When Nvidia warned on earnings last week, the company noted that one reason for the unexpected ~$ 500M shortfall was lower-than-expected sales of its Turing GPUs. The RTX family, which debuted in September, was significantly more expensive than the Pascal GPUs it replaced. Nvidia justified the price increases by appealing to the features these GPUs included, like RTX ray tracing (that’s Nvidia’s name for the DirectX DXR ray tracing capability) and its DLSS anti-aliasing.
Data from the Steam Hardware Survey suggests that Turing cards haven’t sold as well as their Pascal counterparts did — but before we dive into the data, we need to acknowledge some caveats.
Because Steam’s user base grows, percentile changes between periods do not automatically reflect the absolute number of GPUs being sold. If 5 percent of 10,000 customers own one card in 2016 and 2.5 percent of 20,000 customers own a card in 2018, the absolute number of GPUs sold would remain the same.
Second, there’s the fact that Pascal availability was sharply constrained during the summer of 2016, while Turing had far fewer problems. Pascal GPUs were genuinely difficult to find during the periods we’ll be examining (immediately post-launch), which ought to work in Turing’s favor when it comes to comparing uptake.
Third, we have to acknowledge shortcomings in the Steam Hardware Survey itself. Steam has a large “Other” contingent of cards, with some 10-13 percent of cards falling into it at any given time. Currently, “Other” sits at 10.62 percent. But Steam also doesn’t update the survey evenly. AMD’s RX 570 didn’t arrive on the list until December, despite launching in 2017. RX Vega didn’t show up until January 2019, despite an August 2017 launch date. Some of Steam’s calculated month-to-month shifts also don’t make tons of sense — why would Nvidia GeForce 650 GPUs see a 0.15 percent uptick in adoption in January 2019? Whether this reflects better measurements of which cards are installed in user systems or the process of sorting GPUs out of the “Other” category and labeling them more consistently isn’t clear; “Other” tends to hover around 10 percent on an ongoing basis.
In short, the data we’re going to examine should be treated as interesting and useful, but not the absolute final word on the topic.
One advantage of comparing Turing and Pascal is that both families had to live side-by-side with their older siblings for quite some time. In Pascal’s case, this was caused by 14nm production problems. Turing had to contend with a supply glut of Pascal GPUs. But in both cases, Nvidia didn’t immediately pull its older GPU family in the manner you’d expect from a more traditional launch.
The graphs below show RTX versus GTX adoption rates for both the GTX 1080 / RTX 2080 and the GTX 1070 / RTX 2070. They also show the GTX 1080 compared against the RTX 2070, since those GPUs are closer to being matched on launch price. The periods 1-5 on each graph denote the date from launch month to present day. For Pascal, this is May 2016 – September 2016. For Turing, it’s September 2018 – January 2019. Both data sets are available from Steam and the Wayback Machine respectively.
A table of the graphed results is below:
I don’t want to read too much into these graphs. We have a limited number of Turing data points to pull from and only a few months of data to work with. But the pattern suggests that Turing sales are indeed lagging Pascal in ways that may not be entirely explained by either higher prices, previous models still being in-market, or a larger user-base of Steam gamers.
Turing adoption should improve now that Pascal cards are exiting the market, but the GPU family appears to be off to a slow start. All of this is in-line with Nvidia’s market warning, but it’s interesting to see what the differences might be. There is a little good news in the SHS for NV, however — the RTX 2070 posted strong growth, nearly doubling its market share. Other strong movers (in terms of share gain) include the GTX 1070 (up 0.18 percent) and AMD’s RX 580 (up 0.15 percent)