HomeLab Christmas Update 2026

How the my Home-Lab evolved

I published My HomeLab 2025 back in March 2025. At that time, the lab was very much a classic, single-location setup – x86-based ESXi hosts, shared storage, and an environment designed for day-to-day testing, learning, and validation.

Since then, the lab has not grown dramatically in size – but it has evolved in depth and intent. Over the course of 2025 and throughout 2026, the focus shifted toward better storage design, distributed edge-style setups, and on-demand enterprise compute, rather than adding more always-on hardware.

This update captures that evolution.

Storage Evolution – Synology as the Core Backbone

One of the most impactful changes since March 2025 happened on the storage side.

My Synology DS923+ was significantly expanded and now serves as the central storage backbone of the HomeLab.

DS923+ Configuration

4 x 12 TB HDDs for capacity-focused workloads
2 x 512 GB NVMe SSDs used as a dedicated VM workload datastore
10GbE NIC providing high-throughput connectivity to ESXi

Instead of using NVMe as cache, the SSDs are assigned directly to VM workloads where predictable latency and consistent performance matter more than raw capacity.

Storage Services

The DS923+ now provides:

NFS datastores for ESXi
iSCSI LUNs for selected workloads and testing scenarios
VM templates and ISO repositories
Centralized storage for lab services and automation

For a compact 4-bay NAS, this setup strikes an excellent balance between performance, flexibility, and power efficiency.

Distributed Storage for ESXi on ARM

As the lab expanded beyond a single physical location, local storage became increasingly important – especially for lightweight, site-local services.

To support the distributed ESXi on ARM setup, I repurposed two older Synology systems:

Synology DS214+
Synology DS414slim

Both systems are now deployed close to the Raspberry Pi 5 ESXi hosts and provide:

Local NFS datastores
Targets for encrypted backup synchronization
Storage for lightweight, site-local services

This keeps the remote ARM hosts independent and resilient, while still fitting cleanly into the overall lab architecture – very similar to how small edge sites are designed in real-world environments.

On-Demand Compute – Dell PowerEdge R740

The most notable addition in 2025 was the introduction of a Dell PowerEdge R740, which fundamentally changed how heavy workloads are handled in the lab.

Instead of expanding the always-on baseline, this system was added with a clear philosophy:

Enterprise-grade compute power – available only when needed.

The R740 is normally powered off and brought online via iDRAC only for specific scenarios – primarily VMware Cloud Foundation (VCF) lab testing, large-scale simulations, or memory- and CPU-intensive experiments.

Hardware Overview

2 x Intel Xeon Gold 6134 @ 3.20 GHz
512 GB ECC RAM
Local SSD RAID storage
NVIDIA Tesla M10
FusionIO ioDrive2

This approach keeps power consumption, noise, and heat under control, while still allowing realistic enterprise-scale testing when required.

Hardware Reality – Powerful, but with Clear Limits

Not every component in the R740 is modern – and that is very much intentional. This system was added to the lab with a clear understanding of its strengths and its boundaries.

NVIDIA Tesla M10 – Density First, Compute Second

The NVIDIA Tesla M10 is a good example of hardware that still has value, but only in the right context.

Architecturally, the M10 is based on Maxwell (GM107). Its original design goal was VDI density, not compute acceleration:

Four physical GPUs on a single card
Optimized for vGPU consolidation
Low power draw per GPU
Designed for graphics offload, not heavy math

What it does not provide:

No Tensor Cores
No modern CUDA feature set
Very limited FP16 or FP32 throughput
No realistic support for AI or ML workloads

Modern AI pipelines rely heavily on tensor operations, high-bandwidth memory, and newer CUDA features. The Tesla M10 predates this ecosystem entirely, which makes it a poor choice for AI workloads.

In practice, the M10 is still useful for:

Density-focused vGPU testing
Legacy VDI-style labs
Understanding older NVIDIA virtualization models

FusionIO ioDrive2 – Fast Hardware, Frozen in Time

The FusionIO ioDrive2 tells a similar story from the storage side.

From a hardware perspective, the ioDrive2 is still impressive:

Extremely low latency
High IOPS
PCIe flash long before NVMe became standard

The limiting factor today is software support:

Native ESXi driver support is discontinued
The device is no longer usable as an ESXi datastore
Operation is only possible via PCI passthrough
Guest access relies on community-maintained Linux drivers

This means:

No integration with ESXi storage stacks
No hypervisor-level visibility
No production-safe upgrade path

In the lab, that limits the ioDrive2 to:

Passthrough experiments
Linux-based performance testing
Historical and architectural comparisons

It is a strong reminder that raw performance alone is meaningless without long-term platform support.

Why This Still Works

Despite these constraints, the R740 fits perfectly into the HomeLab.

The entire system was free of charge, courtesy of an old friend. I have been doing consulting work for him for more than 10 years, and when this hardware was retired from his environment, it felt more like a continuation of a long professional relationship than a lucky coincidence.

That context matters. It turns the hardware into something very different:

The Tesla M10 becomes a learning tool, not an expectation mismatch
The ioDrive2 becomes an experiment, not a dependency
The R740 becomes a powerful, on-demand platform with clearly understood limits

For a HomeLab, this is often the ideal balance – enterprise-grade hardware, realistic constraints, zero acquisition cost.

Closing Thoughts

This 2026 update reflects a HomeLab that has matured rather than expanded. The focus is no longer on adding hardware for its own sake, but on architecture, efficiency, and intent.

Over time, the lab has become less about raw capacity and more about:

using the right tool for the right job
understanding the limits of hardware
designing environments that resemble real-world constraints

Sometimes the most valuable upgrades are not new components at all – they are better decisions and clearer expectations.

hcidude's blog