Cheapest GPU node to train for a week.
Prices checked
2x NVIDIA A6000 at Lambda does this job for $366 per week, the cheapest of 70 machines that can, checked .
Cheapest 10
all 162 rentals- 1
2x NVIDIA A6000
Own machines
Lambda · 48 GB each$366per week - 2
2A6000.20V · 2x RTX A6000 48GB
Own machines
Verda (DataCrunch) · $139 on spot$279per week - 3
2RTX6000ADA.20V · 2x RTX 6000 Ada 48GB
Own machinesVerda (DataCrunch) · $252 on spot$504per week - 4
4A6000.40V · 4x RTX A6000 48GB
Own machinesVerda (DataCrunch) · $278 on spot$556per week - 5
vcg-l40s-32c-375g-96vram
Own machines
Vultr · 2x L40S · 48 GB each$561per week - 6
2L40S.40V · 2x L40S 48GB
Own machinesVerda (DataCrunch) · $319 on spot$638per week - 7
2x NVIDIA A100 PCIe
Own machinesLambda · 40 GB each$669per week - 8
4x NVIDIA A6000
Own machinesLambda · 48 GB each$732per week - 9
2A100.44V · 2x A100 SXM4 80GB
Own machinesVerda (DataCrunch) · $378 on spot$756per week - 10
2RTXPRO6000.60V · 2x RTX PRO 6000 96GB
Own machinesVerda (DataCrunch) · $444 on spot$889per week
Other ways to pick
What matters
- More than one GPU
- Training anything worth a week of compute is split across GPUs. One card is a fine-tune, not a training run, which is why this list starts at two.
- 40 GB a board or more
- Below that the batch size collapses and the run spends its week moving data instead of learning. It is the floor the training boards of every generation clear.
- What links the GPUs
- NVLink and NVSwitch inside a node, InfiniBand between them. Where a provider states it, the row says so; where it does not, nothing here guesses.
- Checkpoint if you take the spot price
- An interruptible machine can be taken back mid-run. The saving is real and so is the restart, so the two columns are shown side by side rather than blended into one number.
How we count
A week on a multi-GPU node is the unit a real training run is bought in, and the gap between the on-demand price and the interruptible one is the whole decision. We count only machines with two GPUs or more, each with 40 GB or more, and 70 of the 162 rentals we read qualify. The cheapest of them is $366 per week at 2x NVIDIA A6000, which rents for $1.09 per GPU-hour. They are ranked by what the work costs rather than by the hourly rate, because a faster card at twice the price can still be the cheaper way to buy the job.
A week is 168 hours of the whole instance at its published hourly rate, with no commitment discount, because a committed rate is not the standard recurring price this index carries. The interruptible column is the same 168 hours at the spot rate, which assumes the run is never preempted; a real run on spot costs somewhere between the two. Each price is that work at the machine's on-demand rate: 2x NVIDIA A6000 rents for $2.18 an hour, and $366 per week is what the work costs on it.
- Own machines
- The provider owns or leases the machines.
Frequently asked
- What is the cheapest GPU to train for a week?
- 2x NVIDIA A6000 at Lambda, $366 per week. That is 2x RTX A6000 with 48 GB of memory each, renting for $1.09 per GPU-hour. Read on 1 Nov 2026.
- How is the cost worked out?
- A week is 168 hours of the whole instance at its published hourly rate, with no commitment discount, because a committed rate is not the standard recurring price this index carries. The interruptible column is the same 168 hours at the spot rate, which assumes the run is never preempted; a real run on spot costs somewhere between the two.
- Why do only 70 machines qualify?
- The job needs two GPUs or more, each with 40 GB or more. The other 92 rentals we read are still in the index; they just cannot do this job as described. Serverless rates are left out entirely, because they buy a container that runs when it is called rather than a machine you hold.
- Can I pay less on spot?
- 47 of the 70 qualifying machines publish an interruptible rate, and each of those rows shows what the same work would cost on it if the run were never preempted. It can be, so treat the spot figure as a floor rather than a quote.
- How current are these prices?
- Prices are read every 3 days from 10 providers' own public pages and keyless APIs, and this ranking recomputes with them. Last refresh: 1 Nov 2026.