Nvidia is reportedly preparing a significant upgrade to its lineup of laptop GPUs. The forthcoming RTX 50 series, codenamed “GN22,” will feature six different variants, surpassing the five options offered in the current RTX 40 series, according to an apparent leak shared by security researcher Dominic Alvieri.
If accurate, the most notable enhancement in the updated specifications is the increase in video random-access memory (VRAM). The baseline for the RTX 50 series will be 8GB of VRAM, a step up from the 6GB baseline in its current laptop GPUs.
The lineup will reportedly include three models with 8GB of VRAM, two with a robust 16GB, and one with 12GB.
This boost in VRAM will enable the new GPUs to run more artificial intelligence (AI) models, which often require substantial memory resources. For example, Stable Diffusion 1.5 requires at least 4GB of VRAM to run properly. SDXL requires at least 6GB to be able to run.
When Stability AI previewed its forthcoming SD3 model, which will be much more capable and powerful, it used GPUs with 24GB of VRAM. Fortunately, different versions will be available, with 800 million to 8 billion parameters, and different configurations—such as excluding its 4.7 billion parameter T5 text encoder—will not be so demanding.
The RTX 50 series will also leverage faster GDDR7 memory, according to the leak, further improving the performance of AI-related tasks and gaming.
The leaked roadmap suggests that Nvidia plans to continue producing some older GPUs, such as the RTX 4050 and RTX 3050, with 4GB and 6GB of VRAM to meet the market’s need for cheaper hardware. For gaming, 6GB cards are still solid options, as VRAM is not as important as the clock speeds and the number of processors or cores.
These GPUs will reportedly be released in 2025, although no specific timeline was provided. There have been conflicting rumors, but a CES 2025 unveiling is possible.
Buyers looking for cheap cards may have their days numbered, however. Taiwan Semiconductor Manufacturing Company (TSMC), Nvidia’s main chip supplier, is said to be considering a price hike as its own expenses grow.
“We expect our customers to share some of the higher cost with us, and we already started our discussion with our customers,” Wei said, according to Nikkei Asia.
The report said Wei recently referred to Nvidia CEO Jensen Huang as the “three trillion guy,” referencing Nvidia’s soaring market value, which surpassed $3 trillion last week.
TSMC produces an estimated 90% of the world’s most advanced processor chips and is a vital player in Nvidia’s success.
Despite the potential for increased production costs, Huang expressed his support for TSMC’s arguments, saying that raising prices is “consistent with the value that they deliver,” Nikkei Asia reported.
Morgan Stanley analysts have suggested that if Nvidia accepts a price hike from TSMC, other key AI players may follow and raise their prices. They estimate that Nvidia will account for 10% of TSMC’s revenue this year.
Generally Intelligent Newsletter
A weekly AI journey narrated by Gen, a generative AI model.
Source: https://decrypt.co/234675/nvidia-rtx50-leak-gpu-vram-upgrades
