Google Cloud is the first cloud provider to offer virtual machines (VMs) with Nvidia’s new RTX PRO 6000 Blackwell GPUs. These new G4 VMs are suitable for a wide range of applications, such as AI inference, robotics simulations, generative content creation, and advanced graphics processing.
This is according to The Register. According to Google Cloud executives, including Nirav Mehta and Roy Kim, the G4 VMs can be used for both heavy and cost-conscious workloads. The virtual machines combine eight Nvidia RTX PRO 6000 GPUs with two AMD Turin CPUs. They use Google’s own Titanium processors for extra performance.
The RTX PRO 6000 Server Edition, unveiled at Nvidia’s GTC event earlier this year, succeeds the older L40 series. It focuses on AI tasks, model optimization, and data visualization, such as digital twins. Each GPU delivers 3,753 teraFLOPS of computing power at FP4 precision, has 96 GB of GDDR7 memory, and a memory bandwidth of 1.6 TB/s. The G4 VMs thus offer a total of 768 GB of GPU memory, 384 virtual CPU cores, and 12 TiB of local SSD storage. The latter can be expanded to 512 TiB via Hyperdisk. Compared to the previous G2 generation, it offers four times the computing power and memory, and six times the bandwidth.
Fast results
Thanks to their larger memory and higher data speeds, these VMs are better equipped to run complex AI models and deliver results more quickly. The previously released A4 and A4X VMs, which also run on Blackwell GPUs, are more focused on training and inference and lack the graphics processing capabilities of the G4 series.
Market analysis firm IDC also sees the introduction as a sign that Google is serious about powerful cloud solutions. According to IDC president Crawford Del Prete, Google is demonstrating its commitment to supporting diverse customer needs with powerful infrastructure suitable for heavy AI and data-intensive applications. He notes that this is particularly attractive to customers who are willing to invest extra for maximum performance, partly because Google is bringing its AI services together in an integrated ‘AI Hypercomputer’ concept.
Although the technology is already available, interested parties will need to contact a Google Cloud representative for the time being. The G4 VMs are still in a preview phase. Global availability is expected by the end of the year.
Tip: Google Cloud now offers Nvidia Blackwell-supported instances