For the first time in five years, NVIDIA did not announce new GPU models.
It did unveil a particularly interesting update to a technology called DLSS, and the thread connecting both facts is artificial intelligence.
-
Graphics cards were originally invented to satisfy the insatiable demand of young gamers for the best possible graphics as they blasted aliens off their screens with laser cannons.
Since then, they've come a long way — today they power world-changing revolutions like Bitcoin and artificial intelligence — but gamers are still there, expecting their annual pound of performance in the form of record-breaking new GPUs.
This year, NVIDIA launched no new graphics cards.
The reason: NVIDIA cannot supply new cards at a reasonable price, because VRAM production lines are dedicated to manufacturing memory for far more profitable AI servers.
-
NVIDIA did announce a new and interesting version of DLSS.
DLSS technology began many years ago as a primitive AI application for improving graphics in PC games.
It allowed the GPU to render the display at a lower resolution, then synthesized the additional pixels from scratch so the end user received a higher resolution image.
In subsequent generations the technology improved further, gaining the ability to predict the motion of objects on screen and the projection of light from virtual light sources onto the game scene — dramatically improving display quality.
The new version marks the first transition from a CNN engine to a Transformer engine — an architecture similar to the one powering modern AI models.
What makes the Transformer model special is its ability to "understand" the relationship between different parts of the screen simultaneously. If it detects the reflection of a person in a puddle, it understands who the reflected figure is and delivers far more accurate, sharper results.
-
NVIDIA CEO Jensen Huang made a significant statement during an interview at the conference.
According to him, improving capabilities going forward will no longer rely on more powerful hardware, but on AI-based DLSS that will generate graphics from scratch while consuming far less energy and processing resources.
Instead of rendering a scene of smoke or reflections and performing enormously complex calculations for how they move on screen, the AI will receive only a basic graphical skeleton and construct the final scene on top of it.
-
All of these points converge into an interesting picture:
- Demand for AI has caused a RAM shortage that prevents the launch of new graphics cards.
- NVIDIA is using technology to deliver meaningful performance improvements without the need for new graphics cards.
- That technology is built on artificial intelligence — the very thing that caused the shortage in the first place.
And once again, NVIDIA demonstrates a breathtaking ability to maximize a situation to its own advantage, and to define what the future will look like rather than merely trying to predict it.
Video: From the DLSS 4.5 announcement | Source: NVIDIA
--
👋 Hi, I'm Shlomo Strauss — follow me for more interesting content on science and technology.