Nvidia has unleashed a new line of RTX GPUs designed specifically for artificial intelligence (AI) applications on desktop computers. While Nvidia’s RTX GPUs have predominantly been recognized for their gaming and graphics capabilities, the company is now repackaging and configuring them to cater to the growing demand for AI in various industries. Nvidia’s move aims to make GPUs more accessible and readily available for customers requiring them.
The newly announced RTX GPUs are built on the Ada Lovelace architecture, which sets them apart from the Hopper architecture used in the widely sought-after H100 GPUs that are currently in short supply. These GPUs are particularly enticing for enthusiasts looking to explore AI applications on their desktops. In fact, some gaming laptop users have already been utilizing GPUs to run AI-powered applications such as text-to-text or text-to-image models.
Nvidia unveiled these new desktop and workstation designs featuring the RTX GPUs at the recent SIGGRAPH conference. Major computer manufacturers like Dell, Lenovo, and Boxx are set to announce workstations capable of accommodating up to four RTX 6000 data generation GPUs in a single chassis. While Nvidia suggested a retail price of $6,000 for the GPU, vendors such as Dell are selling them for over $9,000, tax included.
Each RTX 6000 GPU, based on the Ada Lovelace design, boasts 48GB of GDDR6 memory and a 200Gbps network-interface card. These GPUs consume 300 watts of power and are based on the older PCIe 4.0 interconnect standard.
In addition to the RTX 6000, Nvidia also introduced the L40S Ada GPU, which can be considered a more affordable alternative to the H100. The L40S GPU demonstrates superior AI training and inference capabilities compared to the previous-generation A100 GPUs, making it an attractive option for users operating within AI-intensive domains. The L40S GPU features 48GB GDDR6 memory and will be integrated into systems based on the OVX reference server design, specifically for metaverse applications.
The L40S GPU offers up to four times faster performance for AI and graphics workloads compared to the previous A40 GPU, which is based on the previous generation Ampere architecture. This advancement is attributed to faster clock speeds, enhanced tensor capability, and improved graphics rendering performance.
It is worth noting that Nvidia’s RTX systems for businesses are designed to cater to both the metaverse and AI markets. The new hardware will come bundled with licenses for the Omniverse and AI Enterprise software. Additionally, Nvidia announced the upcoming launch of AI Enterprise 4.0, which includes the Nemo large-language model.
Unlike the scarcity typically associated with high-end GPUs, Nvidia assures consumers that the L40S GPU supplies will not be as constrained. The GPUs are expected to ship later this year, providing enthusiasts and professionals with more accessible access to powerful AI-capable hardware.
Furthermore, Nvidia revealed that it will release the low-end RTX 4000 GPU in September, retailing at $1,250, followed by the RTX 4500 in October, with a price tag of $2,250.
Nvidia’s commitment to making GPUs readily available for AI applications highlights its recognition of the sector’s importance alongside gaming. The company aims to position GPUs as versatile commodities, enabling enthusiasts to create their own programs and utilize them wherever GPUs are accessible. However, Nvidia’s H100 GPUs remain challenging to find, becoming highly sought-after assets for businesses. For instance, startup CoreWeave has utilized its Nvidia GPUs as collateral to secure funding for its expansion. Furthermore, cryptocurrency miners have begun repurposing their GPUs in data centers to facilitate AI operations.
