Intel IPU E2100 DPU Finally Launched for the Mass Market

4
Intel Mount Evans Vision 2022 With Patrick 2
Intel Mount Evans Vision 2022 With Patrick 2

Today is finally the day. After years of waiting, the Intel IPU E2100 DPU is finally out. Not to be confused with theĀ Intel Xeon E-2100 series, the new DPU uses Arm cores and offers hardware offloads for storage, networking, and security. Codenamed “Mt Evans” we have been following this line for some time.

Intel IPU E2100 DPU Finally Launched for the Mass Market

Here is what the new DPU looks like. Officially, the part number is the Intel E2100-CCQDA2. This has two ports that can be configured as 1x 200GbE, 2x 100GbE, and also supports breaking out to 4x 25GbE.

Intel E2100 CCQDA2 IPU DPU Launch Look
Intel E2100 CCQDA2 IPU DPU Launch Look

There is also a 1GbE out-of-band management port on the cards as well.

Oneboard, the specs are really interesting. The card has 16 Arm Neoverse N1 cores with a 32MB cache and 3 channels of 16GB LPDDR4x memory for 48GB total. Here is the official spec list:

Intel IPU E2100 DPU Specs
Intel IPU E2100 DPU Specs

The idea behind the Mt Evans IPU was to create a DPU that is more akin to an AWS Nitro solution for Google Cloud. That means offloading things like NVMe over Fabric as well as securing the network with Intel QuickAssist derived IP. The Intel IPU E2100 is the first public version of the series for customers outside of Google.

Intel Mount Evans IPU Card
Intel Mount Evans IPU Card

During Hot Chips 33, we covered the Intel Mount Evans DPU IPU Arm Accelerator and first saw the card in person at Intel Vision 2022. Intel initially focused support for the card with Google, but it is now bringing that support to a more general audience.

HC33 Intel Mount Evans DPU IPU Packet Processing P4
HC33 Intel Mount Evans DPU IPU Packet Processing P4

With its programmable packet pipeline, hardware offloads, and more, this is a really interesting platform.

Final Words

Today, Intel showed off deep packet inspection, OpenVINO with NGINX Plus AI, and even Red Hat OpenShift use cases for the IPU. Hopefully, we get a pair of these soon as we are in the middle of testing the BlueField-3 DPUs. Look for that review to hit in the next few weeks. In the meantime, you can see our NVIDIA BlueField-2 DPU and AMD Pensando DSC2-100G Elba DPU hands-on with VMware and our Intel IPU Hands-on with Big Spring Canyon, which was an IPU for another hyper-scaler:

If you want to learn more about DPUs, see ourĀ What is a DPU A Data Processing Unit Quick Primer.

4 COMMENTS

  1. If STH doesn’t have a tray of these, then they’re still vaporware. I liked the FPGA video you did a few years ago so I’d like to see this.

    Imagine that VMware still hasn’t announced support and Pat G’s at Intel now.

  2. These still look like computers on PCIe cards to me. You still need an expensive host machine to plug the expensive computers on PCIe cards into. When reality sinks in then Intel & nVidia & etc will just package the same stuff up in 1U cases at much higher density – more practical and a new love interest for the tech press.

  3. Is it known whether there was some sort of exclusivity period with Google; or was Intel delayed by some aspect of getting it ready for general release(presumably software or firmware that wasn’t substantially Google’s, since they have presumably been shipping working hardware to Google)?

    It just seems like a weirdly long time between being done enough for a specific customer and general availability; especially when none of the individual ingredients(neoverse N1 cores, QAT or QAT-derived accelerators, and Intel 100GbE) are at the bleeding edge engineering samples only stage.

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.