- Vultr is the latest data center operator to add support for AMD's MI300X accelerators
- Cloud providers across the board are looking to expand their GPU diversity
- Hyperscalers are poised to challenge traditional chip players with their in-house silicon
Nvidia may be top dog, but businesses on a budget want options when it comes to running their artificial intelligence workloads in the cloud. That’s great news for AMD, which is starting to see an uptick in adoption of its GPUs across cloud providers of all shapes and sizes.
GPU specialist Vultr is the latest to jump on the AMD bandwagon, adding the chip company’s Instinct MI300X accelerator and ROCm software to its suite of compute options. Microsoft did the same back in May and Oracle also added AMD’s MI300X to its lineup earlier this year.
IDC Research VP Dave McCarthy told Fierce these moves are all about bringing the same kind of silicon diversity that already exists in the CPU market to GPUs to provide choice in how enterprises build their cloud AI infrastructure.
“Silicon diversity has already been happening in the cloud. AMD CPUs are a viable alternative to Intel and their high core counts are an advantage for HPC and other performance-intensive workloads,” he noted.
“While NVIDIA is the undisputed leader in overall GPU performance, not every customer is training the next trillion parameter AI model,” McCarthy continued. “The AMD MI300X will have a lower price point while still offering impressive performance. It will be a good fit for customers looking to optimize price and performance.”
The moves come as cloud companies across the board deploy more accelerated servers to meet demand for AI workloads. Dell’Oro Group noted this week that data center capex jumped 46% in Q2 2024, driven in large part by spending on accelerate servers. These types of servers accounted about 41% of overall server revenues in the quarter.
But AMD isn’t the only one looking to give Nvidia a run for its money in the accelerator market. As Fierce has previously noted, the likes of Google and Amazon Web Services have come out with their own silicon offerings.
Indeed, “servers equipped with NVIDIA Hopper GPUs and custom accelerators, such as Google TPU and Amazon Trainium, saw increased deployment among hyperscale cloud service providers,” in the quarter, Dell’Oro Senior Research Director Baron Fung said in a statement. “Strong demand also came from enterprises and Tier 2 cloud service providers.”