Nvidia today announced three new graphics cards based on its next-generation graphics architecture. Available starting next month, the prices for these will start at $899.
At the top of the stack is the new RTX 4090. This massive new GPU features 16384 CUDA cores with boost clocks that go up to 2.52GHz. The card comes with 24GB of GDDR6X memory and a 384-bit memory interface. Nvidia claims it is 2-4 times faster than the RTX 3090 Ti. The RTX 4090 has a power rating of 450W and runs on a single 16-pin PCIe Gen 5 or 3 x 8-pin PCIe cable. It’s priced at $1599 and will be available starting October 12.
Next is the RTX 4080, which comes in two variants. The more premium variant features 9728 CUDA cores and a 2.51GHz boost clock. It has 16GB of GDDR6X memory with a 256-bit interface. Nvidia claims it is 2-4x faster than the RTX 3080 Ti. Then there’s also a second variant with 12GB of GDDR6X memory and a 192-bit memory interface. This version has 7680 CUDA cores and a 2.61GHz boost clock.
The 16GB model has a power rating of 320W and the 12GB model of 285W. They are priced at $1199 and $899, respectively, and will be available in November.
At the heart of these new graphics cards is the new GPU. Based on Nvidia’s new TSMC 4nm Ada Lovelace architecture, the new GPU features 3rd generation RT cores, 4th generation Tensor Cores, and support for new features such as DLSS 3 and AV1 encoder.
Compared to the 40 shader-TFLOPS of the RTX 3090 Ti, the RTX 4090 has 83-TFLOPS. Thanks to the new Shader Execution Reordering, shader performance is claimed to improve by 2x and in-game frame rates by 25%. Nvidia also claims that the RT performance on the RTX 4090 is around 191-TFLOPS, while the RTX 3090 Ti was capable of 78-TFLOPS. According to reports, the new 4th generation Tensor cores have 5x the throughput of the RTX 4090.
The 40-series cards will also feature two of the 8th generation Nvidia Encoders (NVENC). These can not only decode AV1 but will also be able to encode it. The company has partnered with OBS Studio to bring AV1 encoding to its clients. Nvidia has also bumped the ShadowPlay capture resolution to up to 8K at 60fps in HDR.
One of the more impactful features announced was DLSS 3, which will be exclusive to 40-series graphics cards. The new version uses an Optical Flow Accelerator and data from sequential frames to create new frames on the GPU itself (rather than within the game) to improve frame rates by up to 4x. This method also bypasses any CPU bottlenecks that affect traditional upsampling and image reconstruction methods as the frame rate increases.
Over 35 games have been announced with support for this feature, including Cyberpunk 2077 and Microsoft Flight Simulator. These will start appearing next month. The company also announced a version of Portal with real-time ray tracing, which will be launching in November and be a free update for existing Portal owners. It’s part of the Nvidia RTX Remix modding platform, which features tools for improving the visuals of older titles.
Getting back to the graphics cards, the new 40-series models will replace the 3090 series of cards but will join the existing 3060, 3070, and 3080 series of cards, which will continue to be sold. We do not expect lower-end 40-series models until 2023.
Nvidia will be selling its own Founders Edition cards for the RTX 4090 and the 16GB RTX 4080, alongside models from AIB partners, such as ASUS, Colorful, Gainward, Galaxy, GIGABYTE, INNO3D, MSI, Palit, PNY, and ZOTAC. Most of these cards feature a 3–4 slot design along with beefy power requirements. The cards will have HDMI 2.1 and DisplayPort 1.4a connectivity.