On the occasion of the GeForce Beyond event, Nvidia unveils the Ada Lovelace GPU architecture, the chip that is integrated into the GeForce RTX 4090 and 4080. On the program: finer engraving, the arrival of DLSS 3 and dozens of new billions of transistors. What significantly increase performance in video games.
On the occasion of its GeForce Beyond event and the launch of the GeForce RTX 4090 and 4080, Nvidia lifts the veil on its new Ada Lovelace architecture. It is this architecture that is integrated into the new GeForce RTX 4090 and 4080. The great novelty is the fineness of engraving which goes from 8 nm to 4 nm. In detail, it is the TSMC 4N engraving process.
Double energy efficiency
A finesse of engraving which allows to climb to a total of 76 billion transistors on the chip allowing to integrate more than 18,000 CUDA cores. Compared to the 28 billion transistors of the Ampere architecture of the GeForce RTX 3000. The RAM uses GDDR6X technology, like the old architecture unveiled in 2020 and therefore still not using stacked HBM memory like its competitor AMD.
Nvivia announces double energy efficiency on the Ada Lovelace GPUs of the GeForce RTX 4000 compared to the old Ampere architecture of the GeForce RTX 3000.
Shader processing is evolving with new streaming multiprocessors. These would thus benefit from energy efficiency twice as high as before, with a power of 83 teraflops on the shaders, against 40 teraflops on the Ampere architecture. Ray-tracing is also evolving, with third-generation cores, making it possible to reach a power of 191 teraflops on ray-tracing. It was necessary to count on a power of 78 teraflops for the GeForce RTX 3000.
Finally, Deep Learning is also entitled to improvements, with fourth-generation Tensor cores. Enough to go from a power of 320 teraflops to 1,320 teraflops on Ada Lovelace using the FP8 acceleration.
In total, the Ada Lovelace architecture would be very power-hungry, with a power that could reach 450 watts for the most powerful GPU, against 350 watts for the Ampere architecture.
Improved ray-tracing performance
We can also mention the new Shader Execution Reordering (SER) function which “ improves execution efficiency by rescheduling shading workloads on the fly to better utilize GPU resources“. Nvidia thus promises an improvement in ray-tracing performance “ up to 3x and frame rate in games up to 25%”
DLSS 3: THE function not to be missed
The other great novelty is the arrival of DLSS 3. Presented by Nvidia as ” the next revolution in the company’s Deep Learning Super Sampling neural graphics technology for games“. This is a major evolution of DLSS 2 using AI to speed up calculations in demanding games. Nvivia took the example of Cyberpunk 2077 which goes from 62 FPS in 4K in DLSS 2 to 101 FPS in DLSS 3, thanks to the Ada optical flow accelerator, twice as fast according to the American firm.
The big news is the remastering of Portal with DLSS, called Portal with RTX, free for Portal owners.
Nvidia is also addressing the case of video editing and streamers with the integration of new eighth generation AV1 encoders: it has two in parallel. Encoding efficiency would have been increased by 40%. Which would mean an increase in visual quality for the same bit rate.
GeForce RTX 3000 vs RTX 4000: performance
Finally, Nvidia announces four times higher performance for the GeForce RTX 4090 compared to the GeForce RTX 3090 Ti on Cyberpunk 2077, and twice as high for Microsoft Flight Simulator and Warhammer 40,000: Darktide. All this thanks to the new architecture, but also thanks to DLSS 3, which will be exclusive to this new architecture Ada Lovelace.
Finally, here is a graph summarizing the performance on Resident Evil Village, Assasin’s Creed Valhalla, The Division 2, Warhammer 40,000: Darktide, Microsoft Flight Simulator, Portal and Cyberpunk 2077 with the RTX 3090 Ti, RTX 4080 (12 GB), RTX 4080 (16 GB) and RTX 4090.
To follow us, we invite you to download our Android and iOS app. You can read our articles, files, and watch our latest YouTube videos.