40-core dedicated servers
Dual Xeon Gold 6138 delivers 40 physical cores (80 threads with HT). Designed for high-concurrency streaming, virtualization density and multi-tenant panels.
Best-fit dedicated plans
These are the closest stock plans in our catalogue for 40-core dedicated server workloads. Fully customizable — port, storage, RAM, DDoS profile.
Xeon E5-2690 v3 · 128 GB · 10G
2× Intel® Xeon® E5-2690 v3 · 128 GB DDR4 ECC · 2× 512GB NVMe · 10 Gbit/s unmetered · DDoS-protected
Dual Xeon Gold 6138 · 128 GB · up to 100G
2× Intel® Xeon® Gold 6138 · 128 GB DDR4 ECC · 2× 1TB NVMe · 10/20/40/100 Gbit unmetered
Xeon Platinum 8168 · 256 GB · 4× NVMe
1× Intel® Xeon® Platinum 8168 · 256 GB DDR4 ECC · 4× 1.92TB NVMe · configurable bays
About 40-core dedicated server
40 physical cores across 2 sockets is where "big server" starts. Dual Xeon Gold 6138 gives you the PCIe lanes for 100G networking, the memory channels for bandwidth-hungry workloads, and the raw core count to fill them.
This platform is our default for national-scale IPTV panels (50-100k active lines), Tier-1 CDN origins, live-event streaming, big virtualization hosts (80+ KVM guests), and Kubernetes control-plane + heavy worker nodes on the same box.
AVX-512 is enabled — you get significant speedup on video codec workloads, cryptography, and modern quantized ML inference. NVMe Gen3 storage standard, expandable to 8 drives with backplane.
Uplink options: 10 / 20 / 40 / 100 Gbit/s unmetered on QSFP28 optics. LACP bonding to 200G aggregate on request. In stock at Amsterdam and Kyiv; 72 h provisioning at Warsaw and Miami.
Frequently asked questions
Do I need 40 cores?
For streaming/CDN workloads: usually only if you're at 50k+ concurrent viewers per node OR doing real-time transcode. For pure passthrough, 24 cores handle even 100k viewers.
What's the NUMA topology?
2 NUMA nodes (one per socket). We tune kernel scheduler for NUMA-aware workloads by default. Applications should pin threads per NUMA node for best cache locality.
40 cores vs single-socket 32-core EPYC?
EPYC 7532 (32c/64t Milan) has slightly better single-thread perf and 4× more L3 cache, but no AVX-512. Pick EPYC for compute density, dual-Gold 6138 for PCIe lanes + AVX-512.