Consolidation over scaling out
Migrating from VMware to Proxmox VE is the perfect moment to move away from many older, under-utilised servers toward a few dense nodes. Instead of running a dozen aged dual-socket machines, we recommend consolidating onto a handful of hosts powered by the AMD EPYC 9755 — directly cutting OPEX (power, cooling, rack space) and per-socket support costs.
consolidation ratio
many older nodes
power and support
The point of hardware selection here is not "the most powerful CPU on earth" but the lowest total cost per virtual machine. A dense EPYC node reduces the number of physical hosts — and everything that depends on them.
AMD EPYC 9755 (Turin, Zen 5)
The EPYC 9755 is the flagship of the EPYC 9005 "Turin" line, built on the Zen 5 architecture. It combines extreme core density with a very wide memory and I/O subsystem — making it a universal base for mixed virtual environments.
| Parameter | AMD EPYC 9755 |
|---|---|
| Architecture | Zen 5 (Turin), socket SP5 |
| Cores / threads | 128 / 256 |
| Clock (base / boost) | 2.7 GHz / up to 4.1 GHz |
| L3 cache | 512 MB |
| Memory | 12-channel DDR5-6000 (up to 9 TB / socket) |
| PCIe | 128 lanes PCIe Gen5 |
| Socket configuration | 1 or 2 sockets (2P platform) |
| TDP (configurable) | up to 500 W (cTDP) |
A single EPYC 9755 socket offers 128 cores — as many as several separate servers once required. Under Proxmox's per-socket licensing this directly translates into lower support costs.
Dual-socket (2P) configuration: the SP5 platform supports 2 sockets in a single server, delivering enormous compute — a combined 256 cores and 512 threads of Zen 5 on one physical node. This lets you consolidate even more virtual machines onto a single host while keeping Proxmox per-socket licensing.
Core density vs Proxmox licensing
Proxmox VE is free, and the optional support subscription is billed per CPU socket — regardless of how many cores that socket holds. This is the inverse of VMware's per-core model. The more cores we pack into one socket, the cheaper it gets per core and per virtual machine.
- 128 cores in one socket = one support subscription instead of many
- Fewer physical hosts = fewer sockets to cover with support
- No density penalty — unlike VMware's per-core model (min. 16 cores/socket)
- Simpler cluster: 3 dense nodes are enough for HA quorum and redundancy
Example: an environment that needed 10–16 sockets on old hardware often fits into 4–6 sockets (2–3 dual-socket nodes) after consolidating onto the EPYC 9755 — while keeping headroom for growth.
Performance for general applications
"General application workloads" means a mix: databases, app servers, web, containers, ERP/file systems. This profile benefits most not from a single high clock, but from parallelism, memory bandwidth and fast I/O — and the EPYC 9755 is exceptionally balanced here.
Memory bandwidth
12 channels of DDR5-6000 deliver huge bandwidth, critical with many parallel VMs — eliminating the memory bottleneck typical of dense nodes.
PCIe Gen5 + NVMe
128 PCIe Gen5 lanes drive fast NVMe drives and 100GbE/RoCE cards for shared ZFS storage — low-latency live migration and near-local I/O.
Zen 5 and AVX-512
Higher IPC and a full-width AVX-512 path lift single-thread and compute performance — benefiting both databases and modern analytics workloads.
With a high core count and a NUMA-aware Proxmox/KVM scheduler, workloads of different profiles can be safely mixed on one node, isolated via vCPU pinning and NUMA topology.
Database nuance: for per-core-licensed databases (Oracle Database EE, MS SQL Server Enterprise) the 128/256-core density works against you — every active core is another license. We steer those workloads to the high-frequency EPYC 9175F (16 cores, up to 5.0 GHz) to minimise licensed cores. Open-source databases with no per-core fees — PostgreSQL, MySQL/MariaDB — however thrive on the EPYC 9755's density and bandwidth, isolated on dedicated VMs.
Energy efficiency & TCO
Although a single EPYC 9755 reaches 500 W TDP, what matters is power consumption per virtual machine. One dense node replacing several older servers genuinely lowers total power draw, cooling demand and rack occupancy.
- Fewer physical nodes → lower total data-centre power draw
- Lower cooling costs and fewer rack U units
- Lower OPEX (power, maintenance) and per-socket support costs
- CAPEX savings across the hardware refresh cycle (every ~3 years)
Calculate the exact OPEX and CAPEX savings for your environment in our physical-server migration calculator — including consolidation and post-migration power draw.
When the EPYC 9755 is overkill
To be fair: not every environment needs a flagship 128-core CPU. We propose the EPYC 9755 where consolidation and growth headroom genuinely pay off. For smaller deployments we pick lower-core EPYC models — keeping the same architectural and per-socket licensing advantages.
- Very small environments (a few VMs) — a cheaper, lower-core EPYC is enough
- Single, strictly per-core-licensed apps (e.g. per-core database fees) — need separate analysis
- Mixed, growing application environments — ideal for the dense EPYC 9755
- Consolidating many older servers — the best return on density
Our principle: we match the CPU to the workload profile and growth plan, not the other way round. The EPYC 9755 is the default proposal for general application workloads precisely because it combines density, bandwidth and favourable Proxmox licensing.
Calculate consolidation savings
See how much you'll save by consolidating physical servers onto dense Proxmox nodes — or book a free hardware-selection session for your workloads.
🖥️ Physical-server calculator → TCO Analysis