Nvidia introduced its newest technology GeForce RTX graphics playing cards at its GTC AI convention. The Nvidia RTX 4090 and RTX 4080 are the primary GPUs within the ‘Ada Lovelace’ sequence and must be twice as quick in rasterized video games and 4 instances sooner in ray-traced video games than their last-gen counterparts. .
The RTX 4090 24GB will probably be out there on October 12 for $1599, whereas the RTX 4080 16GB will launch in November and price $1199. There’s additionally an RTX 4080 12GB mannequin that prices $899. The 4090 and 4080 are reportedly two to 4 instances sooner than their Ampere equivalents, the 3090 Ti and 3080 Ti, respectively.
Word that Nvidia RTX 30 sequence playing cards will stay on sale for now and fill the decrease finish of the market till lower-end RTX 40 sequence GPUs arrive.
So how are these GPUs so quick? 40 sequence GPUs use TSMC’s ‘4N’ course of and have as much as 76 billion transistors and 18,000 CUDA cores; that is 70 % greater than the final technology Amps. The brand new course of makes the technology considerably extra energy environment friendly, however we anticipate flagship playing cards to be as energy hungry as they are saying – you get a ton of additional efficiency to offset that by way of effectivity.
The brand new stream multiprocessor makes use of a brand new method known as shader execution reordering (SER), which Nvidia claims offers a ‘two to a few instances acceleration’ for ray tracing and a 25 % enchancment for rasterized video games. SER works by dynamically rescheduling shading workloads to higher make the most of GPU sources. There are related enhancements in customized ray tracing silicon, with doubling the ray triangle intersection output, a brand new opacity micromap engine that doubles the velocity of “ray tracing of alpha check geometry”, and a micromesh engine that will increase geometric richness with out the price of BVH construct and storage. There may be additionally a brand new and extra highly effective Tensor core for AI duties.
It is also Nvidia’s first card to help DLSS 3, a brand new method that creates fully new frames – basically by including interpolated frames between ‘actual’ frames to dramatically improve body charges. This has a detrimental impact on enter lag, so it is mixed with Reflex to scale back enter lag as a lot as doable. This reduces the load on each the CPU and GPU, so you may see framerate advantages in each CPU- and GPU-limited video games – however the enter lag penalty means it will be much less helpful for aggressive FPS video games, for instance.
Nvidia confirmed Cyberpunk 2077 working at ~22fps in 4K with RT enabled and DLSS disabled, then ~100fps with RT enabled and DLSS 3 enabled – which is an amazing speedup, even when it is a cherry demo. The agency additionally performed a video demo of Flight Sim 2020 working at ~60fps with RT + DLSS disabled and ~135fps with RT + DLSS 3 enabled – good proof that the DLSS 3 method eases CPU limitations.
Along with upgrading present video games with RTX and DLSS 3, Nvidia introduced a ‘new’ title: Portal RTX. This remaster, which is 15 years after the unique launch, appears to be like spectacular from the brief teaser trailer. It is going to be launched at no cost to Portal house owners in November. The mod was created with an app that Nvidia calls ‘RTX Remaster’, which helps you to produce mods that add ray tracing and AI-enhanced textures for all kinds of video games – very neat.
Elsewhere, Nvidia has introduced Racer X, a ‘simulated world’ with unimaginable realism – suppose marble demos go as much as 11.
It was a powerful presentation from Nvidia and we won’t wait to check each the efficiency and options of the following technology. It is going to be fascinating to see if DLSS 3 lives as much as Crew Inexperienced’s lofty guarantees and if these new playing cards will truly be out there at their introduced costs if the efficiency claims are confirmed. We’ll share extra data as we get it!