Intel and AWS Drive Cloud Innovation Powered by Xeon 6 Processors

[Daniel Nenni]

What’s New: Intel® Xeon® 6 processors with P-cores are now powering the new Amazon Elastic Compute Cloud (EC2) R8i and R8I-flex instances, available on AWS, delivering the highest performance and fastest memory bandwidth among comparable Intel processors in the cloud.

The new instances reflect years of collaboration between Intel and AWS, bringing advanced silicon innovations and tightly integrated software optimizations to customers seeking scalable, cost-efficient and high-performance cloud infrastructure.

“The launch of the 8th Generation EC2 instances powered by Intel Xeon 6 is a milestone in our collaboration with AWS,” said Ronak Singhal, senior fellow at Intel. “Together, we’ve built an infrastructure that accelerates AI, enhances memory performance, and simplifies deployment – helping customers achieve faster insights and stronger ROI.”

Why It Matters: As enterprises increasingly rely on real-time data processing, AI, and compliance-intensive workloads, the need for versatile, high-performance cloud infrastructure is critical. Intel Xeon 6 processors provide a flexible, general-purpose compute platform that minimizes the reliance on specialized accelerators – streamlining deployment and management for diverse use cases.

“Our customers need infrastructure that matches the pace of their most demanding workloads,” said Nishant Mehta, vice president of EC2 Product Management at AWS. “Through our collaboration with Intel, the new Amazon EC2 R8i and R8i-flex instances deliver breakthrough capabilities, offering 2.5x more memory bandwidth and 15% better price performance compared to previous generations. This is the kind of innovation that helps customers maximize performance while driving down costs.”

How It Works: Intel worked to optimize across every layer of the stack – from processor microarchitecture and firmware tuning to hypervisor tuning, virtualization and software frameworks. This tight integration ensures customers fully benefit from Xeon 6 capabilities across AWS’s global infrastructure.

This collaboration underscores Intel’s role as a foundational technology provider for AWS and the broader cloud ecosystem, enabling developers, data scientists and IT leaders to innovate, scale and deploy with greater speed and efficiency.

Key benefits of the new Xeon 6 based instances include:

• High-Density Compute: Xeon 6 features a rearchitected core design with increased core counts and improved pipelines, delivering greater performance and improved thread scalability. This advancement enables AWS to offer increased instance sizes like 96 x large, giving customers access to greater compute density for their workloads.
• Fastest DDR5 Support in the Cloud: With DDR5 memory speeds up to 7200 MT/s, the new instances deliver significantly higher memory bandwidth and lower latency, ideal for analytics, Machine Learning (ML), and in-memory databases.
• Built-in AI Acceleration: Integrated Intel® Advanced Matrix Extensions (AMX), enable up to 2x AI inference and ML performance gains, reducing the need for external accelerators and simplifying deployment.
• Intel® QuickAssist Technology (QAT): Enables hardware-accelerated encryption and compression, freeing up CPU resources and boosting performance for security-sensitive and regulated workloads across industries including finance, healthcare and public sector applications.

Who’s Using It: Industry leaders – including Netflix and CrowdStrike, among others – are early adopters piloting Xeon 6-based instances to power AI-driven and cloud-native applications. They report measurable gains in compute throughput, AI performance and infrastructure efficiency.

More Context: Press Kit: Intel Xeon 6 Processors | New Memory-Optimized Amazon EC2 R8i and R8i-flex Instances

 

[IanD]

"...delivering the highest performance and fastest memory bandwidth among comparable Intel processors in the cloud."

Note lack of comparison with AMD processors. I wonder why that is? ;-)

 

[samwilde]

This is a definitive win. The new Intel 3 server processors are delivering.

 

[IanD]

This is a definitive win. The new Intel 3 server processors are delivering.

Indeed -- but how are they delivering compared to new AMD processors, not older Intel ones? ;-)

 

[samwilde]

At least Granite Rapids has equal core count with max 128 cores and 256 threads to Turin. Which was not the case with old SKUs.

The real dud IMHO is Sierra Forest which is no match for Turin Dense. It has no SMT.

 

[MKWVentures]

At least Granite Rapids has equal core count with max 128 cores and 256 threads to Turin. Which was not the case with old SKUs.

The real dud IMHO is Sierra Forest which is no match for Turin Dense. It has no SMT.

Intel has been surprisingly open with how bad SF sales are and how it is not competitive. When will Intel CPUs be competitive?

 

[hist78]

"...delivering the highest performance and fastest memory bandwidth among comparable Intel processors in the cloud."

Note lack of comparison with AMD processors. I wonder why that is? ;-)

Amazon AWS uses Intel Xeon, AMD EPYC, and AWS Graviton, Trainium, and Inferentia. It would be vary helpful to compare their performance. But Amazon might not want to release such study.

 

[tomatoma]

Intel has been surprisingly open with how bad SF sales are and how it is not competitive. When will Intel CPUs be competitive?

At least the difference is shrinking

 

[tomatoma]

Indeed -- but how are they delivering compared to new AMD processors, not older Intel ones? ;-)

Well, it's true that AWS hasn't officially introduced a large amount of Zen5 yet.
If you want to compare, it is better to compare the same Intel previous generation processors.

 

[blueone]

Amazon AWS uses Intel Xeon, AMD EPYC, and AWS Graviton, Trainium, and Inferentia. It would be vary helpful to compare their performance. But Amazon might not want to release such study.

Xeon, EPYC, and Graviton are CPUs, so they're comparable. Tranium is comparable to Google TPUs, but not general purpose CPUs. Trainium and Inferentia have similar architectures, but Trainium uses HBM memory, while Inferentia uses DDR memory. So Inferentia is really a cheaper version of Trainium.

The Xeon6 CPUs have been extended with proprietary accelerators. If your applications have been modified to take advantage of them your workload might get a significant advantage over CPUs without these accelerators. (I don't know how many applications have been modified.) Graviton's value proposition is the cheapest CPU power to AWS, which is why AWS is transitioning all of their internal processing to Gravitons. They're Arm Neoverse V2 cores, but don't have special integrated accelerators. So comparing Graviton to Xeon and EPYC is really not logical. The big advantage to the x86 processors in general is that many enterprise applications are native to that instruction set, so you get the highest quality and reliability, even assuming they'll run on Arm-compatible cores.

So a comparison between these chips is not really useful at all. A comparison between Xeon and EPYC CPUs can be useful, but as I've pointed out, it's application dependent.

 

[siliconbruh999]

"...delivering the highest performance and fastest memory bandwidth among comparable Intel processors in the cloud."

Note lack of comparison with AMD processors. I wonder why that is? ;-)

Uh in terms of memory bandwidth at least Xeon Offers the highest so xeon is really better than latest EPYC in this regard. I wonder why did they gimp Xeon 6 to 7200MT/s when they can do 8800MT/s latest Turin can only do 6400MT/s with 1DPC.

 

[siliconbruh999]

Intel has been surprisingly open with how bad SF sales are and how it is not competitive. When will Intel CPUs be competitive?

When the marketing people finally stop segmenting ISA SRF is missing AVX-512 and the per core performance is not great but the 288C variant of SRF went to Hyper scaler and never came to general market.
I thought Clearwater Forest would close the gap but we will have to see cause they delayed it on account of lackluster demand and Packing issue.

 

[blueone]

Uh in terms of memory bandwidth at least Xeon Offers the highest so xeon is really better than latest EPYC in this regard. I wonder why did they gimp Xeon 6 to 7200MT/s when they can do 8800MT/s latest Turin can only do 6400MT/s with 1DPC.

I'm not a board design guy at all, but aren't DDR speeds limited by trace length and other board design considerations? When I've seen Xeon motherboards I'm always surprised by how much acreage the CPUs, and the associated DIMMs and PCIe ports take. Perhaps someone with board design expertise can give us short tutorial, because this is an interesting server design question.

 

[lispwright]

I'm not a board design guy at all....

Neither am I, but I've worked with them at the beginning and end of my career and between that and basic E&M picked up some things. Enough to be horrified at that last job by a bug on a big board where the signal was good at the end of some traces but corrupt in the middle!!! The analog domain is filled with demons and those good at it really earn their pay.

So the key to huge memory banks is easing the electrical load with registered memory that inserts buffering on the DIMM sticks at the cost of some latency.

Otherwise at the low end you usually get two unregistered channels with a max of two DIMMs each, you populate only the last DIMM slots in each for max speed and latency. Which thanks to Moore's Law not quite being dead is not so painful as of late, 64 GB is good enough for most software development and web browsing workstations.