Hardware reality: not every workload needs the highest single-core boost. Virtualization density, database transactions, game ticks and CI builds each bind to a different bottleneck — core count, memory bandwidth, IPC or clock. These are our real nodes; every spec is from the Intel or AMD datasheet.
1. Intel Xeon Platinum 8269CY — Mumbai · best single-thread
A 26-core / 52-thread Cascade Lake-SP part, custom-built for a hyperscaler (never on Intel's public price list). 2.5GHz base, 3.8GHz single-core turbo — the fastest per-core clock in the fleet — with 35.75MB L3, 6-channel DDR4-2666 and Speed Select technology. The node pairs it with 251GB RAM and SSD-RAID pools.
Best for: Minecraft/FiveM ticks, Redis, PHP request latency — anything bound to one fast core rather than many slow ones.
2. AMD EPYC 7C13 — Delhi N3 · density & bandwidth
A Zen 3 "Milan" cloud SKU (2021): 64 cores / 128 threads, 2.0GHz base / 3.675GHz boost, 256MB L3, eight DDR4-3200 channels (~205 GB/s theoretical) and 128 PCIe Gen4 lanes. The newest architecture in the rack — strongest IPC and memory bandwidth by a wide margin, at 225W.
Best for: databases (PostgreSQL/MySQL), compilation, JVM workloads, many-VM density, and memory-bandwidth-bound jobs. VMs land on its NVMe pool.
3. Intel Xeon Gold 6138 — Delhi N1 · the AVX-512 workhorse
Skylake-SP (Q3 2017): 20 cores / 40 threads, 2.0/3.7GHz, 27.5MB L3, 6-channel DDR4-2666, 125W. Its distinguishing feature is two AVX-512 FMA units per core — workloads compiled for AVX-512 (transcoding, scientific compute, some databases) run genuinely faster here than the spec table suggests. 125GB node RAM, dedicated NVMe provisioning pool.
Best for: media transcode, compression, vector-heavy compute, general web hosting.
4. Intel Xeon E5-2699 v4 — Delhi N2, Noida · RAM capacity
Broadwell-EP (2016): 22 cores / 44 threads, 2.2/3.6GHz, 55MB L3, 4-channel DDR4-2400, 145W. The oldest silicon in the fleet — and paired with 504GB of node RAM, the largest in the rack, on a 1.9TB SSD-RAID pool. Thread capacity and memory headroom are its value, not clock speed.
Best for: big memory workloads, many small services consolidated, backup/staging nodes, and anything where "more RAM per rupee" wins.
5. Choosing honestly
- Game servers & remote desktops: favor clock — Mumbai's 8269CY or the EPYC's Zen 3 cores.
- Databases & JVM: favor memory bandwidth — the EPYC's 8 channels are the real feature.
- Windows RDP desktops: any node works; pick by your city's latency to Mumbai vs NCR.
- Bulk RAM, many services: the Noida E5 node's 504GB is there for exactly this.
Node CPUs are printed live from Proxmox on the Mumbai and Delhi location pages — if the fleet changes, the pages change with it. Inside any VM, lscpu tells you which silicon you're actually on.