suilo.xyz

What It Is

Suilo.xyz positions itself as a decentralized compute network, with its main selling point being the ability to run open weights models on unused GPU capacity at lower prices. The site directly compares itself with AWS, Azure, Runpod, and Lambda, claiming support for deploying models such as Llama 3, Mixtral, Stable Diffusion XL, Qwen, and Falcon on H100, A100, and RTX 4090 GPUs. It is aimed at training, fine-tuning, and inference workloads.

Core Capabilities and Use Cases

Based on the available information, Suilo looks more like a GPU rental/deployment platform than a packaged AI SaaS product. The RTX 4090 24GB is listed as suitable for SDXL, Llama-3-8B, and testing; the A100 80GB is positioned for fine-tuning 70B models; and the H100 80GB is intended for larger-scale foundation model training clusters. The site mentions hardware features such as bandwidth, NVLink, Infiniband, and FP8 Transformer Engine, but does not explain capabilities around image management, containers, APIs, storage, monitoring, autoscaling, or job orchestration.

Pricing and Value for Money

Pricing is the biggest highlight: RTX 4090 is listed at $0.29/hr, A100 at $0.99/hr, and H100 at $1.99/hr. In the example shown, using an H100 for 100 hours would cost $199, compared with $4.50/hr on AWS/Azure and $2.89/hr on Runpod/Lambda. If real-world availability and performance are stable, the value proposition is very strong. However, the page notes that the comparison is based on average spot pricing availability, and access is being opened in batches. Actual supply, queue times, failure rates, and long-term pricing still need to be verified.

Pros, Cons, and Limitations

The advantages are clear pricing, GPU models that cover mainstream training and inference needs, and an explicit focus on open-weight models. The downside is limited disclosure: there is no clear information about the company entity, regions, payment methods, data privacy, isolation and security, SLA, customer support, or refund policy. There is also no mention of a Chinese interface or accessibility from China. As a decentralized network, node stability, data security, and performance consistency will be key concerns for enterprise users.

Who It Is For and Access from China

Suilo is best suited to developers, research teams, and AI startups with engineering capabilities who can deploy models and manage environments themselves, and who are highly sensitive to GPU costs. It is less suitable for enterprises that want a ready-to-use experience, compliance audits, or strong SLAs. Access from China is currently unknown. If network access or payment is restricted, alternatives to compare include Runpod, Lambda, AWS/Azure, or China-based options such as Alibaba Cloud, Tencent Cloud, AutoDL, and DCloud.