The Radeon VII makes sense in the lens of Apple
Like many gamers, I felt thoroughly whelmed on reading reviews of the Radeon VII. Here we have a card coming in months after the Geforce 2080, for the same MRSP, with a generations fabrication process advantage and HBM2, and no silicon spent on special tricks like RTX and DLSS. And despite this, performing on average a few percent below the 2080, and at a higher power draw.
At first blush, this seems like little more than the GPU equivalent of a "FIRST" comment, a low effort rehash of the MI50 while being able to claim supremacy on TSMCs 7nm process.
But that's actually where it gets interesting, being largely an MI50 with a soft cap on FP64 (double precision math, essentially how many decimals over a calculation can carry, useful in some scientific compute fields), and no ECC memory.
At first, it was thought the VII would run at 1/8th single precision rate for FP64, but on feedback AMD released more double precision performance, now sitting at 1/4th SP.
The Radeon VII graphics card was created for gamers and creators, enthusiasts and early adopters. Given the broader market Radeon VII is targeting, we were considering different levels of FP64 performance. We previously communicated that Radeon VII provides 0.88 TFLOPS (DP=1/16 SP). However based on customer interest and feedback we wanted to let you know that we have decided to increase double precision compute performance to 3.52 3.46 TFLOPS (DP=1/4SP).
If you looked at FP64 performance in your testing, you may have seen this performance increase as the VBIOS and press drivers we shared with reviewers were pre-release test drivers that had these values already set. In addition, we have updated other numbers to reflect the achievable peak frequency in calculating Radeon VII performance as noted in the charts.
Despite still being soft capped, even 1/4th rate Double Precision leaves the VII as a unique standout in the space. Similar to certain Titan models, this seems more of a compute card under a gaming brand, rather than a pure gaming play.
So compute performance? Rather good
So where does this card existing make the most sense? Perhaps, racking up really well binned, lower power dies for the next iMac Pro, while we get the lesser binned parts. Reportedly, AMD isn't making all too many of these cards. This leaves one question - if the FP64 limit is a soft cap, and Apple writes much of their own video drivers under macOS, would the cap be removed under macOS, performing more like an MI50? "Pro" monikers on Mac hardware usually straddled the line between Pro and Consumer, lacking ECC, and Vega didn't have such delimiters between consumer and compute cards. So where the VII headed for Macs lands on FP64 performance will be more telling of whether they really deserve the "Pro" badge or are just rebranded, lower power consumer cards, as I've speculated was the case for the cylindrical Mac Pro and Macbook Pros.
If the iMac Pro does get a refresh with VII and does fully uncap FP64 performance, it may be a standout value for some compute use cases.