The OpenMetal Bare Metal Dedicated Server XL v4 is the top-tier server in the OpenMetal bare metal lineup, succeeding the XL v3 with 5th Gen Intel Xeon Gold 6530 processors (Emerald Rapids), 1TB DDR5 4800MHz RAM across 16 DIMM slots, and four Micron 7500 MAX 6400GB NVMe drives delivering 25.6TB of persistent, direct-attached storage. The XL v4 is a single-tenant dedicated server — one customer, 64 physical cores, no shared hypervisor, no co-tenant resource contention. Fixed monthly pricing and a 99.96%+ network SLA apply from day one. The XL v4 replaces the XL v3’s Xeon Gold 6430 processors with Emerald Rapids silicon, adding Intel AMX for hardware matrix acceleration, higher per-core TDP, and improved memory bandwidth at DDR5-4800.
Key Takeaways
- 64 dedicated physical cores / 128 threads — all cores belong to one customer; no vCPU oversubscription, no hypervisor tax on bare metal
- 1TB DDR5 4800MHz ECC across 16 DIMM slots — 614 GB/s of memory bandwidth and full DIMM population activates Intel TDX by default for hardware-isolated Trust Domains
- 25.6TB Micron 7500 MAX NVMe (4x 6.4TB) — 1.1M random read IOPS per drive, sub-1ms 6-nines QoS, persistent across reboots and OS reinstalls
- Intel AMX for CPU-based ML inference — BF16 and INT8 matrix acceleration for quantized LLM inference without a GPU; 30B–70B parameter models fit in the 1TB RAM pool
- Intel TDX enabled by default — hardware-isolated, memory-encrypted Trust Domains available from day one; no RAM upgrade required at the XL v4 baseline
- 20Gbps private bandwidth (LACP-bonded), 6Gbps public with 95th-percentile egress billing — burst traffic does not accumulate per-GB charges below the 95th-percentile threshold
Server Configuration at a Glance
| Processor | 2x Intel Xeon Gold 6530 (Emerald Rapids, 5th Gen Intel Xeon Scalable) |
| Total Cores / Threads | 64 cores / 128 threads |
| Base / Max Turbo | 2.10 GHz / 4.0 GHz |
| L3 Cache | 160MB per socket (320MB total) |
| TDP | 270W per socket |
| Memory | 1024GB (1TB) DDR5 4800MHz ECC — 16 DIMM slots fully populated |
| Boot Storage | 2x 960GB SSD RAID 1 (dedicated OS path) |
| Data Storage | 4x Micron 7500 MAX 6400GB NVMe U.3 (25.6TB usable) |
| Max Drive Bays | 10 total |
| Private Bandwidth | 20Gbps (2x 10Gbps LACP bonded) |
| Public Bandwidth | 6 Gbps |
| Confidential Computing | Intel TDX (default, all 16 DIMMs populated) + Intel SGX (128GB EPC max) + TME-MK |
| Remote Management | Full IPMI out-of-band console |
| Pricing | Fixed monthly — see openmetal.io/bare-metal-pricing |
Processor
The Intel Xeon Gold 6530 is an Emerald Rapids processor (5th Gen Intel Xeon Scalable, Intel 7 process node) with 32 cores per socket at 2.10 GHz base and 4.0 GHz max turbo. Two sockets provide 64 physical cores and 128 threads total. L3 cache is 320MB across both sockets , and PCIe 5.0 lanes provide high-bandwidth connectivity to the four NVMe drives and network interfaces.
Key ISA extensions relevant to production workloads: Intel AMX (Advanced Matrix Extensions) provides dedicated matrix multiply hardware for BF16 and INT8 operations — the core arithmetic for quantized LLM inference and recommendation engines. Intel AVX-512 handles SIMD-intensive workloads including vectorized database scans, signal processing, and HPC kernels. Intel QAT offloads TLS termination and AES operations to a dedicated acceleration engine, keeping cryptographic overhead off the compute cores.
TDX and AMX operate concurrently — confidential AI inference inside a TDX Trust Domain runs at full AMX matrix throughput with no throughput penalty from the hardware memory encryption. For GPU-accelerated training, contact OpenMetal about GPU server configurations; the XL v4 serves CPU-based inference at scale.
Memory
1024GB (1TB) of DDR5 4800MHz ECC RAM across 16 DIMM slots delivers 614 GB/s of aggregate memory bandwidth. At full DIMM population, this configuration meets Intel’s requirement for TDX Trust Domain activation — TDX is available by default on every XL v4 without a customer-initiated RAM upgrade. ECC protection ensures bit-error detection and correction for production database, analytics, and compliance workloads.
The 1TB pool accommodates large in-memory workloads: a PostgreSQL instance with a 600GB working set fits entirely in RAM, eliminating disk reads on hot paths. INT8-quantized language models in the 30B–70B parameter range fit within the memory budget without weight swapping to NVMe. VM fleets with 64GB–128GB per VM can run 8–15 concurrent VMs on a single XL v4 without memory contention.
Storage
Boot and data drive isolation
The XL v4 separates boot drives from data drives at the hardware level. Two 960GB SSDs in RAID 1 serve the OS path exclusively — data drives are never co-located with OS partitions. An OS reinstall via IPMI does not touch the four Micron 7500 MAX data drives; data persists across OS changes, maintenance windows, and reboots.
Micron 7500 MAX NVMe drives
Four Micron 7500 MAX 6400GB NVMe U.3 drives provide 25.6TB of usable persistent storage with the following per-drive performance specifications (source: Micron 7500 NVMe SSD Tech Prod Spec Rev. A 10/2023):
| Metric | Micron 7500 MAX 6400GB |
|---|---|
| Sequential Read | 7,000 MB/s |
| Sequential Write | 5,900 MB/s |
| Random Read IOPS | 1,100,000 |
| Random Write IOPS | 400,000 |
| QoS (99.9999%) | Sub-1ms (6-nines) |
| Endurance (TBW) | 16,800 TBW per drive |
| Persistence | Survives reboots, OS reinstalls, maintenance windows |
With 4 drives, aggregate sequential read throughput is 28 GB/s and aggregate random read IOPS reaches 4.4M. Max drive bay capacity is 10 — contact OpenMetal to add drives beyond the standard 4-drive configuration.
Networking
The XL v4 provides 20Gbps of private bandwidth via two 10Gbps uplinks bonded with LACP (IEEE 802.3ad). East-west traffic between XL v4 nodes on the same OpenMetal private VLAN — Ceph replication, database replication, Kubernetes pod-to-pod traffic — traverses the private fabric unmetered.
Public bandwidth is 6Gbps with DDoS protection up to 10Gbps per IP and a 99.96%+ network SLA (actual measured availability typically exceeds 99.99%). Public egress is billed on the 95th-percentile model.
Egress pricing: 95th-percentile billing, not per-GB transfer
OpenMetal measures egress traffic in 5-minute intervals across the billing period and discards the top 5% of measurements (the burst peaks). The billing rate is calculated against the 95th-percentile sustained rate, not total bytes transferred. Workloads with bursty egress patterns — blockchain node traffic, batch export jobs, media APIs — do not accumulate per-GB charges for isolated burst events that fall in the discarded top 5%. Public cloud providers charge per-GB for all egress; at 50TB/month of public egress, AWS us-east-1 charges approximately $4,500/month in transfer fees alone.
Security, Compliance, and Confidential Computing
Intel TDX (Trust Domain Extensions) is active by default on every XL v4 — the full 16-DIMM, 1TB RAM configuration meets Intel’s memory population requirement for Trust Domain activation. Each TDX Trust Domain receives hardware-encrypted memory pages with a unique key held only by the CPU and the guest OS; the host operating system, hypervisor, and OpenMetal operators cannot read Trust Domain memory. For workloads requiring hardware-level isolation from the infrastructure operator, TDX on bare metal provides a stronger isolation posture than enclave services on shared cloud hypervisors.
Intel SGX is available alongside TDX with up to 128GB of Enclave Page Cache (EPC) for application-level confidential workloads. TDX and SGX can run simultaneously — a key management service can run in an SGX enclave while the broader application stack runs inside a TDX Trust Domain on the same physical server. TME-MK (Total Memory Encryption — Multi-Key) provides per-VM AES encryption across the full 1TB DRAM range beneath TDX.
- Intel AES-NI — hardware AES acceleration, zero CPU overhead on encryption workloads
- Intel Boot Guard — UEFI firmware attestation before OS load, prevents pre-OS compromise
- Intel CET — Control-Flow Enforcement Technology for hardware stack and branch protection
HIPAA and Regulatory Compliance
OpenMetal executes HIPAA Business Associate Agreements (BAAs) at the organizational level — one BAA covers all workloads running on OpenMetal infrastructure. Facility-level certifications are held by data center operators, not by OpenMetal directly.
XL v4 servers deployed in Ashburn are hosted at an NTT DATA facility that holds HIPAA as a facility-operator certification, alongside SOC 1/2 Type II, ISO 27001, PCI DSS, and NIST 800-53 HIGH. Los Angeles is hosted at a Digital Realty facility with SOC 2, SOC 3, ISO 27001, and PCI DSS — HIPAA at Los Angeles is OpenMetal organizational-level only, not a facility cert. Amsterdam (Digital Realty) holds SOC 1/2, PCI-DSS, ISO 27001, ISO 50001, and ISO 22301. Singapore (Digital Realty) holds BCA Green Mark Platinum.
Recommended Workloads
Databases and transaction processing
PostgreSQL, MySQL, SQL Server, and Oracle workloads that require large buffer pools, high random IOPS, and consistent per-core latency benefit from the XL v4’s 1TB RAM, 1.1M IOPS per NVMe drive, and dedicated 64-core allocation with no hypervisor scheduling overhead. ACID-compliant transaction workloads that cannot tolerate CPU contention from co-tenants run predictably on bare metal.
Kubernetes and container orchestration
Kubernetes clusters running on Talos Linux, Flatcar, or K3s benefit from the 64-core, 1TB compute budget. A single XL v4 can host a large Kubernetes node with 60+ allocatable cores and 900GB+ of allocatable memory, or serve as a worker node in a multi-server cluster where each XL v4 contributes dedicated resources without interference. Boot/data isolation ensures container image layers and persistent volumes are on separate NVMe paths from the OS.
Blockchain infrastructure
Ethereum archive nodes, validator clusters, and high-throughput RPC endpoints use the 25.6TB NVMe pool for chain state, the 20Gbps private bandwidth for peer traffic, and the 95th-percentile egress billing to avoid unpredictable per-GB charges for bursty RPC response traffic. IPMI provides OS reinstall and node recovery without contacting a cloud support ticket queue.
Virtualization and multi-tenant hosting
Proxmox VE and KVM-based multi-tenant hosting providers use the XL v4’s 64 cores and 1TB RAM to host dense VM fleets. The Level1Techs Proxmox architecture on OpenMetal bare metal provides a proven configuration for MSPs and hosting providers building private cloud offerings on dedicated hardware.
ML inference (CPU-based, quantized models)
Quantized LLMs in the 30B–70B parameter range fit in the 1TB RAM pool. Intel AMX provides BF16 matrix multiply acceleration for inference throughput. TDX wraps inference workloads in hardware-encrypted Trust Domains for confidential inference pipelines processing private documents or medical records. For GPU-accelerated training, contact OpenMetal about GPU server configurations.
Data analytics and batch processing
Apache Spark, ClickHouse, and Presto clusters that require large memory-mapped datasets, high sequential NVMe read throughput (28 GB/s aggregate across 4 drives), and predictable egress costs for query result exports run efficiently on the XL v4. Fixed monthly pricing simplifies cost-per-query modeling compared to pay-per-query cloud analytics services.
Ready to Deploy a XL v4?
Tell us about your workload and we’ll help you configure the right deployment — bare metal or Hosted Private Cloud, in any of our four data center regions.
XL v4 vs AWS
The closest AWS equivalent by RAM and NVMe storage profile is the AWS i4i.metal (128 vCPUs, 1024GB RAM, 30.72TB ephemeral NVMe). The fundamental structural differences: OpenMetal provides 64 dedicated physical cores on single-tenant hardware; AWS i4i instances use shared-hypervisor vCPUs. OpenMetal NVMe is persistent across reboots; AWS i4i instance storage is ephemeral and lost on stop/terminate. OpenMetal bills a fixed monthly rate; AWS charges $10–$13/hr on-demand plus per-GB egress and EBS storage for persistence.
For a full structural cost comparison including TCO modeling and egress analysis, see the OpenMetal XL v4 vs AWS i4i comparison page.
XL v3 vs XL v4: What Changed
| Component | XL v3 | XL v4 | Improvement |
|---|---|---|---|
| CPU Model | 2x Xeon Gold 6430 (Sapphire Rapids) | 2x Xeon Gold 6530 (Emerald Rapids) | Architecture generation +1 |
| Base Frequency | 2.10 GHz | 2.10 GHz | Same baseline |
| Max Turbo | 3.4 GHz | 4.0 GHz | +17% peak clock |
| L3 Cache | 60MB per socket | 160MB per socket | +167% L3 cache per socket |
| Intel AMX | Yes (Sapphire Rapids) | Yes (Emerald Rapids, improved) | Available in both generations |
| PCIe Generation | PCIe 5.0 | PCIe 5.0 | Same |
| Memory Speed | DDR5 4800MHz | DDR5 4800MHz | Same speed, same capacity |
| Data NVMe | 4x Micron 7500 MAX 6400GB | 4x Micron 7500 MAX 6400GB | Same drives |
| Boot Drives | 2x 960GB RAID 1 | 2x 960GB RAID 1 | No change |
| Intel TDX | Eligible (1TB RAM, upgrade path) | Enabled by default | Default on XL v4, upgrade on XL v3 |
The headline improvement from XL v3 to XL v4 is the jump in L3 cache per socket (60MB to 160MB), which reduces memory latency for cache-resident working sets in database and analytics workloads, and the increase in max turbo from 3.4 GHz to 4.0 GHz for single-thread-sensitive applications.
Deployment Options
Bare Metal Dedicated Server
Full root access, IPMI out-of-band console, no shared tenancy. Pre-built OS images available (Ubuntu, CentOS, Rocky Linux, Debian, and others). Price locks available up to 5 years; ramp pricing for migrations from other providers. Deploy an XL v4 bare metal server in Ashburn, Los Angeles, Amsterdam, or Singapore.
→ View pricing: openmetal.io/bare-metal-pricing
Hosted Private Cloud
Three-node OpenStack + Ceph cluster on XL v4 hardware. Day 2 operations (monitoring, alerting, platform updates, storage management) handled by OpenMetal. No VMware licensing, no OpenStack or Ceph licensing. Deploys in under 45 seconds via the OpenMetal control plane. Proof of Concept clusters available for testing.
→ View pricing and configuration: openmetal.io/cloud-deployment-calculator
Where to deploy
Deploy an XL v4 bare metal server or XL v4 Hosted Private Cloud in Ashburn, Los Angeles, Amsterdam, or Singapore.
| Location | Region | Certifications | Location Page |
|---|---|---|---|
| Ashburn, VA | US-East | SOC 1/2 Type II, ISO 27001, PCI DSS, NIST 800-53 HIGH, HIPAA (facility) | Ashburn |
| Los Angeles, CA | US-West | SOC 2, SOC 3, ISO 27001, PCI DSS | Los Angeles |
| Amsterdam | EU-West | SOC 1/2, PCI-DSS, ISO 27001, ISO 50001, ISO 22301 | Amsterdam |
| Singapore | Asia | BCA Green Mark Platinum | Singapore |
Get a XL v4 Quote
Tell us about your infrastructure needs and we’ll provide a custom quote for the XL v4.
- Bare metal: Single-server or multi-server deployments with full root access and IPMI
- Hosted Private Cloud: Three-node OpenStack + Ceph clusters with Day 2 operations included
- Custom configurations: RAM upgrades, additional NVMe drives, TDX enablement
Ramp pricing available for migrations. All deployments include fixed monthly pricing, 99.96%+ network SLA, and DDoS protection.
Product specifications, pricing, and availability may change due to market conditions and other factors. For the most current information, please contact the OpenMetal team directly.