Crusoe NVIDIA H200 HGX vs Modal Nvidia H200 SXM.
machine vs machine · prices as of
The same H200 SXM at two providers, so the board is identical and everything else is not. Crusoe NVIDIA H200 HGX takes 2 of 4 rounds against Modal Nvidia H200 SXM. It is also the cheaper of the two by the hour, at $4.29/GPU-hr against $4.54/GPU-hr.
NVIDIA H200 HGX
$4.29/GPU-hr
$4.29/hr for the machine · $721 a week
1x H200 SXM · 141 GB each · $0.0304/GB-hr
Nvidia H200 SXM
$4.54/GPU-hr
$4.54/hr for the machine · $763 a week
1x H200 SXM · 141 GB each · $0.0322/GB-hr · Serverless
Price per GPU-hour (lower wins)
$4.29/GPU-hr
$4.54/GPU-hr
Interruptible rate (lower wins)
not published
not published
Price per GB of VRAM (lower wins)
$0.0304/GB-hr
$0.0322/GB-hr
VRAM per GPU
141 GB
141 GB
GPUs per instance
1x H200 SXM
1x H200 SXM
vCPU
not stated
not stated
RAM
not stated
not stated
Included storage
none · volumes extra
none · volumes extra
Regions listed
regions not stated
regions not stated
What a run costs on each
- An hour of the machine$4.29/hr vs $4.54/hr
- A day$103 vs $109
- A week$721 vs $763
- A week on spotnot published vs not published
Crusoe then Modal, at the published rate with no commitment discount
Go further
Add a third or fourth machine, or switch the weighting, in the tool.
Frequently asked
- Which is cheaper, Crusoe NVIDIA H200 HGX or Modal Nvidia H200 SXM?
- Crusoe NVIDIA H200 HGX, at $4.29/GPU-hr against $4.54/GPU-hr, both the on-demand rate for one GPU for one hour, read on 17 Sept 2026. On the whole machine that is $4.29/hr against $4.54/hr.
- Which holds more GPU memory?
- Both add up to 141 GB across the instance, at 141 GB and 141 GB a board.
- What would a week cost?
- $721 for Crusoe NVIDIA H200 HGX and $763 for Modal Nvidia H200 SXM, at 168 hours of the whole machine at its published rate, with no commitment discount. Neither publishes an interruptible rate.
- What does the hourly price leave out?
- Crusoe NVIDIA H200 HGX: disk none, billed separately, no regions published, billed per hour. Modal Nvidia H200 SXM: disk none, billed separately, no regions published, billed per second. Egress and tax are extra on both.
Every rate is the on-demand price for one GPU for one hour, read 17 Sept 2026 from each provider's own page, so a node of eight is compared at what one of its GPUs costs. Storage, egress and tax are extra unless a row says otherwise, and reserved or committed rates are not indexed. Bars are scaled to the larger value in each row.