Interested in our v5 hardware?
v5 hardware is currently available for deployment in our Ashburn, VA data center and can be ordered directly through our Bare Metal and Private Cloud catalogs. For deployments in other data center locations or for custom configurations, please contact our team.
Every team that runs serious infrastructure eventually hits the same wall. A workload that was fine on a shared cloud instance starts behaving unpredictably: latency spikes that don’t track your traffic, a database that’s quick until a neighbor on the same host gets busy, an IOPS ceiling you didn’t know existed until you hit it. That’s usually the moment bare metal stops looking old-fashioned and starts looking like the answer.
OpenMetal v5 hardware makes that case stronger than it’s been in a while, with Intel Xeon 6 processors, DDR5-6400 memory, and high-endurance NVMe attached directly to the server. This is the honest version of which workloads belong on bare metal, and what about the hardware actually makes the difference.
Why bare metal in the first place
The case for a dedicated server comes down to a few things you can’t get from a shared instance.
You get the whole machine. No hypervisor tax, no CPU steal, no noisy neighbor stealing memory bandwidth during your checkpoint flush. Performance is deterministic, which for latency-sensitive work matters more than any single benchmark number.
You get control. Full root plus BIOS and IPMI access means you can pin cores, disable hyperthreading per workload, set NUMA balancing, or boot an immutable OS from a signed image, without filing a ticket and waiting.
And you get the storage on the box. Every v5 server ships with high-endurance Micron 7500 MAX NVMe directly attached, plus a separate RAID 1 boot pair so an OS rebuild never touches your data. There’s no network storage round-trip and no per-IOPS meter running in the background.
With that as the backdrop, here’s where v5 bare metal earns its place.
Databases and transaction processing
Relational primaries are the textbook bare metal workload. Once the working set lives in memory, large PostgreSQL, MySQL, SQL Server, or Oracle instances bottleneck on memory bandwidth and storage latency, which is exactly what v5 improves. DDR5-6400 feeds the cores, the on-chassis NVMe delivers around 1.1 million random read IOPS per drive at sub-millisecond latency, and the boot/data isolation keeps write-ahead logging and checkpoint I/O on dedicated drives. Dedicated single-tenant hardware means no variance during vacuum, replication catch-up, or backup windows.
The Large v5 is often the right fit here, with the highest base clock in the lineup at 3.2 GHz and nearly double the L3 cache of the previous generation. For very large in-memory primaries, the XL v5 and its 1 TB of RAM gives you room to keep the whole dataset resident.
Virtualization and multi-tenant hosting
If you’re running Proxmox, KVM, or OpenStack compute nodes yourself, core count, ECC memory bandwidth, and hardware virtualization features (VT-x, VT-d, RDT, MBEC) are what determine density and isolation. The XL v5, with 64 cores and 1 TB of DDR5-6400, packs a lot of guests into one host; the Large v5’s higher clock makes individual VMs feel more responsive. OpenMetal publishes a full Proxmox reference architecture for bare metal, demonstrated by Wendell Wilson of Level1Techs on the previous generation, and it carries straight over to v5 with more headroom.
Choose v5 hardware as standalone bare metal servers or as a fully managed three-server hyperconverged private cloud. v5 is available now in Ashburn, VA (US East). Order today from the catalog and deploy in minutes.
Bare Metal Catalog Hosted Private Cloud Catalog
Kubernetes on bare metal
Running Kubernetes directly on hardware (Talos, Flatcar, K3s, or full distributions) removes the hypervisor layer entirely, which is the point for teams chasing density and predictable latency. Full IPMI and BIOS access supports immutable, PXE-booted operating systems, the boot and data drive isolation keeps container image churn off your persistent volumes, and the on-chassis NVMe backs Ceph RBD, Rook, or Longhorn without a separate storage tier. Three or more nodes on the 40 Gbps private fabric make a production-grade cluster.
CPU-based AI inference
Not every inference workload needs a GPU. The Granite Rapids P-cores in v5 include Intel AMX, AVX-512, and DL Boost, which accelerate INT8 and BF16 matrix operations, vectorized retrieval, and embedding pipelines. With up to 1 TB of DDR5-6400 you can hold large model weights in memory and serve many concurrent sessions, and the on-chassis NVMe gives low-latency access to vector indexes. When a workload does need a GPU, these servers pair with OpenMetal GPU servers over the same private fabric.
Data analytics and batch processing
Spark, Presto, and ClickHouse live and die on memory bandwidth and storage throughput. The Large v5 moves roughly 819 GB/s of aggregate memory bandwidth, AVX-512 accelerates vectorized and columnar scans, and the deep L3 cache keeps hot segments on-die. The Micron 7500 MAX handles spill-to-disk at hundreds of thousands of mixed IOPS, and multiple nodes on the private mesh handle distributed queries with low inter-node latency.
Blockchain infrastructure
Validators, archive nodes, and RPC endpoints want fast local NVMe for chain state, hardware crypto acceleration (AES-NI, SHA-NI) for signature and hash work, and predictable monthly cost. The piece that bites teams on public cloud is egress: a hot RPC endpoint racks up per-gigabyte charges fast. On v5, egress is billed on the 95th percentile rather than per gigabyte, and storage IOPS are included rather than metered, which gives node operators a stable cost model.
Regulated and confidential workloads
For data that needs protection while it’s in use, the XL v5 ships TDX-ready, because its 1 TB of memory populates one DIMM on every channel across both sockets — the full-channel layout TDX requires; a BIOS toggle activates it, with no RAM upgrade (unlike the Large and Medium tiers). Trust domains run in hardware-isolated, encrypted memory that the host OS, hypervisor, and OpenMetal staff cannot read. Intel SGX is enabled by default across the line for narrower enclave use, and the Large and Medium tiers can reach TDX with a memory upgrade. Combined with organization-level HIPAA and a BAA, that makes v5 a fit for regulated financial, healthcare, and sovereign workloads.
The common thread
What ties these together is determinism, control, and a cost model that doesn’t surprise you. You get the full machine, fixed monthly pricing, egress billed on the 95th percentile, and storage that’s part of the server rather than a metered add-on. If your workload is performance-critical, latency-sensitive, or simply expensive to run on metered cloud, bare metal is worth a serious look.
Read More Blog Posts