Nvidia to release GeForce RTX 4090 in October, RTX 4080 GPUs to follow in November
Nvidia has announced its first GeForce RTX 40 video cards. These will be based on a new Ada Lovelace architecture. The company comes with an RTX 4090 for 1959 euros and two RTX 4080 variants for 1099 and 1469 euros respectively.
Nvidia positions the GeForce RTX 4090 as the top model in the new line-up. The RTX 4090 has 16,384 CUDA cores with a base clock of 2.23 GHz and a boost clock of 2.52 GHz. Nvidia supplies this video card with 24GB GDDR6X memory and a 384bit memory bus. The card gets a tbp of 450W. Nvidia claims that the card is “up to two times faster than the RTX 3090 Ti in current games.”
In addition, the company comes with two different GeForce RTX 4080 models. The ‘entry-level model’ will be a GeForce RTX 4080 12GB, which has GDDR6X memory and a Lovelace GPU with 7680 CUDA cores. It has clock speeds of 2.31GHz with a turbo frequency of 2.61GHz. The RTX 4080 12GB gets a 192bit memory bus and a tbp of 285W.
In addition, Nvidia introduces a 16GB variant of the RTX 4080. In addition to extra memory, this will also receive a more powerful gpu with 9728 CUDA cores. The clock speed with this card is 2.21 GHz with a boost clock of 2.51 GHz. Furthermore, this card will have a 256-bit memory bus and a tbp of 320W. Both RTX 4080 video cards will be released in November. Nvidia will produce Founders Editions of the RTX 4090 and RTX 4080 16GB. Only modified variants of board partners of the RTX 4080 12GB will be released on the market.
TSMC and the Lovelace Architecture
The GeForce RTX 40 video cards are produced on TSMC’s 4N process, which the company has developed together with Nvidia for the production of GPUs. The previous RTX 30 generation was produced by Samsung at 8nm.
The top model in the GeForce RTX 40 line-up, the RTX 4090, has 76 billion transistors, according to Nvidia. The manufacturer claims that the Lovelace architecture is up to two times faster than the current GeForce RTX 3090 Ti in games with rasterization. In the case of ray tracing, Nvidia claims a performance gain of four times over the RTX 3090 Ti.
The new video cards will be based on a new Ada Lovelace architecture. With this, Nvidia is renewing, among other things, the streaming multiprocessors. According to the company, these each deliver up to twice the performance. In addition, Nvidia is introducing a new shader execution re-ordering feature. It should accelerate shading in the GPU pipeline by rescheduling workloads in real time to run them as efficiently as possible. According to Nvidia, this is two to three times faster for ray tracing and delivers up to 25 percent better overall gaming performance.
Also part of Lovelace is a third generation of ray tracing cores, which should deliver twice the throughput on ray-triangle interceptions. A fourth generation of Tensor cores must also deliver up to four times higher throughput than before. This is achieved by leveraging the fp8 transformer engine that Nvidia introduced to the Hopper architecture for data centers. Furthermore, the RTX 40 series will receive support for AV1 encoding. Upscaling technique DLSS is getting a new 3.0 version, which for the first time can also generate its own frames for higher frame rates. DLSS 3.0 only works on RTX 40 cards, and not on GPUs from previous generations.
Prices and availability
The presented RTX 40 video cards will be available in the coming weeks. The GeForce RTX 4090 will be released on October 12. This video card will then receive a suggested retail price of 1959 euros; that is 410 euros more than the RTX 3090 at release. Both RTX 4080 variants will follow in November. The 12GB variant will then be available for 1099 euros, where the RTX 3080 had to cost 719 euros at release. The 16GB variant has a suggested retail price of 1469 euros, which is close to the suggested retail price of the RTX 3090 at release.
The company is keeping its RTX 30 series in the running for the time being. The GeForce RTX 3080, 3070 and 3060 will all remain available alongside the three new RTX 40 video cards, Nvidia said during its presentation. Nvidia is currently not talking about the release of lower-ranked GeForce RTX 40 video cards, such as an RTX 4070 or RTX 4060. The expectation was that Nvidia would only present its highest-ranked models during this presentation.
Nvidia GeForce RTX 40 cards to their RTX 30 counterparts | ||||||
video card | RTX 4090 | RTX 3090 | RTX 4080 16GB | RTX 4080 12GB | RTX 3080 | RTX 3070 |
Architecture | Ada (TSMC 4N) | Ampere (Samsung 8nm) | Ada (TSMC 4N) | Ada (TSMC 4N) | Ampere (Samsung 8nm) | Ampere (Samsung 8nm) |
CUDA cores | 16,384 | 10,496 | 9728 | 7680 | 8704 | 5888 |
Boost Speed | 2.52GHz | 1.70GHz | 2.51GHz | 2.61GHz | 1.71GHz | 1.73GHz |
Memory | 24GB GDDR6X | 24GB GDDR6X | 16GB GDDR6X | 12GB GDDR6X | 10GB GDDR6X | 8GB GDDR6 |
Memory bus | 384bit | 384bit | 256bit | 192bit | 320bit | 256bit |
Tbp | 450W | 350W | 320W | 285W | 320W | 220W |
Release date | October 12, 2022 | September 2020 | Nov 2022 | Nov 2022 | September 2020 | October 2020 |
MSRP | 1959 euros | 1549 euros | 1469 euros | 1099 euros | 719 euros | 519 euros |