The video discusses the anticipated release of Nvidia’s RTX 5090 graphics card at CES 2025, highlighting its expected advancements in memory capacity and bandwidth, which could significantly enhance local AI performance. The host emphasizes the potential for the RTX 5090 to outperform its predecessor, the RTX 490, particularly in inference capabilities, and encourages viewers to consider upgrading despite concerns about pricing and power requirements.
The video discusses the anticipation surrounding the rumored release of Nvidia’s RTX 5090 graphics card, which is expected to be unveiled at CES 2025. The host expresses excitement about the potential capabilities of this new GPU, especially for developers and local AI enthusiasts. They mention that Nvidia has been focusing more on AI GPUs rather than traditional gaming GPUs, but the RTX 5090 could represent a significant advancement for local computing. The host has sold their previous RTX 490 cards in preparation for this new release, highlighting the long wait for an upgrade.
The video references leaks and rumors about the RTX 5090, including speculation about its specifications. It is suggested that the card will feature 32 GB of GDDR7 memory, which would be double the capacity of its predecessor, the RTX 580. The host emphasizes the importance of memory bandwidth for performance, noting that the RTX 5090 is expected to deliver an impressive 1.8 terabytes per second of bandwidth, thanks to its 512-bit memory bus and high-speed GDDR7 memory. This could lead to significant improvements in performance for local AI applications.
The host also discusses the power requirements of the RTX 5090, which is expected to have a thermal design power (TDP) of 575 watts, significantly higher than the RTX 490. This increase in power draw may necessitate new power supply units for users looking to upgrade. The video highlights the potential for the RTX 5090 to outperform the RTX 490 in terms of inference performance, with estimates suggesting an 80% improvement. This performance boost is attributed to the increased memory bandwidth, which is crucial for running larger AI models more efficiently.
In terms of practical implications, the host provides a comparison of token processing capabilities between the RTX 5090 and the RTX 4090, suggesting that the new card could handle up to 128 tokens per second on a 7 billion parameter model, compared to 71 tokens per second for the 4090. This performance leap makes the RTX 5090 an attractive option for those involved in local AI development. The host encourages viewers to consider upgrading, despite potential concerns about pricing and Nvidia’s business practices.
The video concludes with the host expressing enthusiasm for the upcoming CES event and the potential for the RTX 5090 to revolutionize local AI capabilities. They invite viewers to join a live stream of the event and share their thoughts on the new GPU. Overall, the video captures the excitement and speculation surrounding Nvidia’s latest offering, emphasizing its potential impact on the AI landscape and local computing.