Intel 4th Gen Xeon Powers New Amazon EC2 M7i-flex and M7i Instances

[GFreeman]

Today, Amazon Web Services (AWS) announced the general availability of new Amazon Elastic Compute Cloud (Amazon EC2) instances powered by custom 4th Gen Intel Xeon Scalable processors. This launch is the latest on a growing list of 4th Gen Xeon-powered instances that deliver leading total cost of ownership (TCO) and the most built-in accelerators of any CPU to fuel key workloads like AI, database, networking and enterprise applications.

"Intel worked closely with AWS to bring our feature-rich 4th Gen Xeon processors to its cloud customers, many of which have benefited from its performance and value for months in private and public preview. Today, we're happy to bring that same real-world value to cloud customers around the globe," said Lisa Spelman, Intel corporate vice president and general manager of the Xeon Products and Solutions Group.



These new Amazon EC2 instances - Amazon EC2 M7i-flex and M7i - bring 4th Gen Xeon's accelerator engines to the masses via AWS's expansive global footprint. Built-in accelerators like Intel Advanced Matrix Extensions (Intel AMX) offer a much-needed alternative in the market for customers with growing AI workload demand. 4th Gen Xeon with AMX can also meet inference performance metrics for large language models (LLMs) below 20 billion parameters, making LLMs both cost-effective and sustainable to run on general-purpose infrastructure.

More broadly, Intel and AWS offer customers a series of product options to best match their workload, performance and pricing needs, along with the ease and flexibility to address future digital infrastructure demands.

M7i-flex and M7i instances are available in the following AWS Regions: US East (Ohio), US East (N. Virginia), US West (Oregon) and Europe (Ireland).

The M7i-flex instances offer:

• The easiest way for customers to get price-performance benefits for a majority of general-purpose workloads.
• Designs to seamlessly run general-purpose workloads, including web and application servers, virtual desktops, batch processing, microservices, databases and enterprise applications.
• Up to 19%1 better price performance compared with M6i instances.

The M7i instances offer:

• Price performance benefits for key workloads, such as large application servers and databases, gaming servers, CPU-based machine learning and video streaming.
• Larger instance sizes (up to 192 vCPUs and 768 GiB memory).
• New built-in accelerators that enable efficient offload and acceleration of data operations that help in optimizing performance for databases, encryption and compression, and queue management workloads.
• Up to 15%1 better price performance compared with M6i instances

View at TechPowerUp Main Site | Source

 

[john_]

I think it's a long time since the last time I've seen a press release about Xeon and not EPYC. Either these accelarators do work, or Amazon got a deal from Intel that couldn't refuse.

 

[lemonadesoda]

They are on a budget. Because a name like Amazon EC2 makes no sense. And Elastic Compute is just silly. It means a department staffed by interns. Ie cheap

 

[AnarchoPrimitiv]

I think it's a long time since the last time I've seen a press release about Xeon and not EPYC. Either these accelarators do work, or Amazon got a deal from Intel that couldn't refuse.

Based on the benchmarks I've seen, I'd imagine Amazon got a deal....

 

[cvaldes]

They are on a budget. Because a name like Amazon EC2 makes no sense. And Elastic Compute is just silly. It means a department staffed by interns. Ie cheap

There are companies of various sizes and at different places in their timeline that take advantage of Amazon AWS.

For some early stage, minimally staffed startups, an EC2 instance is a better use of limited cash rather than hiring a bunch of IT staff and stocking up your own server room with hardware that is idle half of the time.

Twice I used a low-end Amazon EC2 instance for home office purposes, each time for a 12-month free trial.

Hardware cost: $0
Software cost: $0
Time spent on Windows system administration: 0 hours 0 minutes

My Windows office needs have evolved yet if I didn't do any PC gaming, I could get by with a $200 mini PC running an Intel N100 SoC.

Laugh all you want at Amazon EC2 but for some companies at certain points in their history, EC2 provides a useful solution, just like a five-year-old Honda FIT might be more useful than a big honking brand new SUV would be depending on the buyer and what moment of their life is.

 

[lemonadesoda]

@cvaldes Hey pal you picked up the wrong end of the stick. I was talking about the name EC2. Sorry if i wasn’t clear

 

[unwind-protect]

Good opportunity to give them a spin with my own benchmarks.

 

[Minus Infinity]

I think it's a long time since the last time I've seen a press release about Xeon and not EPYC. Either these accelarators do work, or Amazon got a deal from Intel that couldn't refuse.

IIRC their were some benchmarks posted a few months ago for the AI accelerators and for the task(s) chosen they were pretty big leaps in performance, like > 5x. AMD won't have them until Epyc Turin.

 

[john_]

IIRC their were some benchmarks posted a few months ago for the AI accelerators and for the task(s) chosen they were pretty big leaps in performance, like > 5x. AMD won't have them until Epyc Turin.

And here is a question I have. With all the focus on GPUs for AI, do server CPUs really need AI functionality? Maybe it's simpler to program a few typical AI functions when also the CPU implements AI functions, but are they really needed? Just throwing thoughts here and while the simple answer would be "Obviously there must be advantages" still wonder if these are true advantages. Even 1000 CPUs with AI functions might not be as fast a 5 top GPUs.

 

[Minus Infinity]

And here is a question I have. With all the focus on GPUs for AI, do server CPUs really need AI functionality? Maybe it's simpler to program a few typical AI functions when also the CPU implements AI functions, but are they really needed? Just throwing thoughts here and while the simple answer would be "Obviously there must be advantages" still wonder if these are true advantages. Even 1000 CPUs with AI functions might not be as fast a 5 top GPUs.

Something like Hopper costs $70KUS. These server class CPU's about $14K in top spec. Not everyone is running LLM's for training. Lot's of other AI tasks don't need that power. The companies will know exactly what capabilities they need. Different segments will require things like H100 Mi300, Gaudi 2/3 etc for sure, others will find AI accelrated cpu's more than sufficient.