GPU cloud knowledge base

We pulled every offer from our own marketplace feed across 8 providers. Picking the right provider saves less than you think. Picking the right host on the same provider saves more.

The MI300X has 192GB of HBM3, more memory and bandwidth than an H100 or H200, and published hourly rates from $1.71 to $12 per GPU depending on where you rent it. Verified prices, the sparsity trap in every spec comparison, and an honest read on ROCm.

Across 80 live offers the same RTX 5090 ranges from $0.158 to $2.004 per hour. What drives the spread, what the 32GB and FP4 support actually buy you over a 4090, and the CUDA version that trips people up on arrival.

RunPod has three storage types and they fail in three different ways. Exactly what survives a stop, what survives a terminate, what each costs per GB, and the constraints the pricing page doesn't mention.

ComfyUI custom nodes reinstall every session on a fresh cloud GPU. Why it happens, what ComfyUI-Manager's snapshot really saves, and what actually fixes it.

Five real ways to stop paying for a cloud GPU when idle, each mechanism, and the specific catch that keeps your meter running.

Terminate a cloud GPU box and the disk is deleted, with no backup to restore from. Why it happens, why a volume is not a backup, and how to stop losing work.

Moving a GPU workload to another cloud provider is easy to start and hard to finish. What actually has to come with the box, and how to move it clean.

Idle time is 30 to 95 percent of most cloud GPU bills. Here is the actual per-hour and per-month math by GPU class, and where the waste hides.

A checkpoint saves your model, not the machine that trained it. When you have to move to a different provider mid-run because the one you were on ran out of capacity, here is the difference between saving a model and saving the whole training environment.

When your provider runs out of capacity and you have to rent from someone else, a fresh box makes you re-resolve torch, CUDA, and every wheel from scratch. Here is why requirements.txt won't save you, and what actually does.

A GPU volume is region-locked and bills while idle, which is exactly what keeps you stuck on one provider. Why storage can't follow you to a cheaper or more available one, what moving it costs, and what portable state actually requires.

When your provider runs out of H100s and you have to rent from someone else, a volume snapshot brings back your files, not your working box: venv, custom nodes, models, OS state. Here is the difference, and what actually restores your setup.

A reclaimed spot GPU instance wipes the whole environment, not just your data. What gets lost, why the usual checkpoint-and-restart advice falls short, and why recovery is something you switch on before the reclaim — never after it.

A preempted spot GPU can vanish with seconds of notice, and nothing recovers it afterwards. Why network volumes and disk snapshots only bring back your data, and what has to be capturing your box beforehand to make the next one a restore instead of a rebuild.

Stopping a pod does not stop the bill, and suspend expires. What idle GPU time really costs, and how to pause a box and resume the exact setup anywhere.

Spot GPUs are cheap until a reclaim wipes your run. Why hand-rolled checkpoint scripts fail, and what has to already be capturing your box for an interruption to cost you one interval instead of the whole environment.

Committing to one GPU provider means rebuilding everything the day you leave. Here is how to build a persistent environment that isn't locked to a single provider's region, so switching is a restore, not a rebuild.

A RunPod network volume keeps billing while the GPU is off and can't leave RunPod's datacenter, which is what locks you in even after a cheaper or more available box shows up elsewhere. Six alternatives, ranked by idle cost, portability, and setup time.

Moving ComfyUI between GPU providers is not a storage problem, it is a version problem. Here is what actually breaks on the new box, and how portability works.

Keep your ComfyUI models between cloud GPU sessions. The workarounds people try, where each breaks, and what actually persists your library across providers.

Cloud GPUs are stateless. Your ComfyUI setup isn't. Here's why a persistent cloud GPU for ComfyUI matters, and how to stop rebuilding your studio on every box.

Six RunComfy alternatives for running ComfyUI on cloud GPUs, priced and compared: managed clouds, raw GPU pods, buying your own card, and portable studios.

A real training run moved across three GPU boxes: process data on an A100, snapshot, resume training on a 5090, then pick it up again on another VM.

A step-by-step walkthrough of running FLUX diffusion inference faster on a multi-GPU cloud VM with xfuser (xDiT): CUDA setup, install, and parallel inference.

Vertex AI, SageMaker, Azure Foundry, Hugging Face and the GPU-native clouds compared on supply, governance and cost — and which one fits your workload.

Idle compute, not hardware, drives your GPU bill. Make training interruptible, checkpoint properly and migrate when prices shift — the levers behind a 40% cut.

A100 vs H100 vs H200 on VRAM, throughput and cost per completed run — which one actually fits your training or inference job, and when upgrading pays off.

Shadeform shows you which GPUs are available; Aquanode runs and moves the workload. Where the two differ on deployment, monitoring and workload mobility.

H100 rates swing from under $2 to over $8 an hour. Three practices that keep you at the low end: cross-provider search, idle-free training, checkpointing.

Real hourly rates for H100, A100, 4090 and more on Aquanode vs RunPod, plus the throughput and cold-start costs a price-per-hour table always hides.