Building or upgrading a workstation desktop for serious work?

It depends on workload. A solid CAD workstation runs ₹1.2–2.5 Lakh, a serious video-editing build ₹2.5–5 Lakh, a 24GB-VRAM AI build ₹3.5–7 Lakh, a 48GB-VRAM ML build ₹7–12 Lakh. Refurbished workstations bring entry points down 40-60%. We’ll quote both pre-built and custom paths after understanding the workload.

Pre-built wins on warranty, ISV certification (SolidWorks / AutoCAD validate against specific OEM models), and corporate procurement. Custom wins on price-per-spec by 25-40% and lets you pick exactly the cooling / case / quietness you want. If the build needs to pass a corporate IT review — pre-built. If it’s your own workstation — custom usually wins.

Yes for inference, sometimes for fine-tuning. A refurbished HP Z840 with dual Xeon E5-2697 v4 and an added RTX 3090 24GB runs Stable Diffusion, Llama-3-8B inference, and PyTorch training on small datasets fine. Limits: no PCIe Gen4, slower NVMe, more power draw. Honest tradeoff: 40-60% cheaper, 10-20% slower for most ML workloads.

Both. Hyderabad gets full doorstep service — ₹149 visit, on-site diagnostics, walk-in at our MG Road Secunderabad store. India-wide customers WhatsApp us for spec consults; for repairs you courier the unit to Secunderabad, we diagnose, quote, and return-ship after repair.

30 days on labour and assembly from us. Each component (CPU, GPU, RAM, SSD, PSU, motherboard) keeps its manufacturer warranty — typically 3 years on consumer parts, 5 years on workstation-grade parts. We honour cross-shipment for warranty claims with most major brands.

VRAM is the deciding number. 24GB (4090 / 5090) handles most fine-tuning of 7-13B models with QLoRA, full inference of 7-30B models, and Stable Diffusion / Flux. 48GB (A6000) lets you fine-tune larger 13-30B models without paging or train smaller models from scratch. Above 48GB — A100 / H100 / multi-GPU rigs, real budget territory.

Depends on age and fault. If the motherboard is dead on a 6+ year-old Xeon E5 workstation, replacing the platform usually makes sense. If just the SMPS or a fan failed, 80% of the time repair is cheaper. We always quote both paths so the decision is yours.