The AMD Radeon VII’s Core Configuration Has Been Misreported

CES is always a hectic time for reviewers and companies — I’d argue it’s one of the busiest weeks of the year if you’re a tech reporter. Because of this, it’s not unheard of for incorrect information to leak into the channel, despite the best efforts of all involved.

In this case, there’s a correction we have to issue regarding AMD and its just-announced Radeon VII. Earlier today, wfoojjaec ran a story claiming that this GPU would have a core configuration of 3840:240:128 (GPU cores:texture units:render outputs). This information was based on reporting from other sites who have attended the show and attested to the accuracy of this information. The data was reported in the context of AMD disclosing further details about the GPU while at the show, not as rumor or unverified reporting, which is why we didn’t present the usual caveats when we gave this data.

According to an AMD spokesperson we’ve since spoken to, the number of reported ROPs for the Radeon VII is incorrect. The 128 figure, while widely and credibly stated, is wrong. “Radeon VII is 64 ROPs,” the AMD spokesperson stated to us. The GPU’s actual configuration is therefore 3840:240:64.

The good news is, this explains some things. It wasn’t at all clear why AMD would double up ROPs but leave GPU core count and texture unit resources unchanged. Typically, AMD and Nvidia maintain a rough parity between resources within a given GPU family, reducing or increasing the number of GPU cores, texture units, and ROPs equally. There are occasional exceptions to this, to be sure, but of the four basic resources: GPU cores, texture units, ROPs, and memory bandwidth, AMD had supposedly doubled or more-than-doubled two, while the number of GPU cores and the GPUs texturing capability remained identical apart from improved performance based on clock.

This suggested one of two things. Either the Vega 64 wasn’t remotely well-balanced in terms of resources, or Vega 7nm had been tweaked to meet the requirements of some specific customers. Given that AMD is heavily invested in its semi-custom business and the 7nm variant of the core launched first in the enterprise and HPC segment, the latter seemed more likely than the former.

We don’t know yet if AMD reps at the show were misinformed or exactly how the incorrect evidence spread, but wfoojjaec regrets the error.