Node families

Every plan you can choose when creating a node belongs to a family: E1, E1WC, C3, M3, SDC3, SQL, GDC, and GDC3. Each family is a different mix of CPU, memory, storage, and, for the GDC families, graphics hardware. Pick the family that matches your workload, the job or software the node will run.

Node families side by side

Family	Best for	What makes it different
E1	Development, testing, and small services where cost and flexible storage matter most	Root storage is configured on its own, from 75 GB up, and no default public IPv4
E1WC	The same jobs as E1, on Windows	The Windows edition of E1
C3	CPU-intensive work and production services that need steady uptime: web servers, batch processing, analytics	Disk size and IOPS come bundled into the plan; growing storage means moving to a bigger plan
M3	Memory-heavy work: in-memory databases, caching, real-time big data processing	Built for RAM headroom rather than raw CPU
SDC3	Production workloads that need dedicated, predictable performance	Shown as Linux Smart Dedicated Compute or Windows Smart Dedicated Compute; Chennai offers only the Linux version
GDC	Graphics-intensive and parallel-processing work, or AI/ML workloads	Dedicated GPU compute with guaranteed resources
GDC3	Deep learning, LLM inference, and HPC	Next-generation dedicated GPU instances for deep learning, LLM inference, and HPC
SQL	Windows workloads that need SQL Server	Adds SQL Server to a Windows node, shown as SQLWEB or SQLSTANDARD

How to choose, by bottleneck

If your main concern is	Start with
Lowest cost for development or testing	E1, or E1WC on Windows
Steady, CPU-heavy work	C3
RAM headroom	M3
Dedicated, predictable performance	SDC3
A Windows node running SQL Server	SQL
GPU work: training, inference, or rendering	GDC, or GDC3

Then compare the exact plan: CPU count, RAM, disk size, and IOPS, plus GPU memory and the CUDA version on a GPU plan. IOPS is input/output operations per second, how fast the disk can read and write. CUDA is Nvidia’s toolkit for running compute jobs on its GPUs; the version tells you which toolkit the drivers support. HPC, high-performance computing, means using many processors or GPUs together on a demanding task like a simulation. Two plans in the same family can still differ a lot; the Select Plan screen shows the real numbers before you commit. Check the Estimated Cost before you launch; Create a node shows where it is.

E1 and E1WC, for a small test server

For a small test server or a development box, pick E1: it is the cost-effective family, built for development and testing, with the flexibility to size storage on your own terms. An E1 node keeps its root storage apart from the plan, so you can start small, from 75 GB, and grow the disk later without recreating the node. It starts with a private IP only, so you attach a reserved IP once you need one from outside. E1WC is the Windows edition of E1, for the same cost-sensitive jobs on a Windows node.

Root storage starts at a default of 150 GB, runs up to a maximum of 2,400 GB, and grows in steps of 25 GB below 150 GB and 50 GB from 150 GB up. The root disk only grows. If you end up needing less, create a new node at the size you want.

E1 root storage carries its own IOPS, separate from the plan. At the default 150 GB, a root disk gets 15,000 read and 10,000 write IOPS, and every GB above 150 GB adds 5 read and 5 write IOPS, so a larger disk is also a faster one:

Root storage	Read IOPS	Write IOPS
150 GB	15,000	10,000
200 GB	15,250	10,250
250 GB	15,500	10,500
300 GB	15,750	10,750

The portal is the source of truth for the final root storage price and IOPS it shows before you launch or upgrade.

If you know from the start that you need predictable, dedicated CPU, memory, or storage I/O, pick a bundled family like C3 or SDC3 instead.

C3, once that test becomes a production service

C3 is compute-optimized, built for steadier, CPU-intensive work like web servers, batch processing, and analytics. Its disk size and IOPS come bundled into the plan itself, so you grow storage later by moving to a bigger plan. Move to C3 once your workload needs constant uptime rather than occasional bursts. If memory is what your workload is short on, M3 fits better.

M3, for memory-heavy work

M3 is memory-optimized, for workloads where RAM is the bottleneck rather than CPU: built for in-memory databases, caching, and real-time big data processing. Pick it for a database or cache that needs headroom a CPU-optimized plan would not give you. For a mostly CPU-bound workload, use C3; M3’s extra RAM would sit unused.

SDC3, for dedicated production capacity

SDC3 gives you dedicated compute with guaranteed performance, the right choice for a production workload that needs its capacity to itself. It is shown as Linux Smart Dedicated Compute or Windows Smart Dedicated Compute, depending on the OS you pick; in Chennai, only the Linux category is offered. For short-lived development or low-cost testing, use E1.

GDC and GDC3, for GPU work

Both put a graphics card on the node for work that a CPU alone does slowly: machine training, inference, or rendering. Pick one once your job runs better on a GPU; choose from the GPU tab in the create flow.

GDC is dedicated GPU compute with guaranteed resources, for graphics-intensive and parallel-processing work as well as high-performance AI or machine-learning workloads that would otherwise share capacity. GDC3 is the newer generation, aimed at deep learning, LLM inference, and high-performance computing. The plan list shows GPU memory and the CUDA version for every plan, so you can compare cards directly before you commit. Skip both if your application only needs CPU and RAM; a GPU family is wasted spend otherwise.

SQL, for Windows nodes that need SQL Server

SQL plans add SQL Server to a Windows node, shown in the plan list as SQLWEB or SQLSTANDARD. If a normal Linux or Windows server is all you need, choose another family; SQL plans cost more because they include SQL Server.