Bare metal that moves
at cloud speed
Single-tenant AMD EPYC and Intel Xeon servers with DDR5 ECC and NVMe RAID — racked, imaged, and answering SSH in hours, from $0.37/hr. Every server ships with 20 TB of egress included, and virtual CPU servers start at $14/mo when you don't need the whole box.
Hourly or monthly reserved · IPMI access included · Four regions · No setup fees
Chassis allocated · rack A14
US East (Ashburn) · single tenant
2× 1.92 TB NVMe imaged
RAID1 mirror · Ubuntu 24.04 LTS
Configuring uplinks + IPMI
25 Gbps · 20 TB egress
Handing off SSH keys
Bare metal dedicated servers
Four chassis, one meter. Current-generation EPYC and Xeon silicon, always single-tenant, with DDR5 ECC and mirrored NVMe in every configuration. Pay by the hour or reserve monthly — release whenever, and the meter stops.
Metal 4E Starter
AMD EPYC 4344P (8C/16T Zen 4)
Single-tenant entry point for game servers, build farms, and small clusters.
Threads
16
Memory
64 GB ECC
Storage
2× 960 GB NVMe RAID1
Network
10 Gbps · 20 TB egress
$0.37/hr
$229/mo reserved
Metal X16
Intel Xeon Gold 6526Y (16C/32T Emerald Rapids)
DDR5 ECC for steady production workloads.
Threads
32
Memory
128 GB ECC
Storage
2× 1.92 TB NVMe
Network
10 Gbps · 20 TB egress
$0.62/hr
$379/mo reserved
Metal E32
Most popularAMD EPYC 9354P (32C/64T Genoa)
The sweet spot for databases, analytics, and virtualization hosts.
Threads
64
Memory
256 GB ECC
Storage
2× 1.92 TB NVMe
Network
25 Gbps · 20 TB egress
$1.05/hr
$649/mo reserved
Metal E64
AMD EPYC 9554P (64C/128T Genoa)
768 GB ECC — consolidation-class hardware, delivered in hours.
Threads
128
Memory
768 GB ECC
Storage
4× 3.84 TB NVMe
Network
2× 25 Gbps · 20 TB egress
$2.99/hr
$1,899/mo reserved
Every bare-metal price includes 20 TB of monthly egress, IPMI/KVM-over-IP access, and RAID configuration at no charge. Reserved monthly rates shown alongside hourly metering.
Virtual CPU servers
For everything around the big jobs — APIs, queues, control planes, cron. Shared vCPU from $14/mo, or dedicated cores with zero steal time when latency matters. Same images, same API, same console as the rest of the fleet.
Full price listVC-S2 Shared
Dev and test boxes, small web apps, CI runners, bastion hosts.
- vCPUs 2
- Memory 4 GB
- Storage 80 GB NVMe
- Transfer 4 TB
$0.02/hr
$14/mo
VC-S4 Shared
Small production APIs, staging environments, low-traffic SaaS backends.
- vCPUs 4
- Memory 8 GB
- Storage 160 GB NVMe
- Transfer 5 TB
$0.04/hr
$28/mo
VC-D4 Dedicated
Dedicated vCPU with no noisy neighbors — production databases and latency-sensitive services.
- vCPUs 4
- Memory 16 GB
- Storage 200 GB NVMe
- Transfer 6 TB
$0.10/hr
$69/mo
VC-D8 Dedicated
Kubernetes control planes and data-prep pipelines feeding your GPU jobs.
- vCPUs 8
- Memory 32 GB
- Storage 400 GB NVMe
- Transfer 8 TB
$0.20/hr
$135/mo
Server-grade by default
Every chassis in the fleet ships the same way: one tenant, ECC memory, mirrored NVMe, and real bandwidth. The spec sheet is the baseline, not the upsell.
One tenant per box: you
There is no hypervisor layer and no oversubscription. Your OS boots directly on the metal, your workloads own every core and every byte of cache, and benchmarks are reproducible run after run — because the host never changes underneath you.
- IPMI / KVM-over-IP access on every server, no add-on fee
- Ubuntu 24.04, Debian 12, Rocky 9 — or boot your own ISO
- Simpler compliance story: dedicated hardware, auditable by serial number
Hypervisor: none · CPU steal time: 0.0%
Memory that corrects, disks that mirror
Every chassis carries DDR5 ECC memory — from 64 GB to 768 GB — and enterprise NVMe in RAID1 by default. A silent bit flip or a dying drive becomes a log line, not an outage.
- RAID1 standard; RAID0 or RAID10 on request at no charge
- Hot-spare drive pool in every rack row
- Failed parts swapped by on-site engineers, usually within the hour
10–25 Gbps up. 20 TB out. Included.
Dedicated uplinks on every server — no burst credits, no traffic shaping — and 20 TB of monthly egress in the sticker price. Traffic to your Reviosa GPU pods and storage volumes rides the private fabric and never touches the meter.
- 10 Gbps standard, 25 Gbps on E-class, 2× 25 Gbps on the E64
- 20 TB egress included per server, flat overage after — no regional tiers
- Free private VLANs between your servers, VMs, and GPU instances
peak 18.4 Gbps of 25 Gbps uplink
Private fabric to GPU pods · unmetered
Racked, imaged, and online in hours
Dedicated hardware used to mean quote calls, POs, and a six-week lead time. On Reviosa you order a server the way you launch a VM: pick a chassis in the console, and we rack, image, and hand you SSH the same day.
- Order from the console, CLI, or Terraform — no sales call
- Median time from order to SSH: 2 h 40 m across all four regions
- Release anytime on hourly billing — no 12-month contract
-
Order placed
10:42 · console
-
Chassis racked & cabled
10:44 · US East (Ashburn), rack A14
-
RAID1 imaged · Ubuntu 24.04
11:26 · 2× NVMe mirrored
-
SSH-ready
13:07 · 2 h 25 m from order
Hyperscaler compute, without the hyperscaler math
Dedicated hosts on the big clouds bury the real price in egress tiers, support plans, and multi-year reservations. We publish one number and put the hardware behind it.
Tenancy
Single-tenant metal — your cores, your cache, your NUMA topology
Shared hosts where a neighbor's batch job throttles your p99s
Egress
20 TB included with every server, one flat overage rate
Metered per GB from the first byte, tiered by region and service
Provisioning
Ordered from the console, SSH-ready in hours
Quote requests and multi-week lead times for dedicated hosts
Commitment
Hourly metering or monthly reserved — release anytime
One- to three-year reservations to reach a workable rate
Support
24/7 engineers who can walk to the rack and pull a drive
Ticket tiers sold separately, escalation measured in days
$0.37/hr
entry dedicated server
20 TB
egress included, every server
128 threads
in the largest chassis
Elena Voss
Head of Infrastructure, Northfork Render
“We replaced sixty reserved hyperscaler instances with sixteen Metal E32s. Same render throughput, 41% off the monthly bill, and frame times stopped spiking every time a neighbor got busy. Procurement was an afternoon, not a quarter.”
41%
lower monthly infrastructure spend after moving the render fleet to bare metal
Read customer storiesThe rest of the fleet
Bare metal is one corner of the platform. Wire your servers into GPU pods, replicated storage, and serverless endpoints — all on the same private fabric, all on one bill.
GPU Cloud
On-demand B200, H200, and H100. From a single card to 8× NVLink nodes on 3.2 Tbps InfiniBand.
Storage
NVMe blocks and S3-compatible objects. Triple-replicated, attaches to any server in seconds, free ingress.
Serverless AI
Inference that scales to zero. Open-weight models behind autoscaling endpoints — pay per token.
Start training in minutes
Create an account, add a card, and launch your first GPU instance. Per-second billing means you only pay for what you use.
No minimum commitment · Cancel anytime · $10 free credit for new accounts