Skip to content

What are the best Denvr Dataworks alternatives for training?

Denvr Dataworks alternatives for H100 training: Nodus, Together AI, Runpod, Lambda and CoreWeave priced on the same 72-hour, 8-GPU job.

For cheaper NVIDIA H100 training than Denvr Dataworks, Nodus is the lowest on-demand alternative we priced, listed at $2.60 per GPU-hour versus Denvr’s $3.00, ahead of Together AI GPU Clusters, Runpod and Lambda. If you still hold Denvr credits, you may not need to leave at all, because Nodus can run single-node jobs inside your own Denvr account (Beta). Stay with Denvr for Canadian data residency, Intel Gaudi 2, a long reservation you will keep busy, or clusters larger than 64 GPUs.

We build Nodus, so read this with that in mind. Every competitor number links to that company’s own page, and every total is written out so you can redo it with the rates you see today.

What does Denvr Dataworks offer for training today?

Section titled “What does Denvr Dataworks offer for training today?”

Denvr sells GPU machines by the hour, on demand or reserved. Its pricing page lists an 8-GPU NVIDIA H100 SXM node at $3.00 per GPU-hour with 3,200 Gbps InfiniBand. On-demand instances are billed by the minute, and reserved contracts save up to 20% on multi-month or annual terms. The same page lists the A100 SXM 80 GB at $2.00, the GH200 at $3.87, Intel Gaudi 2 at $2.00, and the H200 SXM as reserved only.

Its homepage describes data centers in Canada and the USA, clusters of up to 1,024 GPUs, managed inference endpoints and shared NVMe storage. It also runs a startup program that starts with a $1,000 credit and goes up to $500,000 for eligible teams.

That is a solid offer for steady cluster work. The reasons people look elsewhere are usually the on-demand H100 rate, a wish to stop managing machines, or a job that must keep going when a machine disappears.

How do the alternatives compare at a glance?

Section titled “How do the alternatives compare at a glance?”

The table lines up each provider’s published H100 terms. Rows show what each page states; a blank means the page we checked does not say.

Denvr Dataworks Nodus Together AI GPU Clusters Runpod Lambda CoreWeave
H100 on-demand $3.00/GPU-hr (SXM) $2.60/GPU-hr (SXM) $3.99/GPU-hr $4.39/GPU-hr Secure Cloud, from $2.69 Community $4.59/GPU-hr on the 8-GPU instance $49.24/hr per 8-GPU instance
Cheaper tiers Reserved, up to 20% off Reserved $3.69 (3-30 days), $3.45 (31-90 days); preemptible $1.99 1-Click Clusters from 16 GPUs, $5.85 at 64 GPUs Spot $19.71/hr per 8-GPU instance
Billing unit Per minute Prepaid, with a spend cap per job Per second
Largest training setup Up to 1,024 GPUs Up to 8 nodes (Beta) 16 to 2,000+ GPUs
Recovery when a machine is lost Restores the job’s saved state on new capacity

Prices checked November 20, 2026: Denvr pricing, Denvr homepage, Nodus pricing, Nodus multi-node docs, Nodus checkpoint docs, Together AI pricing, Runpod pricing, Lambda pricing and CoreWeave pricing. CoreWeave’s figures are for its 8-GPU NVIDIA HGX H100 instance.

What does the same 72-hour, 8 x H100 run cost on each?

Section titled “What does the same 72-hour, 8 x H100 run cost on each?”

We priced one job everywhere: a fine-tune or continued-pretraining run on one 8 x H100 machine for 72 hours. That is 8 x 72 = 576 GPU-hours. Each total is 576 times the per-GPU rate, except CoreWeave, which prices the whole 8-GPU instance, so its total is 72 x $49.24.

Provider H100 rate 72 hours on 8 GPUs
Nodus $2.60/GPU-hr $1,497.60
Denvr Dataworks on-demand $3.00/GPU-hr $1,728.00
Together AI GPU Clusters on-demand $3.99/GPU-hr $2,298.24
Runpod Secure Cloud $4.39/GPU-hr $2,528.64
Lambda 8 x H100 instance $4.59/GPU-hr $2,643.84
CoreWeave on-demand $49.24/instance-hr $3,545.28

Same sources and check date as the first table. GPU compute only: storage and data transfer are not included for anyone.

Cost of 576 H100 GPU-hours (8 GPUs for 72 hours) at on-demand list prices on November 20, 2026: Nodus $1,497.60, Denvr Dataworks $1,728.00, Together AI $2,298.24, Runpod $2,528.64, Lambda $2,643.84 and CoreWeave $3,545.28. At list prices the 72-hour run costs $230.40 less on Nodus than on Denvr, about 13% less.

Two cheaper tiers sit outside the chart because they are not on-demand. Together AI lists preemptible H100 at $1.99, which would make the job $1,146.24, and CoreWeave lists spot at $19.71 per instance-hour, which would make it $1,419.12. Both can beat Nodus on paper if your training code saves state often and you accept those providers’ terms for interruptible capacity.

Is a Denvr reservation cheaper than paying on demand?

Section titled “Is a Denvr reservation cheaper than paying on demand?”

A reservation wins only if you keep the machine busy almost all the time. Denvr says reserved contracts save up to 20%, so the best case for the H100 is $3.00 x 0.8 = $2.40 per GPU-hour. That is below the $2.60 Nodus listing, and it is the one place in this comparison where Denvr’s own price beats ours.

The catch is that a reservation bills every hour of the term. Take one 8 x H100 node for a 730-hour month. At the best-case discount the reservation costs 8 x $2.40 x 730 = $14,016 whether you use it or not.

Nodus on demand costs 8 x $2.60 = $20.80 for each hour you actually run. It stays cheaper until you use $14,016 / $20.80 = 673.8 hours, about 92% of the month. Denvr’s own on-demand rate of $24.00 per node-hour crosses at 584 hours, which is 80%.

Hours used in the month Nodus on demand Denvr on demand Denvr reserved, best case
200 $4,160 $4,800 $14,016
400 $8,320 $9,600 $14,016
600 $12,480 $14,400 $14,016
730 $15,184 $17,520 $14,016

Rates from the Denvr pricing page and Nodus pricing page, checked November 20, 2026. The 20% is Denvr’s stated maximum; your quoted discount may be smaller, which moves the crossover later.

Monthly cost of one 8 x H100 node by hours used, November 20, 2026 list prices: Nodus on demand $4,160 at 200 hours, $8,320 at 400 and $15,184 at 730; Denvr on demand $4,800, $9,600 and $17,520; Denvr reserved at the best-case 20% discount $14,016 at every level. A best-case Denvr reservation only pays off past about 674 busy hours a month, roughly 92% utilization.

Our decision rule from this math: reserve on Denvr if one node will run above about 92% of every month for the whole term. Below that, pay on demand. For the wider version of this tradeoff, see reserved vs on-demand GPUs.

Can you keep spending Denvr credits while you move to Nodus?

Section titled “Can you keep spending Denvr credits while you move to Nodus?”

Yes, for single-node work. The Nodus cloud-accounts docs list Denvr among the providers you can connect, first to view your existing instances and then, as a separate Beta step, to run work in your own account. Denvr asks for an API key, the cluster your credits or quota are in (Msc1 or Hou1) and optionally a VPC.

Once you choose “Run work in my account” and set a spending limit, Nodus places your single-node Jobs in that account first. Denvr bills the machine and disk against your credits, and Nodus adds a route fee of $0.02 per device-hour. On the 72-hour job that is 576 x $0.02 = $11.52. Sandboxes and multi-node runs still use Nodus capacity.

This matters most if you joined Denvr’s startup program. You can burn down the credit balance with the same job definitions you will use afterwards, instead of keeping two workflows. Check what you have connected with the command from the docs:

Terminal window
nodus get cloudaccounts

For other ways to stretch credits, see how startups get GPU credits.

What happens to a multi-day run when a machine fails?

Section titled “What happens to a multi-day run when a machine fails?”

On Nodus, the job comes back with its saved state. Your script writes weights, optimizer state and the step counter under /nodus/state. When the capacity under a job is reclaimed, the next attempt starts with that directory restored, as the checkpoint guide describes. The script must load what it saved, because a restored attempt starts your command from the beginning.

On a machine you hold yourself, at Denvr or any provider in the table, recovery is your job: notice the failure, start a new machine and relaunch from your last checkpoint. That is fine when you watch the run. It is costly over a weekend, because the idle hours before you notice are still billed.

Where is Denvr Dataworks the better choice?

Section titled “Where is Denvr Dataworks the better choice?”

Denvr is the better pick in several specific cases, and we would tell you to stay in them.

  • You need data to stay in Canada. Denvr runs sovereign Canadian infrastructure and supports Canadian organizations qualified for the ISED Canada AI Compute Access Fund. If residency is a hard requirement, that alone is a reason to stay.
  • You train across more than 64 GPUs. Denvr describes clusters up to 1,024 GPUs on 3,200 Gbps InfiniBand. Nodus multi-node training is in Beta with a ceiling of 8 nodes.
  • You will keep a reserved node above roughly 92% busy. At the best-case discount Denvr’s $2.40 per GPU-hour beats our $2.60 listing.
  • You want Intel Gaudi 2 or NVIDIA GH200. Denvr lists both; the Nodus pricing page lists neither.
  • You want one vendor for the whole stack, including managed inference endpoints and shared NVMe storage next to the cluster.

If none of these apply and you mostly run single-node jobs of hours or days, the on-demand math points elsewhere.

How do you move a Denvr training job to Nodus?

Section titled “How do you move a Denvr training job to Nodus?”

Keep your training script and change how it is launched. Install the CLI and sign in, as the install docs show:

Terminal window
pip install nodus-compute
nodus login

Then submit the run. This is the eight-GPU example from the nodus run reference, detached so it keeps going after you close the terminal. Raise --max-cost from the example’s $40 to your budget; the 72-hour job above is $1,497.60 at list price.

Terminal window
nodus run --gpu H100:8 --max-cost 40 -d -- torchrun train.py

Point your checkpoint path at /nodus/state so a restored attempt can pick up where it stopped. If you are coming from another provider too, Runpod alternatives and SF Compute alternatives cover the same ground for those platforms.

Is Denvr Dataworks cheaper than Lambda for H100s?

Section titled “Is Denvr Dataworks cheaper than Lambda for H100s?”

At on-demand list prices, yes. Denvr lists the H100 SXM at $3.00 per GPU-hour and Lambda lists its 8-GPU H100 instance at $4.59, so the 72-hour job is $1,728.00 against $2,643.84.

What is a cheaper alternative to Denvr for A100 training?

Section titled “What is a cheaper alternative to Denvr for A100 training?”

Nodus lists the A100 80 GB at $1.09 per GPU-hour against Denvr’s $2.00 for the A100 SXM 80 GB. For 576 GPU-hours that is $627.84 on Nodus and $1,152.00 on Denvr.

Does Nodus charge extra when it runs jobs in my Denvr account?

Section titled “Does Nodus charge extra when it runs jobs in my Denvr account?”

Nodus adds a route fee of $0.02 per device-hour for GPU Jobs in your own account, and Denvr bills the machine and disk as usual. Your account spending limit uses Nodus’s estimate of cloud costs, which can differ from Denvr’s actual invoice.

Can I run multi-node training in my own Denvr account through Nodus?

Section titled “Can I run multi-node training in my own Denvr account through Nodus?”

Not today. Running in your own account covers Jobs, Workspaces and other single-node work; multi-node runs use Nodus capacity, up to 8 nodes in Beta.

No. Denvr’s pricing page says on-demand instances are billed by the minute. Runpod bills by the second, and Nodus enforces a dollar cap you set per job.

More from the blog