Frank Denneman

Chief Technologist AI | Distinguished Engineer at VCF Broadcom | Publishing at | Author vSphere Deep Dive books

[Frank Denneman @FrankDenneman]

Do you create vSphere clusters that contain ESXi hosts with different CPU and memory configurations?

[Frank Denneman @FrankDenneman]

@alexellisuk haven't looked at that yet. Will post some results soon

[Frank Denneman @FrankDenneman]

DGX Spark arrived Friday.

First thing I did was run NemoClaw with Nemotron locally. No cloud, just a small box on my desk acting like a personal AI runtime.

What stood out wasn’t that it works, but how different unified memory feels. The usual GPU signals no longer apply.

Wrote up what I found: https://t.co/xywEsIivhM #DGX #NVIDIA

[Frank Denneman @FrankDenneman]

Part 10 of Architecting AI Infrastructure is live.

Multi-GPU inside a single host is not just about adding more GPUs. Once a model spans GPUs, communication becomes part of the runtime path. This article walks through the full spectrum:

PCIe
NVLink domains
4 GPU NVLink
8 GPU split domains
HGX with NVSwitch

From independent accelerators to a unified GPU fabric inside a single server. Understanding this is key to scaling distributed inference and MoE models.

Part 10: Understanding Multi-GPU Topologies Within a Single Host:
https://t.co/6wVpeZcimr

[Frank Denneman @FrankDenneman]

Agentic AI changes how GPUs are used.

Instead of short prompt–response interactions, agents keep sessions alive while they reason, call tools, and loop through workflows. The KV cache doesn’t disappear; it accumulates.

That changes GPU memory planning.

New article
https://t.co/etQL262bWU

Explains why durable sessions, multi-step reasoning, and tool calls create persistent GPU memory pressure, and why infrastructure teams need to start planning for it.

#AI #GPU #NVIDIA #VCF

[Frank Denneman @FrankDenneman]

New tool: Same-size vs Mixed-size Placement Simulator
See how vGPU profile choices shape cluster-wide GPU capacity over time. Not just per GPU.
Walkthrough:
https://t.co/5wJpsB1iCV
Simulator:
https://t.co/ugaH15ZYXJ

[Robert Scoble @Scobleizer]

@sama In depth analysis (wrapup of everything that happened here on X too) on why you are able to make a deal where @DarioAmodei can't: https://t.co/RJpwZ0pPzP

[Frank Denneman @FrankDenneman]

@DCasota AFAIK, right now we do not have support for ASICs. Just 'regular' GPUs from NVIDIA, AMD, and Intel. Are you talking about the Taalas device, right? Looks really promising.

[Frank Denneman @FrankDenneman]

Part 5 of the Architecting AI Infrastructure series is live.
This article shows how startup order and assignment policy shape fragmentation, consolidation, and future placement opportunities.
https://t.co/NXBDNxmFh1

[Frank Denneman @FrankDenneman]

Started an 'Architecting AI Infrastructure' series. Part 4 published today: https://t.co/450tKEyfBx

[Frank Denneman @FrankDenneman]

new blog post: Which Multi-GPU Configurations Are You Planning to Deploy? https://t.co/m8mhYJOf37 #AI #GPUs #NVIDIA

[Frank Denneman @FrankDenneman]

Explore was incredible again. Attendees were locked in. Great conversations with customers, full focus during break-out sessions. It was an awesome week, one of the best I've ever had. #vmwareexplore2025

[Frank Denneman @FrankDenneman]

Why you want to use Enhanced vMotion for vGPU VMs in VCF 9.0 for LLM workloads:

https://t.co/6UMkZYrSLT

#AI #LLM #VCF #NVIDIA #VMware #vMotion

[Sarbjeet Johal @sarbjeetjohal]

A WHOLE NEW 'kitten kaboodle' of software systems!
An excerpt from 'Cloud Native AI' paper by @CloudNativeFdn, shared by @FrankDenneman of @VMware [by @Broadcom]

Link to paper: https://t.co/vURjpbdqvT

[Christian Mohn™ @h0bbel]

Hey it’s @FrankDenneman in the @virtspeaking studio at #VMwareExplore

[Waldemar Pera (@[email protected]) @WaldemarPera]

@VMware (by @Broadcom) #VCF #LLM #RAG - Building an Efficient AI Ingestion Pipeline: A Guide for Infrastructure Teams on Data Transformation and Vector Database Optimization (by @FrankDenneman) @VMwareCloud @VMwareVCF #vExpertHybridCloud @vExpert https://t.co/FZly3Xl4un

[Frank Denneman @FrankDenneman]

Check out my latest article: Building an Efficient AI Ingestion Pipeline: Data Ingestion Strategies https://t.co/EtVxRXyFql via @LinkedIn

[Frank Denneman @FrankDenneman]

New blog post: Building an Efficient AI Ingestion Pipeline: A Guide for Infrastructure Teams on Data Transformation and Vector Database Optimization

https://t.co/PSq5fKb8N6

#AI #RAG #VMware

[Alan Renouf @alanrenouf]

@jamesrvh724 @lamw I watched the AI session from @vhojan and @FrankDenneman - that’s a must watch!

[Frank Denneman @FrankDenneman]

@johjustinmurray helping that audience understand the key concepts of VMware Private AI Foundation with NVIDIA. Waiting line outside #AI