Earlier this week, various media of the Southeast Asia (including myself) were invited by NVIDIA here in Kuala Lumpur to see what is essentially a showcase of the GPU giant’s latest innovations in the consumer space so far. No particular announcements were made in this event, although we get to see some of the newly-introduced features in action.
NVIDIA Is All-In On Neural-Based Graphics
At least this sub-title is our takeaway from the comments made by NVIDIA’s representatives, as they pointed out that a majority of gamers enabled DLSS in their games when available to boost framerates significantly. With monitors now far outpacing the raw power GPUs of today can provide, they believe that DLSS is a way for gamers to fully utilize the display’s capabilities (assuming you own one of those high-end 1440p 360Hz or 4K 240Hz displays, that is).
While the demonstrations (primarily featuring RTX 50 gaming laptops) are generally pretty by the book, one worth pointing out is Alan Wake 2 running on one of these laptops, featuring what NVIDIA referred as “RTX Mega Geometry”, which is essentially a way to simplify object complexity to reduce the computational workload required for ray-traced graphics.
Naturally, this demonstration has all the DLSS goodies applied to boost performance, but looking at the metrics up close we saw a framerate of 170 FPS or so, while the latency is approaching 100ms (meaning the native framerate is no more than 10 FPS). While Alan Wake 2 is not exactly falling under the competitive shooter category, I don’t think input lag of this extent is anywhere near enjoyable either.
I do not have the luxury of time on my side to fully investigate the reasons behind this latency deficit, but if I have to make an educated guess, perhaps the GPU’s VRAM has been exhausted or the graphics settings are too demanding to begin with. Are games getting too difficult to run these days, or are GPUs not keeping up fast enough? It’s a question worth debating for sure.
Still, that bit of nitpicking aside, some of the features which utilize neural computing are definitely interesting prospects, with Neural Rendering being one promising solution that can at least fix some of the VRAM woes seen in mid-range GPUs today (we really should’ve moved on from 8GB, though). Another one is Neural Radiance Cache, which aims to reduce computational demands of path tracing workloads.
Outside of gaming uses, NVIDIA also presented the common use cases with their GPUs, including running LLMs (large language models) locally for increased data privacy, and its Broadcast suite of features with all the AI features you see in laptop webcams today, and then some. We also saw one of the upcoming laptops feature the unreleased RTX 5070 Laptop GPU, which you should be able to see in retail storefronts soon.
While that is all from the event itself, keep in mind NVIDIA will be on Computex in less than a month’s time – the yearly trade show will once again have Jensen Huang himself do the honors, and we’ll be there to see what’s new from Team Green, along with other tech companies showcasing their latest innovations.
