Computing Power Package

Pay As You Go

Experience the power of DataCanvas Alaya NeW, expertly managing and segmenting massive GPU computing resources into smaller, bite-sized units - "Computing Power Packages".

Each package is measured in “DCU", ensuring you only pay for what you use. Our innovative pre-charging and real-time deduction model makes it easy to access reliable, cost-effective, and inclusive computing resources, eliminates waste and maximizes ROI.

Start Free Trial

Product Advantages Pain Points Application Scenarios Start Free Trial

Product Advantages

Charging Mode

Pay-by-compute-consumption model, where costs are only incurred when compute tasks are running, resulting in zero idle computing resource

Flexibility

Flexible GPU resource allocation, on-demand GPU resource scaling

Toolset

Out-of-the-box toolkit, providing a complete development toolchain for LLMs and Agents

Pain Points

Limited Budgets

Skyrocketing costs for intelligent computing servers, equipment, and maintenance fees have led to a desperate need for pay-by-usage models.
Severe Resource Waste

Throughout the lifecycle of LLMs, computing resource demands and usage are often intermittent. However, traditional bare-metal leasing models that charge by the month or year can result in significant waste of computing resources.
Inflexible Resources

Computing resource demands fluctuate greatly across LLMs’ lifecycle. Users urgently need a flexible way to scale up or down computing resources in response to changing stage-specific needs.
High Technical Barriers

Users require an elastic and easy-to-use cluster environment, as well as a ready-to-use AI LLM training and fine-tuning toolchain on top of it. This can lower the technical threshold for configuring, managing, and maintaining intelligent computing software and hardware infrastructure.

Application Scenarios

LoRA Model Fine-tuning

Fine-tuning a 13B-parameter open-source pre-trained model on 3B tokens was completed in under 17 hours using a 2xH-series GPU setup

Industry Model Fine-tuning

Fine-tuning an industry-scale 33B-parameter open-source pre-trained model on 4B tokens was completed in under 2 days using an 8xH-series GPU setup

Industry Model Fine-tuning

Fine-tuning an industry-scale 175B-parameter open-source pre-trained model on 7B tokens was completed in under 2 days using an 64xH-series GPU setup

Model Training

Training a 7B-parameter model from scratch on 1.8 trillion tokens was completed in approximately 10 days using a 128xH-series GPU setup

Start Free Trial

DataCanvas Computing Power Package，precision-split massive computing power into bite-sized chunks, perfectly suited to each user's unique requirements.

Our comprehensive large model toolchain empowers you to harness the full potential of computing, ushering in a new era of inclusive and accessible computing power for all.

免费试用