According to Nvidia’s CEO Jen-Hsun Huang, streaming services like GeForce Now are the best way to deliver AAA game experiences to mobile players. Ray tracing mobile chips, he claims, are a less effective way to meet that goal.
“Ray tracing games are quite large, to be honest. The data set is quite large, and there’ll be a time for it,” Huang said. “When the time is right we might consider it.”
Huang’s comments are in reference to AMD’s announcement earlier this week that the mobile GPU it has developed in partnership with Samsung will support features like ray tracing and variable rate shading.
On the one hand, this is exactly what we’d expect Nvidia to say. The company is unlikely to publicly declare AMD had stolen a march on it by bringing ray tracing and VRS to mobile GPUs, even if Nvidia privately believed this was the case.
It’s interesting, however, to see Jen-Hsun claiming the issue was game size (it is unclear if he meant VRAM requirements or in terms of storage capacity) as opposed to die size or power efficiency. Modern, high-end Samsung smartphones already offer 128GB to 512GB of storage and 8GB to 16GB of RAM.
The addition of ray tracing capabilities sent die sizes soaring, though it did so for different reasons. In Nvidia’s case, the company has added specialized hardware blocks for ray tracing and tensor calculations. AMD has added a large L3 cache as part of its strategy for maximizing performance.
Both of these strategies have increased absolute die sizes.
GPU die sizes have increased significantly with the advent of ray tracing for both AMD and Nvidia GPUs.
This trend holds up lower in the stack as well. The GTX 1060 was a 200mm sq. die, while the TU116 — its nominal replacement — was a 284 mm sq. chip. The 6700 XT is substantially larger than the 5700 XT it replaces.
We still have more questions than answers regarding the future role of this IP across AMD and Samsung products. The originally announced agreement only states that Samsung will sell the graphics IP in “mobile devices, including smartphones, and other products that complement AMD product offerings.” Previous reports have suggested Samsung’s RDNA-derived IP will be limited to smartphones and tablets. It isn’t clear if AMD will be building any chips of its own, but a new low-power GPU could theoretically be of use to the company in embedded, low-end mobile, and ultra-low-power segments.
Samsung will likely be the first mobile company to ship a ray tracing-enabled GPU to consumers, though there’s no telling when we’ll see games or apps optimized to take advantage of it. We may need to see broad ray tracing support across the entire market first. Jen-Hsun’s reference to potentially considering such a chip is also interesting, given that Nvidia walked away from the consumer mobile market some years ago to focus on automotive. It’s not hard to imagine Nvidia developing its own graphics solutions to pair with ARM CPUs and making those chips available for licensing if the deal is allowed to go through.
Now Read: