# TapGPU Review Hourly cloud GPU rental for AI training, inference, and generative workloads, priced in CNY with Alipay support. ## Editor's Score: 58/100 Capability scores 16 because TapGPU covers a genuinely wide GPU range from RTX 3090 to H200, but it is a pure infrastructure rental with no tooling, templates, API, or developer features beyond raw instance access, and it sits in a coding tools category where it has no relevant feature overlap with any directory competitor. Ease of use scores 15 because the self-serve console and hourly billing are simple in concept, but the non-obvious storage billing behavior (disks charge after instance stop, requiring explicit deletion) is a real usability trap, and the Chinese-only interface excludes a large portion of potential users. Value scores 18 because the raw CNY hourly rates are genuinely low, with RTX 3090 at ¥0.58/hr and A100 PCIe at ¥1.88/hr undercutting many Western providers, but the lack of a free tier, the storage billing gotcha, and the CNY-only payment wall reduce practical value for the broader audience. Delivery scores 9 because the tool is newly launched (GitHub submission November 2026), has no findable user reviews or trust signals, Community Cloud stock is explicitly described as volatile, and the storage-keeps-billing-after-stop behavior is a documented trust concern that a buyer would want resolved before committing real workloads. ## Pros - RTX 3090 Community Cloud starts at ¥0.58/hr, making it one of the cheapest entry points for GPU compute among self-serve providers. - Wide GPU range in a single platform: from consumer RTX 3090/4090/5090 up to H200 with 141 GB VRAM, covering both hobbyist and serious training workloads. - No subscription or minimum commitment; pure pay-as-you-go from a prepaid balance suits irregular or experimental workloads. ## Cons - Storage disks (system, data, and shared model disks) continue billing after an instance is stopped and require explicit deletion to fully end charges, a billing behavior not common on Western GPU clouds and a real risk of unexpected costs. - Community Cloud stock and node stability fluctuate, meaning the cheapest tier is not reliable for time-sensitive or production workloads. - The entire platform, including documentation and support, appears to be Simplified Chinese only, making it effectively inaccessible to non-Chinese-speaking developers. ## Pricing Paid. - RTX 3090 (Community Cloud): ¥0.58/hr - RTX 3090 (Secure Cloud): ¥2.58/hr - RTX 4090 (Community Cloud): ¥1.60/hr - RTX 4090 (Secure Cloud): ¥4.27/hr - RTX 5090 (Community Cloud): ¥2.16/hr - RTX 5090 (Secure Cloud): ¥6.84/hr - A40 (Community Cloud): ¥1.89/hr - A40 (Secure Cloud): ¥4.44/hr - A100 PCIe (Community Cloud): ¥1.88/hr - A100 PCIe (Secure Cloud): ¥6.08/hr - A100 SXM4 (Community Cloud): ¥2.16/hr - A100 SXM4 (Secure Cloud): ¥8.04/hr - H100 (Community Cloud): ¥9.34/hr - H100 (Secure Cloud): ¥24.18/hr - H200 (Community Cloud): ¥17.52/hr - H200 (Secure Cloud): ¥32.69/hr Pricing reflects what we saw at the time of review (2026-11). Confirm current pricing on the official site. Category: AI Coding & Dev Tools (https://aitoolseekers.com/category/ai-coding-dev-tools) Official site: https://www.tapgpu.com/ Last verified by a human: 2026-11-20 Reviewed by a human at AI Tool Seekers, a hand-reviewed AI tools directory. Every listed tool is verified by a person before publishing. Scores are editorial opinions from a fixed rubric (capability, ease of use, value, delivery on promise) and cannot be bought; there is no pay-to-rank. Directory: https://aitoolseekers.com This review: https://aitoolseekers.com/tools/tapgpu Methodology: https://aitoolseekers.com/how-verification-works ## Verdict TapGPU is the right pick for China-based developers who want cheap, self-serve GPU access billed in CNY without a subscription commitment. Its Community Cloud tier undercuts many Western providers on raw hourly cost, but stock availability fluctuates and the platform appears to serve a Chinese-language audience almost exclusively, limiting its reach. Developers who need English-language support, reserved capacity, or per-second billing should look at RunPod or Lambda Labs instead. Best for: China-based AI developers and researchers needing cheap, no-commitment hourly GPU compute. Not for: Teams needing English-language support, reserved/committed pricing, or guaranteed GPU availability. ## Overview TapGPU is a Chinese cloud GPU rental platform that sells compute by the hour, drawing down from a prepaid CNY balance. There is no subscription, no reserved tier, and no minimum spend. You pick a GPU model, choose between Community Cloud or Secure Cloud, and pay only for the time the instance runs. The GPU lineup covers many workloads: RTX 3090 (24 GB, ¥0.58/hr Community Cloud) for entry-level inference and fine-tuning; RTX 4090 (24 GB, ¥1.60/hr) as the main image-generation and LoRA training workhorse; RTX 5090 (32 GB, ¥2.16/hr) for higher-VRAM consumer work; and data-center cards including A40 (48 GB), A100 PCIe and SXM4 (both 80 GB), H100 (80/94 GB), and H200 (141 GB, up to ¥32.69/hr Secure Cloud). The two-tier Community/Secure split is the platform's defining structural choice: Community Cloud nodes are cheaper but stock and node stability fluctuate more, while Secure Cloud costs roughly two to four times more per hour but offers more predictable availability. One important billing detail sets TapGPU apart from per-second providers: stopping an instance does not necessarily stop all charges. Persistent instances may need to be explicitly deleted, and storage disks (system disk, data disk, shared model disk) can keep billing after the GPU instance itself is stopped. That is a real gotcha for anyone used to cleaner billing semantics from Western providers. The platform is built for a Chinese-market audience. The pricing page is in Simplified Chinese, Alipay is the highlighted payment method, and all prices are in CNY. There is no evidence of English-language documentation or support channels, which matters if your team is not Chinese-speaking. ## Who should use it A solo researcher in China fine-tuning a LoRA model on an RTX 4090 for ¥1.60/hr, or a small team running batch inference jobs on an A100 without wanting to commit to a monthly plan, will find TapGPU genuinely cheap and frictionless. The RTX 3090 at ¥0.58/hr Community Cloud is one of the lowest entry points for GPU compute available anywhere, making it reasonable for students or hobbyists who need occasional bursts of compute. Anyone outside China, or anyone who needs English documentation, per-second billing granularity, or guaranteed GPU availability for production workloads, should look elsewhere. GitHub Copilot is the better fit if your actual bottleneck is AI-assisted coding rather than raw compute. If you need raw GPU infrastructure with English support and a more mature ecosystem, RunPod or Lambda Labs (not currently in this directory) are the practical alternatives; within this directory, Cursor is worth considering if your GPU need is driven by AI-assisted development rather than training or inference at scale. ## FAQ Q: What is the cheapest GPU available on TapGPU? A: The RTX 3090 on Community Cloud starts at ¥0.58/hr as of the November 15, 2026 pricing snapshot. It has 24 GB VRAM and is listed as an entry-level consumer option. Q: How does billing work, and how do I stop charges? A: Charges deduct hourly from a prepaid CNY balance. Stopping an instance is not always enough: persistent instances may need to be explicitly deleted, and storage disks (system, data, shared model) can keep billing after the GPU instance stops. Delete unused disks to fully end charges. Q: What is the difference between Community Cloud and Secure Cloud? A: Community Cloud is cheaper but has more variable stock and node stability. Secure Cloud costs roughly two to four times more per hour for the same GPU model but offers more predictable availability, making it better for sustained training runs. Q: Does TapGPU support payment methods outside of China? A: Alipay is the highlighted payment method. There is no publicly documented support for international credit cards or PayPal, which effectively limits the platform to users with Chinese payment accounts. Q: Is the H200 available, and what does it cost? A: Yes. The H200 with 141 GB VRAM is listed on the pricing page. Community Cloud is ¥17.52/hr and Secure Cloud is ¥32.69/hr, making it the most expensive option on the platform. Q: Is there a free trial or free tier? A: No. TapGPU is purely pay-as-you-go with no free tier or trial credits documented on the pricing page. You fund a balance and pay per hour of use.