ABS & Phison - Bigger Data, Smaller Machine

AI Training PC

Optimized for LLM training up to 13 billion parameters and faster inference.

Tech Specs:

NVIDIA RTX 5070 TI
Intel Core Ultra 7
Phison aiDAPTIV+ AI Software
Phison 320GB aiDAPTIVCache
64GB RAM
2TB SSD Storage
Ubuntu Linux OS

aiONE Workstation

Ultimate desktop machine for local AI fine-tuning up to 100 billion parameters and accelerated inference.

Tech Specs:

4x NVIDIA RTX 6000 Ada
Intel Core Xeon W5
Phison aiDAPTIV+ AI Software
Phison 2TB aiDAPTIVCache
512GB RAM
2TB SSD primary storage
4x 2TB SSD additional storage
Ubuntu Linux OS

What is Phison aiDAPTIV+?

Trains 10x larger models than GPU-only systems, Supports RAG + Fine-tuning, & Compatible with PyTorch and NVIDIA NeMo

Extends GPU Memory

Enables larger LLMs on local hardware

Affordable Flash Memory as Cache

Fits your budget

Private. On-Premises AI.

You keep control over your data

Watch & Learn

Already Have an NVIDIA GPU-based Desktop?

Upgrade for More AI Processing Power.

AI100E aiDAPTIVCache SSDs Contain:

aiDAPTIVCache Memory
aiDAPTIVLink Memory Management Middleware
aiDAPTIVPro Suite all-in-one AI toolkit

320GB M.2

1TB M.2

2TB M.2

1TB U.2

2TB U.2

Fits Your
Budget

Simple to Use and Deploy

Keeps Data in Your Control

Affordable

Ofﬂoads expensive HBM and GDDR memory to cost-effective ﬂash memory. Eliminates the need for large numbers of high-cost and power-hungry GPU cards.

Ease of Use

Easily deploys in your home, office, classroom or data center with a small footprint while using commonplace power & cooling.

Offers command line access or intuitive GUI with all-in-one toolset for model ingest, ﬁne-tuning, and validation and inference.

Control

Enables LLM training behind your ﬁrewall. Give you full control over your private data and peace of mind with data sovereignty compliance.

Teach Yourself LLM Training with an AI Training PC

Provides a cost-effective AI Training PC for individuals and organizations to learn how to ﬁne-tune LLMs beyond just simple inference. Fills shortage of skilled talent to train LLMs locally with your own data.

Download Solution Brief

LLM Training Use Cases

LLM training on-premises enables organizations and individuals to enhance general knowledge models with domain-specific data. This provides better usability, relevance and accuracy for a wide range of specialized fields such as medical diagnostic, financial forecasting, legal analysis and product development.

Download Solution Brief

Phison aiDAPTIV+ LLM Training Integrated Solution

Use a Command Line or leverage the intuitive All-in-One aiDAPTIVPro Suite to perform LLM Training

Supported Models

Llama, Llama-2, Llama-3, CodeLlama
Vicuna, Falcon, Whisper, Clip Large
Metaformer, Resnet, Deit base, Mistral, TAIDE
And many more being continually added

and/or

Built-in Memory Management Solution

Experience seamless PyTorch compliance that eliminates the need to modify your AI application. You can effortlessly add nodes as needed. System vendors have access to AI100E SSD, middleware library licenses, and full Phison support to facilitate smooth system integration.

aiDAPTIV+ BENEFITS

Transparent drop-in
No need to change your AI Application
Reuse existing HW or add nodes

aiDAPTIV+ MIDDLEWARE

Slice model, assign to each GPU
Hold pending slices on aiDAPTIVCache
Swap pending slices w/ finished slices on GPU

FOR SYSTEM INTEGRATORS

Access to ai100E SSD
Middleware library license
Full Phison support to bring up

and

Seamless Integration with GPU Memory

The optimized middleware extends GPU memory by an additional 320GB (for PCs) up to 8TB (for workstations and servers) using aiDAPTIVCache. This added memory is used to support LLM training with low latency. Furthermore, the high endurance feature offers an industry-leading 100 DWPD, utilizing a specialized SSD design with an advanced NAND correction algorithm.

SEAMLESS INTEGRATION

Optimized middleware to extends GPU memory capacity
2x 2TB aiDAPTIVCache to support 70B model
Low latency

HIGH ENDURANCE

Industry-leading 100 DWPD with 5-year warranty
SLC NAND with advanced NAND correction algorithm

Improves Inference

aiDAPTIV+ enhances the inferencing experience by accelerating Time To First Token recall for faster responses. Furthermore, it extends the token length which provides greater context for lengthier and more accurate answers.

Unlocks Training of Larger Model Sizes

No longer limit your model size ﬁne-tuning due to the HBM or GDDR memory capacity on your GPU card. aiDAPTIV+ expands the memory footprint by intelligently incorporating ﬂash memory and DRAM into a larger memory pool.

This enables larger training models, giving you the opportunity to affordably run workloads previously reserved for the largest corporations and cloud service providers.

BIGGER DATA, SMALLER MACHINE

Supercharge Your AI Training. At Home, In Office, & The Lab.

With GPU Memory Extension by Phison aiDAPTIV+, You Unlock Advanced AI model Performance.

Designed for Real World AI Creators!