Exploring AWS’s New Graviton5 Processors: M9g and M9gd Instances

AWS recently launched its M9g and M9gd instance types, marking a significant upgrade in its Graviton line of processors. This new generation of instances is tailored for demanding workloads, particularly those associated with agentic AI. Here’s a deep dive into their specifications, capabilities, and market impact.

Specifications and Features of M9g and M9gd Instances

The M9g and M9gd instances double the core count to 192 per chip. These CPUs are constructed using four chiplets built on TSMC’s advanced 3nm process node. Key specifications include:

• L3 Cache: A whopping 192 MB of L3 cache, which is five times more than the previous Graviton4.
• Memory: Support for DDR5-8800 RAM, noted as the fastest DDR5 available in any cloud instance. This capability enhances data access speeds, crucial for high-performance applications.
• Storage and I/O: PCIe Gen 6 interfaces and up to 11.4 TB of NVMe SSD storage with 30% higher IOPS than the previous generation.
• Hypervisor Isolation: The introduction of the first formally verified hypervisor isolation in a production cloud environment, enabling stronger security for multi-tenant workloads.

Purpose-Built for Agentic AI Workloads

According to AWS, the Graviton5 is specifically engineered for agentic AI tasks. The About Amazon blog underscores its suitability for real-time reasoning, code generation, and multi-step task orchestration. The key here is the ability to handle numerous concurrent environments simultaneously, keeping all attached accelerators in motion.

However, opinions among tech enthusiasts are mixed. Some users on Reddit challenge AWS’s marketing framing, asserting that while Graviton chips excel at their jobs, labeling them as “AI processors” might be an overreach. One user succinctly noted:

“They’re normal ARM cores… the only reason those are AI cores is because somebody in marketing thinks that will sell better with the word AI attached.”

Another commentator provided a practical perspective, emphasizing the chip’s general-purpose advantages rather than its marketed AI capabilities, remarking on its performance benefits for traditional applications such as databases and web servers.

Real-World Performance Benchmarks

Customer feedback from the six-month preview period paints a promising picture. Notably, ClickHouse reported a 36% performance increase compared to M8g instances with zero code modifications. Honeycomb achieved similar results with a 36% rise in throughput per core relative to Graviton4. HubSpot also shared success stories, citing a dramatic 60% reduction in query durations after deploying M9g instances for MySQL databases.

Feedback from industry professionals reinforces the exceptional performance of the chips. Denis Sheahan, a principal performance engineer at Airbnb, remarked that M9g instances are “some of the fastest EC2 instances we have ever tested.”

Enhanced Security with the Nitro Isolation Engine

Another significant advancement is the integrated Nitro Isolation Engine within the sixth-generation Nitro System. This feature provides mathematically verified isolation between virtual machines, enhancing security in multi-tenant environments. The formal verification ensures that the isolation boundary is not just tested, but proven mathematically, thus bolstering protection against potential vulnerabilities.

Cost Analysis and Price Performance

When it comes to cost, AWS’s published prices show M9g instances are priced 9% higher than their M8g counterparts across all sizes. However, the increase in compute performance—estimated at around 25%—results in a roughly 15% improvement in price-performance ratio. Actual benefits will vary depending on workload.

M9gd instances are tailored for data-intensive applications, such as media processing and batch jobs, leveraging their extensive local storage capabilities to run more efficiently.

Future expansions in the Graviton Family

While AWS has made significant strides with the M9g and M9gd instances, the expansion of the Graviton family is anticipated to continue. Compute-optimized C9g and memory-optimized R9g variants are projected for release in late 2026. Enterprises with demanding inference or analytics requirements might benefit from waiting for these specialized configurations to enter the market.

Competitors and Market Context

In the broader context, AWS’s advancements with the Graviton line position it as a leader in custom silicon adoption. While companies like Google and Microsoft are venturing into custom processing with their offerings—Axion for GKE and Cobalt 100 for Azure, respectively—AWS has carved out a substantial market footprint. The Graviton processors alone accounted for more than half of all new CPU capacity added over the last three years, with its custom silicon business now surpassing a $20 billion annual run rate.

Availability of Instances

M9g and M9gd instances are available immediately through multiple purchasing options, including On-Demand, Savings Plans, Spot, Dedicated Instances, and Dedicated Hosts. Presently, these instances can be accessed in various regions including US East (N. Virginia, Ohio), US West (Oregon), and EU (Frankfurt), showcasing AWS’s commitment to global service coverage.

This rich array of features and advancements demonstrates AWS’s commitment to optimizing cloud operations. The M9g and M9gd instances present both robust performance and enhanced security for enterprises looking to harness the power of custom silicon in their journey toward digital transformation.

Inspired by: Source