Driving AI Innovation Forward: How The AMD Instinct™ MI350 Series Is Powering The Future

The growth of AI’s presence requires every organization to be at the forefront of innovation in the field, and AMD is no exception. Now, it is looking forward to its next step in its AI journey and is ready to take its advances directly to partners and consumers.

CRNtv’s Jon Alba spoke with Mahesh Balasubramanian, director, Datacenter GPU Product Marketing for AMD to explore how AMD is navigating the ever-changing AI landscape and look at how the new AMD Instinct™ MI350 series is changing the game.

Where does AMD currently stand in its AI journey?

Mahesh Balasubramanian: AI moves at a rapid pace, and I feel like AMD is the real inflection point right now. We've moved from establishing our credibility with our epic series of CPUs across a broad set of industries to now establishing our instant GPUs in this market, which is widely available through Supermicro servers and Supermicro as a partner cloud ecosystem. The cadence of silicon with which we are delivering data center experience to our customers has accelerated quite rapidly to keep pace with this AI innovation that's happening, as well.

The last few years we've been shipping at MI300X and MI325 AMD Instinct™, in volume with Supermicro and our partners jointly. And this year, we are really excited to launch our mid-level series product jointly as well, being the 350X leap in high-performance gen-on-gen from the previous generation products, also in a broadly available set of servers from Supermicro, both in liquid and air-cooled solutions.

So, let's build off that a little bit. What do AMD's AI advances mean for Supermicro customers?

Mahesh: Supermicro is integrating all our EPYC CPUs, as well as inserting GPUs, into a broadly available set of servers, as mentioned, both as liquid and air-cooling solutions.

They could go into super dense racks for greenfield or hyperscale-use cases where they are maximizing space constrained, as well as power and cooling, or into your existing infrastructure where you might be very dependent on air source solutions. You can still maximize your AI utilization with a choice of servers in the air-cooled racks as well, in several chassis offices, with certified power and cooling from Supermicro.

Beyond the hardware innovation you're speaking of, what else should customers expect on the software side of things?

Mahesh: The first thing is, that open-source software R&D is deeply, deeply integrated with our ROCm™ software stack, and it brings in all the latest optimization and support for innovating on AI. This includes support for the latest frameworks, including PyTorch, Jax, OpenAI, Triton, inference frameworks and all of these open solutions.

The second thing is the native support for frameworks that I mentioned gives researchers portability that enables them to use not only AMD's hardware, but a choice of hardware that brings in this portability. So, you're not stuck using a single GPU or a single solution and that again accelerates your time to develop.

And the third part is we have a growing catalog of fully optimized models, including day-zero support for all the latest models. When Lama 3 or the latest Llama for Scout and Maverick came out, they were supported. Similarly, Deepseek R1 and V1 ran out of the box on instant hardware with rapid optimization that increased performance by over 3x within a two-week window.

This software innovation is possible when you have an open ecosystem. With the partnership, that ecosystem that AMD has participated in helps bring these latest models to customers for them to choose and deploy for their workloads.

How can partners and customers actually test drive all these solutions?

Mahesh: One is jointly with Supermicro. There is remote, POC access or sandbox access, that are available for our partners to test drive. Instinct GPUs as well as EPYC CPUs are available in their choice of location, whether it be in the hybrid cloud, or on-prem solution or through the cloud.

The second option is, Supermicro has a jumpstart program in various locations. That provides easy access for evaluating these GPUs or their workloads.

And the third is, Supermicro namely has a pretty large partner ecosystem with several neoclouds using Supermicro servers. So that's another easy option for access to this hardware. And a lot of these options come with free credits or extended support to ensure that that sandbox environment gives our customers the best experience to test drive these joint solutions for their use cases.

To learn more about the AMD Instinct™ MI350 series, check out their products here.