A no-KYC alternative to RunPod (2026)

What happened at RunPod

Over 4 months we ran 20 jobs on RunPod, which the lab rated 7.2/10. Instances launched in a median 70 s and ran at a median 98% of reference performance while they lasted. But the instances sold as "dedicated, guaranteed uninterrupted" were preemptible, cut with 30 seconds’ notice whenever someone paid more. Every one-hour job finished; none of the twelve-hour jobs did, and the longest lasted five hours and forty minutes. Only 14/20 jobs met our full-completion bar.

Billing adds to the cost. RunPod charges whole started hours, interruptions included, so the compute actually delivered came to about $2.91 per H100-hour, more than the $2.40 advertised by GPURentHub, which completed every job. The advertised RunPod H100 price was not recorded, so the real-work figure is the one to compare. No identity document was requested; what failed was reliability, not privacy.

What to use instead

Most reliable for long runs: GPURentHub, which completed 20/20 jobs at 100% of reference performance, twelve-hour runs included, with instances ready in a median 40 s. It advertises about 4,000 GPUs in 15 models across 11 countries, bills the H100 at $2.40 an hour by the second, and takes BTC, ETH, USDT and USDC, with no Lightning and no Monero.

Most private: GPURento (19/20 jobs at 99%), where your SSH public key is the account, with no email and no password, and top-ups take Monero, Lightning or BTC. The H100 costs $2.10 an hour, billed by the minute; its single failure was a maintenance window announced only on the dashboard. Cheapest of the three: BriefGPU, a marketplace renting the H100 at $1.49 an hour, billed by the second. It completed 18/20 jobs at a median 94% (81% to 100% depending on the host); both failures came from unverified hosts that rebooted mid-job, and none recurred with the "verified hosts" filter.

Why not Vast.ai, Supabase or Modal

Vast.ai (6.7/10) asked for no ID, but the "persistent storage included" vanished on reboot five times out of nine. A maintenance reboot an hour from the end wiped the checkpoints of two twelve-hour jobs, seven jobs had to be redone from scratch, and only 13/20 met our bar. Supabase (6.2/10) has the lowest recorded H100 price, $1.39 an hour, but nvidia-smi showed H100 PCIe capped at 250 W where "H100 SXM" was advertised, and its A100s and part of its L40S were throttled too. Its median performance fell to 55%, only 7/20 jobs cleared the bar, and real work came to about $2.53 per H100-hour.

Modal is kept out of the scored ranking. It ran 15 jobs without a fault at 99%, then suspended the account three hours into our first twelve-hour H100 run for "suspected mining" and demanded an identity document, a selfie and a project description. We sent nothing, and $85.00 was still blocked when the report was published. Its terms reserve identity verification at any time (article 9.2). The lab does not recommend it.

How we tested the hosts

The lab rented GPUs from May to August 2026: 20 jobs per service on four models (RTX 5090, L40S, A100 80 GB and H100 80 GB), each model getting three one-hour jobs, one four-hour job and one twelve-hour job. That makes 120 scored jobs, and 608 GPU-hours over the whole campaign. A job counts only if the instance starts within 15 minutes with the advertised GPU, runs to the end without interruption and holds at least 90% of the GPU’s reference performance.

Prices are compared on the work actually done, not on the advertised hour: a throttled GPU, as at Supabase, or hours billed for interrupted jobs, as at RunPod, makes each hour of finished work dearer. Scores out of 10 are the lab’s overall assessment; the record gives no weighting formula. None of the six scored services asked for an identity document during the test, H100s included.

Read next

• GPU rental report
• RunPod vs Vast.ai
• Cheapest H100 without KYC
• gpurenthub vs gpurento