AMD’s Radeon RX Vega GPU is the company’s most significant consumer launch in at least two years. High-end consumer cards don’t own the mass market and they aren’t as significant a profit center as professional GPUs, but they’re important halo products in their own right. One user has begun posting 3D benchmark results for the new GPU — but they need to be taken with a huge grain of salt.
First and foremost, it’s important to keep in mind Vega’s Frontier Edition is explicitly intended for professional markets, not gaming. This matters when considering the GPUs current 3D gaming performance, because AMD’s earliest driver iterations are going to be focused on the apps and programs where the card is expected to compete. For consumer cards, that’s going to be games — but this is the first GPU launch where AMD has led with a professional card, and that means 3DMark optimizations probably aren’t where the company is going to focus.
Second, previous comments from AMD have implied that Vega Frontier Edition is running at a lower clock than the consumer variants will. This is unlikely to have a huge impact on the final product, but if Vega FE runs at 1.6GHz and the consumer RX Vega is clocked at 1.7GHz, that could translate to a 3-6% difference in GPU performance depending on the game. Greater clock speed differentials will, of course, translate to larger performance gaps.
Define’s own comments imply that power management may not be completely nailed down in the latest driver for the GPU but that performance in professional applications is excellent, “almost on par with Quadro P6000 and P5000 with the same settings. Biggest difference is that VEGA FE price was £977 and is running in my PC and both Quadro benchmark results are taken from spec.org as follows: P6000 and P5000.”
The P6000 is currently listed for between $ 4,700 and ~$ 5,800 according to Google, while the P5000 is $ 1,849 – $ 2000. A £977 price for Radeon Vega Frontier Edition would work out to a $ 1,263 price tag; a $ 999 list price is actually expected for US cards. AMD is clearly playing a value angle here, albeit at a high price point relative to the consumer market.
First, here’s a table of all three sets of results (the two official SPEC runs and Define’s Vega FE tests. Keep in mind that as a professional GPU benchmark, ViewPerf is typically designed not to vary much based on CPU clock or core count, though this doesn’t mean some variation is impossible). Define does note that he’s using a different CPU than the official SPEC results (a Core i7-4790K), and that these tests are run at the official resolution of 1900×1060 as opposed to a 4K data set that an updated version of the same test might use.
While we’ve accurately quoted Define above, I wouldn’t really agree that the FE is particularly well positioned against the P6000 on performance alone. The P5000 is the better point of comparison, so we’ve gathered up this data and color-coded which GPU wins and by what margin.
Again, keep in mind that this is very early days for the Vega FE and driver updates could meaningfully improve these scores in the future.
What about gaming tests?
The only gaming results seem to be from FireMark (regular, Extreme, and Ultra). Overclockers3d report scores of 22,916, 9,449, and 5,216. This would put the Vega FE closer to par with a GTX 1080 as opposed to the GTX 1080 Ti. The 3DMark Fire Strike Ultra score of 5216 is roughly 1.32x faster than Fury X’s performance in the same benchmark.
But again — with extremely early drivers that haven’t been optimized for consumer games, and a potential clock deficiency relative to the consumer part, I just wouldn’t put a ton of weight on these scores. Pricing is also going to be critically important to evaluating the consumer card’s overall attractiveness, and we don’t have any guidance on that issue yet.
Overall, based on these early figures, I’d say AMD has the beginnings of a potent GPU. We’ll need to see how driver updates evolve the product, but it’s certainly better than anything Team Red has fielded since 2015. And we haven’t seen the card stretch its legs in AI or HPC benchmarks, where GCN was historically a strong performer.