
The other big feature with this video card is the inclusion of 16GB of HBM2 memory at 1TB/sec of memory bandwidth. The real question is if 16GB of VRAM is really going to help today’s gaming or if it is overkill. From a "Content Creator" or professional application scenario, it could very well help, but we are focused on the gaming performance. All of this is wrapped up in a TDP of 300W, slightly above Vega 64.