Nvidia held its annual Analyst Day yesterday, with a number of announcements and discussions of various aspects of its business. One particular claim the company made caught our eye. According to Nvidia, Turing sales revenue over the eight-week post-launch period was 45 percent higher than Pascal over the same time frame. The company also took care to highlight another fact: 90 percent of gamers who bought into RTX purchased a more expensive GPU.
We don’t doubt the factual truth of both of these statements, but there’s important context around both. Nvidia’s Pascal was quite popular from launch, but also difficult to find in-market. Turing availability was markedly better from the start, and the GPUs themselves commanded higher prices, even compared with the Founders Edition versions of Pascal hardware. Turing also launched later in the year, and closer to the Christmas holiday, whereas Pascal launched part way through Q2. Finally, Nvidia launched the RTX 2080 Ti immediately and alongside the other Turing cards and at a substantially higher price point than ever before.
Context matters, in this case, because Nvidia has already told investors that Turing sales were below expectations in the back half of the year. This agreed with publicly available data points from companies like Amazon and Newegg, where the best-selling GPU lists last fall were full of GTX 1080s and 1070s, with far fewer Turing cards making the cut. Nvidia may indeed be earning more money from Turing at launch than it did from Pascal, but the company’s own remarks indicate Turing didn’t meet Nvidia’s initial expectations. Nvidia’s gaming revenue also fell like a rock in Q4 2018 as part of the overall crypto hangover. This doesn’t speak to a strong launch period for its new architecture in terms of unit volumes, even if revenue rose.
With this in mind, the fact that 90 percent of gamers chose to trade up isn’t surprising. RTX GPUs don’t offer a performance improvement per dollar over their GTX counterparts. If you owned a GTX 1080 and bought an RTX 2070 to replace it, the only thing you got for $ 500 was the promise of RTX support in future games and a very small performance increase. I don’t know how many people are willing to drop $ 500 for a new GPU solely for features they can’t use yet, but I doubt it’s very many. If you bought an RTX card, you probably stepped up a GPU rank to do it.
As for the claim that 90 percent of Nvidia’s market is below the performance of a 1660 Ti? That looks pretty true. Using the Steam Hardware Survey’s numbers (and classifying the 1070 and 1660 Ti as equivalent), there are just four GPUs with any significant market share above the 1660 Ti’s performance: GTX 1070, 1070 Ti, 1080, and 1080 Ti. Add them up, and that’s 10.65 percent. The only other GPU that might beat the 1660 Ti would be the Maxwell-based GTX 980 Ti. (The original GTX 980 was almost exactly equal to the performance of a GTX 1060, so the 980 Ti has a shot at beating the 1660 Ti, though I haven’t done the comparison.) The number of GTX 980 Ti’s in-market is probably pretty small at this point. And since the Steam Hardware Survey isn’t identical to the total GPU market, I think we can call this one pretty straight.
And that’s kind of remarkable in and of itself. The GPUs that drive marketing forward are a far cry from the cards that most people practically purchase. While a full 50 percent of Nvidia’s customer base is using Pascal hardware, the bulk of them bought in at the 1060 level or below. Of course, the flip side to this is that it’ll be a long time indeed before RTX features meaningfully hit the mainstream. If 90 percent of Nvidia’s market is plugged in below the new $ 280 price point and RTX features start on $ 350 cards, the best RTX owners can reasonably expect is for 10-15 percent of their fellow gamers to have access to RTX by the time Nvidia’s next generation launches. That’s definitely enough support to drive some adoption, but not enough to build ray tracing into a mainstream feature this generation.