NVIDIA has officially announced the GeForce RTX 40 series video cards, based on the new Ada Lovelace architecture. The new chips are made in TSMC factories on a 4nm manufacturing process, and the performance improvements from generation to generation are, according to NVIDIA, up to 4X. Of course, there are certain games and certain conditions for these improvements to be achieved. The company has announced three video card models to start: the Geforce RTX 4090, the new flagship, the GeForce RTX 4080 16GB and the RTX 4080 12GB. Although the two XX80 models have similar names, they are more different from each other than we expected.
GeForce RTX 4090 could deliver a 400% performance boost over RTX 3090 Ti
The GeForce RTX 4090 is the new gaming “monster” from NVIDIA, promising the biggest next-generation performance boost yet. NVIDIA has compared the performance of this GPU to that of the flagship RTX 3090 Ti so far, and promised framerates between 2 and 4 times higher in certain titles.
The new GPU is based on a 16,384-core CUDA chip, running at a base frequency of 2.23 GHz, with boost to 2.52 GHz. With 24GB of 384-bit bandwidth GDDR6X memory, this is the most powerful gaming graphics card yet. Curiously, however, the memory has not increased to 32 GB, being at the same capacity as the previous generation.
Internally, the architecture has not changed significantly, but the dedicated cores for ray-tracing and AI processing (Tensor), have reached 3rd and 4th generation respectively. Thus, in ray-tracing games, the differences will be immediately noticeable, as there is now an ability to process “rays” almost three times faster. And the new Tensor cores can now process more information faster to enable the new DLSS 3 technology.
DLSS 3.0 differs from generations 1 and 2 by being able to create intermediate frames between those that the GPU creates. So not only will you be able to improve the resolution of a game by upscaling with AI, but you’ll be able to improve framerate directly by generating extra frames. So not only games that were limited by GPU performance will benefit from improved framerates, but also those that are limited by CPU performance. With the RTX 4090, NVIDIA aims to improve framerates in particular through DLSS 3.0, allowing top games such as Cyberpunk 2077 to run in 4K at very high frame rates (120 Hz). Even in 8K we could say that some games will be “playable” at 60 FPS. It remains to be seen what the real performance contribution will be when the boards hit the market.
But all these improvements come with a huge power consumption. NVIDIA recommends an 850W power supply for a GeForce RTX 4090 system, with the video card alone having a claimed 450W at full load. Power is supplied with three 8-pin PCIe cables, or a new generation PCIe Gen 5 cable. Pricing starts at 9,499 lei, according to NVIDIA’s website, and will launch on October 12.
RTX 4080 models come in two variants with different performance
In the case of the GeForce RTX 4080, we have a two-way discussion. Instead of releasing two cards with different names, i.e. RTX 4080 and RTX 4070, NVIDIA preferred to name them the same, but equip them with quite different hardware. It’s possible that both are based on the same chip at their core, but it’s clear that one is not as high performing as the other. Also, the memory has less bandwidth on the lower model.
So we have the top model, the RTX 4080 16 GB, which integrates a 9,728 CUDA core chip and comes with a base frequency of 2.21 GHz and boost to 2.51 GHz, with 16 GB GDDR6X with 256-bit bandwidth. This model will start at 7,129 lei and will be released sometime in November. The lower model, with 7,680 CUDA cores, 2.31 GHz frequency and boost to 2.61 GHz comes with 12 GB GDDR6X with 192-bit bandwidth, priced from 5,349 lei. Even the power consumption is different. The 16GB model recommends 750W power supply and has a claimed power consumption of 320W (power supply with 3 8-pin cables or one PCIe Gen5), while the 12GB model recommends 700W power supply for claimed power consumption of 285W (power supply with 3 8-pin cables). The launch of these models could be simultaneous.
Theoretically, the base model, with 12 GB, should be able to outperform the RTX 3080 Ti of the past by almost twice in some games when ray-tracing and DLSS 3.0 technologies are used. practically, even at raster rendering level, the boards have higher claimed power of 40 and 50 TFLOPS respectively, compared to the 3080 Ti with 34 TFLOPS. The RTX 4090 has about 82 TFLOPS, so it could be twice as powerful even compared to the new 4080 with 12 GB.
Of course, these prices are only for Founder’s Edition variants, produced and sold directly by NVIDIA. Boards from partners like ASUS, GIGABYTE or MSI will most likely come at higher prices, but also with higher performance profiles. Thus, we expect even higher power consumption than the official specs for overclocked variants.