# Tips for renting GPUs. > Keep your storage near where you pack, rent in quiet hours, pick from a list of GPU models instead of waiting for one card, and don't restore on a GPU a generation newer than the one you packed on. Canonical page: https://renest.ai/docs/practical-guide ## Put your storage near where you pack Uploads slow down with distance; downloads hardly do. So keep the drive or bucket near the machines you usually pack on, and don't worry much about where you restore. Restores pull large files over several connections at once, which makes distance matter even less. ## Rent in quiet hours, and keep a backup cloud Cheap cards sell out at busy times and the same GPU can cost several times more. Pack and run test restores that aren't urgent in quiet hours. If no cheap card is free, try another cloud instead of waiting: a nest isn't tied to one provider, and the pre-flight check tells you whether the new machine can run it. ## Pick from a list of GPU models Getting a machine at all is often the hard part. Keep a short list of acceptable GPU models and a price cap rather than waiting for one exact card. ## Want the exact same image? Use the same GPU model and CPU brand Every file comes back checked byte for byte. Images also depend on the hardware: the same GPU model with the same CPU brand (Intel or AMD) gives the same image back. A different CPU brand or GPU model gives small differences — the same ones every time, not random noise. For pixel-exact work, go back to the GPU model you packed on; to simply keep working, switching cards is fine. ## Don't restore on a newer GPU generation than you packed on A GPU that is too new can't run the older PyTorch the nest was packed with: the files check out, and rendering fails. The pre-flight check in `renest restore` refuses that pairing before anything downloads, and you can check ahead of time by pointing `renest doctor` at the nest's manifest. Picking by hand, avoid the newest flagships. The measurements behind these tips are in the [field reports](/blog).