Amazon announced that they will be using Nvidia’s NVSwitch to create new rackscale AI platforms. This allows customers to use Nvidia technology like Grace Hopper or build something using AWS Nitro DPUs and Elastic Fabric Adapaters. That last combination doesn’t use InfiniBand and moves to Ethernet. Two new chips are coming out from the Amazon labs. The first is Graviton4, the latest generation of Arm processor. Graviton4 has 50% more cores and 75% more memory bandwidth. This version of Graviton is based on the Demeter Neoverse V2 core. On the AI front Amazon also announced the next revision of their AI acceleration chip, Trainium2. This update has a 4x performance increase from the first generation and allows for liquid cooling. Trainium2 is designed to be deployed in clusters of 16 chips. In the data space, AWS made news with the GA of Bedrock, which enables customers to run foundational ML models from companies like Anthropic, Meta, and of course AWS itself. But many companies are worried about hallucination and want to include their own data in these models. That’s what AWS is delivering with Guardrails, and what AWS partners are leaning into as well. This and more on this week’s Gestalt IT Rundown.
1:18 – DAOS Foundation Launched for Object Storage
The Linux Foundation has launched a new initiative focused on object storage. The new foundation is focused on Distributed Asynchronous Object Storage, which gives the foundation its acronym of DAOS. DAOS is a high performance open source object storage system that is designed to work with HPC and AI. Intel has been a contributor to DAOS since 2012 and has donated their work to the new foundation to keep the project moving forward. Stephen, what do you think about DAOS?
4:57 – Autonomous Purple Team Announced by Skyhawk at AWS re:Invent
This week Skyhawk Security announced a new platform that is empowering attacking red teams and defending blue teams with AI. Their new feature is Adaptive Cloud Threat Detection and Response. It continuously analyzes customer cloud infrastructure looking for holes and entry points. What’s new is that the red team is AI-based and so is the blue team. This melding of the two aspects of threat analysis makes a very pretty purple for security teams.
8:26 – Couchbase’s Capella Columnar Service Revealed
Couchbase announced a new columnar data service this week at re:Invent. The new service, named Capella, allows for real-time analytics on the same platform as operational workloads. According to Scott Anderson, SVP of Product Management, this will reduce the latency between databases and analytics and will allow development teams to more quickly get statistics and feedback to make their applications better.
10:59 – Hackers Were Inside NXP For Two Years Before Detection
Netherlands-based semiconductor company NXP is in the news this week thanks to a new report that says that a Chinese-backed hacking group had the run of their network for over two years. The report states that the Chimera group had unfettered access starting in late 2017 and was only detected after breaching another area of the network and having their activity traced back. The attackers reportedly snooped through mail boxes and shared drives looking for chip designs and flaw reports. A cloud security research firm posted a report two years ago with details about Chimera using services like Dropbox to exfiltrate the information. The report was later deleted but the unnamed organization was later confirmed to be NXP.
Read More: Hackers spent 2+ years looting secrets of chipmaker NXP before being detected
14:47 – AWS re:Invent Announcements
We’re mostly talking hardware today, but there are other announcements like a better S3 and Guardrails around ML Bedrock.
Read More: Live updates from AWS re:Invent 2023
Read More: Here’s everything Amazon Web Services announced at AWS re:Invent
15:26 – NVSwitch For AI Nodes
Amazon announced that they will be using Nvidia’s NVSwitch to create new rackscale AI platforms. This allows customers to use Nvidia technology like Grace Hopper or build something using AWS Nitro DPUs and Elastic Fabric Adapters. That last combination doesn’t use InfiniBand and moves to Ethernet. Does this play into the announcement of Trainium2?
Read More: AWS Taps NVIDIA NVSwitch for Liquid Cooled, Rackscale AI GPU Nodes
20:46 – ARM V2 for Graviton4 and Trainium2
Two new chips are coming out from the Amazon labs. The first is Graviton4, the latest generation of Arm processor. Graviton4 has 50% more cores and 75% more memory bandwidth. This version of Graviton is based on the Demeter Neoverse V2 core. On the AI front Amazon also announced the next revision of their AI acceleration chip, Trainium2. This update has a 4x performance increase from the first generation and allows for liquid cooling. Trainium2 is designed to be deployed in clusters of 16 chips.
Read More: AWS Adopts Arm V2 Cores for Expansive Graviton4 Server CPU
Read More: AWS Graviton4 is an Even Bigger Arm Server Processor and Trainium2 for AI
25:22 – AWS Guardrails for Bedrock
In the data space, AWS made news with the GA of Bedrock, which enables customers to run foundational ML models from companies like Anthropic, Meta, and of course AWS itself. But many companies are worried about hallucination and want to include their own data in these models. That’s what AWS is delivering with Guardrails, and what AWS partners are leaning into as well.
30:10 – The Weeks Ahead
Cloud Field Day 19 – January 31 – February 1
Cisco Live EMEA – February 6 – February 7
AI Field Day 4 – February 21 – February 22
Networking Field Day 34 – February 28 – February 29
The Gestalt IT Rundown is a live weekly look at the IT news of the week. Be sure to subscribe to Gestalt IT on YouTube for even more weekly video content.
