SyncBio AI Factory · Infrastructure

Sovereign, on-premise AI compute — built and running in Puducherry.

A Slurm-orchestrated HPC cluster, GPU-ready inference nodes, and a local model stack that runs open-source and enterprise models on demand — CPU and GPU resources provisioned as workloads need them.

8
Compute nodes
86
Physical cores
172
Parallel threads
960 GB
Cluster ECC memory
RTX-ready
4090 / 5090 GPU
20+
Analyst workstations
01 · Hardware

Technical Specifications

A dedicated DDR4 ECC HPC cluster, GPU inference workstation, an enterprise-API inference node, and a fleet of engineering workstations — all owned and operated locally.

HPC Compute Cluster — 7 nodes, DDR4 ECC
78 CORES · 156 THREADS · 960 GB ECC
Processor
Qty
Cores / Threads
Role
Xeon E5-2680 v4
14C / 28T
Heavy compute node slurmd
Xeon Silver 4210
10C / 20T
Specialised compute node
Xeon E5-2650 v3
10C / 20T each
Symmetrical compute workers
Xeon E5-2680 v3
12C / 24T each
Symmetrical compute workers
Cluster Controller
Dedicated Core i7 · 6 Cores / 12 Threads. Runs the Slurm controller — scheduling, accounting & job dispatch across the cluster.
1× · slurmctld
AI Production Workstation
Core i9-9900K · 128 GB RAM. Primary AI inference & production node — RTX 4090 / 5090 GPU-ready.
1× · GPU Partition
Enterprise Inference Node
Dedicated Mac Mini · M2 Pro · 64 GB, reserved exclusively for enterprise model APIs — OpenAI, Anthropic, Mistral, Moonshot, DeepSeek, NVIDIA & Hugging Face. Available on demand.
1× · Inference Gateway
Engineering Workstations
20+ local workstations on Core i5 / i7 processors, supporting a substantial in-house team of analysts and developers.
20+ · Access Layer
02 · Architecture

Cluster Topology & Slurm Orchestration

Every job flows through the Slurm workload manager, which schedules across CPU partitions and allocates GPU resources on demand via generic resource (GRES) scheduling.

Access Layer
Analyst & Developer Workstations
20+ · Core i5 / i7 · submit jobs & sessions
AI Production Workstation
i9-9900K · 128 GB · RTX 4090/5090
Head / Controller Node
Dedicated Core i7 · 6 Cores / 12 Threads
slurmctld · scheduler + accounting
SLURM WORKLOAD MANAGER
Job queue · CPU partitions · GPU (GRES) allocation · fair-share scheduling · on-demand provisioning
Compute Partition — slurmd workers
Xeon E5-2680 v4
14C / 28T · heavy compute
Xeon Silver 4210
10C / 20T · specialised
Xeon E5-2650 v3 ×3
10C / 20T each · workers
Xeon E5-2680 v3 ×2
12C / 24T each · workers
Model Serving & Inference
GPU Inference
RTX 4090 / 5090 · serving & fine-tuning
Local Runtimes
Ollama · Goose · Odysseus
Enterprise API Gateway
Mac Mini M2 Pro · 64 GB · OpenAI · Claude
03 · Elasticity

Provision CPU & GPU On Demand

Resources are allocated per workload and released back to the pool when done — no idle capacity, no lock-in.

Deploy
Spin up environments
Containerised jobs and services launched on any node via Slurm batch or interactive sessions.
Allocate
Right-size the request
Request exactly the cores, memory and GPUs a task needs; fair-share scheduling keeps the cluster balanced.
Provision
GPU when it matters
GRES scheduling hands GPU nodes to inference and training jobs on demand, then reclaims them automatically.
04 · Models

AI Models & GPT Support

Open-source models run locally on the cluster; enterprise models are reached through the dedicated inference gateway — one stack, both worlds.

Open-Source · run locally
Served on-premise via Ollama, Goose, Odysseus & Hugging Face — no data leaves the cluster.
Llama Mistral / Mixtral DeepSeek Qwen Gemma Phi Kimi (Moonshot) Hugging Face Hub
Enterprise · via gateway
Proprietary frontier models accessed on demand through the dedicated Mac Mini inference node.
OpenAI GPT Anthropic Claude Mistral AI Moonshot AI DeepSeek NVIDIA NIM ElevenLabs
Supported third-party inference layers
NVIDIA
AI hardware, GPU architecture & software ecosystem
OpenAI
Foundation models, AI APIs, consumer AI
Anthropic
Foundation models & AI safety research
Mistral AI
Open-weight & proprietary large language models
Moonshot AI
Foundation models, AI APIs, consumer AI
Hugging Face
Open-source model hub & ML collaboration
DeepSeek
Open & proprietary reasoning models
ElevenLabs
Voice AI, text-to-speech & voice agents
05 · Platform

Supported Technology Stack

The same battle-tested bioinformatics and MLOps stack SyncBio runs in production — now backed by dedicated local compute.

Orchestration & HPC
Slurm MPI Nextflow Snakemake CWL WDL
Containers & DevOps
Docker Kubernetes Singularity CI/CD
ML / AI Frameworks
PyTorch TensorFlow Keras Scikit-learn vLLM
Local Model Runtimes
Ollama Goose Odysseus llama.cpp
Languages
Python R Java JavaScript Bash Perl
Data & Cloud
PostgreSQL MongoDB Neo4j Redis AWS · GCP · Azure

Ready-to-deploy sovereign AI compute.

The infrastructure is already built, running and growing in Puducherry — available to partners, government and research today.

GET IN TOUCH
www.syncbio.in
info@syncbio.in
+91 63699 75526
Atal Incubation Centre, Pondicherry
Engineering College Campus, Puducherry 605 014