Maingear, a maker of high-performance custom PCs, launched its aiDAPTIV+ package for its AI-enabled Pro RS and Pro WS workstations.
The company announced a new aiDAPTIV+ package, co-developed with Phison Electronics, a maker of NAND flash controllers and storage solutions, for its Pro RS and Pro WS workstations.
The aiDAPTIV+ add-on enables full-parameter fine-tuning and large-model inference on mainstream GPUs, helping teams move faster while keeping data private and on-prem.
Live demos of a Maingear workstation equipped with aiDAPTIV+ will be available at Phison’s booth during the Nvidia GTC event in Washington, D.C.
AI teams need on-premise training and inference performance without the unpredictability of cloud costs or the exposure of sensitive data. The aiDAPTIV+ package combines Maingear’s enterprise-ready workstations with Phison’s aiDAPTIV+ intelligent SSD caching to expand effective VRAM, enabling larger models and longer contexts at the edge, with predictable costs and IT-friendly deployment.
“On-prem AI shouldn’t require a research lab,” said Wallace Santos, CEO of Maingear, in a statement. “With aiDAPTIV+ on our Pro workstations, teams can fine-tune real models, keep data private, and move from idea to insight faster, on hardware they can actually buy, service, and scale.”
What capabilities does aiDAPTIV+ add:
● Full-parameter training & fine-tuning on-prem: Iterate rapidly on real workloads while keeping IP and sensitive datasets in-house.
● Bigger models on mainstream GPUs: aiDAPTIV+ extends effective VRAM, unlocking larger models and contexts without exotic hardware.
● Predictable TCO & easy scaling: Avoid surprise cloud bills; scale incrementally with familiar, serviceable components.
● Broad platform compatibility: Offered on MAINGEAR Pro RS and Pro WS workstation PCs, and compatible with Intel Core, Intel Xeon, AMD Ryzen, and AMD Threadripper CPU options.
● Simple to deploy: Delivered as a pre-validated MAINGEAR package with room to grow memory, storage, and GPU capacity over time.
● User management: Pro Suite supports multiple users over a local intranet with role-based access to support concurrent users.
Performance and AI capability
For representative workloads, systems support full-parameter fine-tuning on up to 16B-class models and scale to larger model classes. Inference remains responsive for 7B–16B class models locally, with 70B (AWQ4-quantized) suitable for batch or background inference. By extending effective VRAM with high-performance storage, aiDAPTIV+ helps smooth burst performance during multi-stage pipelines. The package is designed for multi-session development and QA environments, with achievable concurrency
depending on selected GPU configurations and precision.
“aiDAPTIV+ helps unlock bigger models without the need for as many GPUs used in traditional training, enabling access to high-parameter training for a wider audience,” said Michael Wu, president of Phison US, in a statement. “Together with Maingear, we’re giving developers a practical path to train and deploy at the edge, combining performance, predictability, and data control in a single workstation footprint.”