RTX 40 Series
The new generation of RTX graphic cards is built upon a new 4nm manufacturing process developed by TSMC. NVIDIA designed the architecture, which was based on their new Ada Lovelace platform that was named after the famous English mathematician and the world's first computer programmer. Even though her work was purely theoretical, it was the first ever accurate application of a computing device, which she called the Analytical Engine.
The chips based on this design pack up to 76 billion transistors and 90 teraflops of processing power. The new design and smaller die requirements make it possible for them to pack twice the ray-triangle intersection throughput for double the performance of the last gen RTX cards. To help things along, they introduce a new streaming multiprocessor that sorts through the different rendering requirements and helps optimize the processes more efficiently.
Another new highlight is the introduction of a new Tensor Core that features the Hopper FP8 Transformer Engine, capable of generating 1.4 Petaflops of processing power. The great thing is that you get all these performance upgrades and more while still maintaining the same 450 watts power input as the previous generation Ti model.
The premier model of the third generation RTX series is the RTX 4090, which comes with 16,384 CUDA cores and 24GB GDDR6X memory. The Boost Clock Speed is set to 2520MHz, but we will have to see if we can further overclock the setup in future reviews. The other model on showcase was the RTX 4080, which comes in 16GB and 12GB models. They pack in 9,728 and 7,680 Cuda Cores, respectively, while also maintaining a healthy 320 or 285 watts of power input requirements.
All the aforementioned cards will be officially launched in October at a starting price of $900 for the base model 4080. Check out the full spec list, along with the prices down below.
DLSS 3
AI enhancements are a big part of graphical performance, which is why updating the software side is equally as important as the hardware. For this purpose, NVIDIA has released the third generation of their DLSS tech. DLSS 3 brings new levels of machine learning output to your games, with the ability to predict actual frames instead of just pixels. This means that the system can now understand the frame in its entirety and then decide the best areas to optimize.
Since the predictions are much smarter than before, they also take less time to generate. Especially since the company has combined three key existing technologies - DLSS Super Resolution, DLSS Frame Generation, and Nvidia Reflex - to train the AI on creating a more efficient workflow, with help from this new optical flow accelerator. This makes RTX-based DLSS 3 four times faster than traditional rendering tools, with the ability to give a 2-3 times push to your frame rates.
Even if you are playing a CPU-limited game, you can still enjoy an upgrade in graphical performance, thanks to these new tools as well as beefed-up hardware. The company is claiming that they will personally optimize more than 35 games to support these new features right from launch. For other developers, NVIDIA announced a new application called RTX Remix. They can use it to capture their game in a universal scene descriptor, and then optimize it for RTX gen 3 on their own. This means that almost any game other there can now get the gift of AI-based graphical output with the new RTX 40 series.
Graphic Cards Specifications
Type | Specification |
---|---|
RTX 4090 |
|
CUDA Cores | 16,384 |
Memory | 24GB GDDR6X |
Boost Clock Speed | 2520MHz |
Bus Width | 384-bit |
Power | 450W |
Price | $1,599 |
RTX 4080 16GB |
|
CUDA Cores | 9,728 |
Memory | 16GB GDDR6X |
Boost Clock Speed | 2505MHz |
Bus Width | 256-bit |
Power | 320W |
Price | $1,199 |
RTX 4080 12GB |
|
CUDA Cores | 7,680 |
Memory | 12GB GDDR6X |
Boost Clock Speed | 2610MHz |
Bus Width | 192-bit |
Power | 285W |
Price | $900 |