GPU Cloud
Choose your GPU rental.
Start with one DGX Spark or a linked two-node cluster. Rent by the minute, or reserve by the month for 1 to 24 months, discounted from 1 month.
Updated . Prices in USD, excluding applicable tax.
DGX Spark rentals
Choose your configuration
Choose by memory needs and runtime compatibility. Both configurations use ARM64 hosts with NVIDIA GB10; x86-only binaries need an ARM-compatible build.
Single Spark
One node with 128 GB unified memory and 1 TB NVMe storage. Start here when your evaluation fits on one machine.
Linked cluster
Two nodes, each with 128 GB memory and 1 TB storage. Choose this only when your runtime supports distributed execution.
| Configuration | Memory | Architecture | Hourly rate | Configure rental |
|---|---|---|---|---|
| 1× DGX SparkDedicated single node | 128 GB unified | ARM64 · GB10 | $0.79 / hour | Deploy now 1× DGX Spark |
| 2× DGX SparkLinked cluster | 256 GB across two nodes | ARM64 · linked GB10 | $1.79 / hour | Deploy now 2× DGX Spark |
| DGX Spark 1× (Partner)Same machine type, hosted by an independent partner | 128 GB unified | ARM64 · GB10 | $0.79 / hour | Deploy now DGX Spark 1× (Partner) |
The 2-Spark cluster requires a runtime that supports distributed execution. Its 256 GB is total memory across two nodes, not a single memory pool.
DGX Spark 1× (Partner) is the same machine type on a partner's machine, not GPUwerk's own. How the Partner tier differs
Choose a rental duration
Use one Spark or a two-Spark cluster short-term. For predictable use, reserve by the month for 1 to 24 months, discounted from 1 month.
x86 GPUs
Looking for an x86 GPU?
Every GPU below is rented first come, first served: deploy when capacity is free, or join its waitlist. The GPU is yours alone while you rent it, billed per minute like every GPUwerk tier. The labels below are read live; status.gpuwerk.com shows every tier and how long its waitlist has waited.
| GPU | GPU memory | Hourly rate | Deploy or join the waitlist |
|---|---|---|---|
| RTX 5090Check availability | 32 GB GDDR7 | $0.89 / hour | Deploy now for RTX 5090 |
| L4Check availability · Monthly terms on request for L4 | 24 GB GDDR6 | $0.49 / hour | Deploy now for L4 |
| RTX 6000 AdaCheck availability | 48 GB GDDR6 ECC | $0.75 / hour | Deploy now for RTX 6000 Ada |
| L40SCheck availability · Monthly terms on request for L40S | 48 GB GDDR6 | $0.99 / hour | Deploy now for L40S |
| RTX PRO 6000 BlackwellCheck availability | 96 GB GDDR7 ECC | $1.89 / hour | Deploy now for RTX PRO 6000 Blackwell |
| H100 SXMCheck availability · Monthly terms on request for H100 | 80 GB HBM3 | $2.69 / hour | Deploy now for H100 SXM |
| H200Check availability | 141 GB HBM3e | $3.49 / hour | Deploy now for H200 |
| B200Check availability · Monthly terms on request for B200 | 180 GB HBM3e | $5.49 / hour | Deploy now for B200 |
| B300Check availability · Monthly terms on request for B300 | 288 GB HBM3e | $6.49 / hour | Deploy now for B300 |
Whole servers
8-GPU servers
One HGX or DGX server with all eight GPUs, rented by one customer. The rate is eight times the single-GPU rate, billed per minute.
| Server | GPU memory | Hourly rate | Deploy or join the waitlist |
|---|---|---|---|
| 8× H100 serverCheck availability | 8 × 80 GB HBM3 | $21.52 / hour | Deploy now for 8× H100 server |
| 8× H200 serverCheck availability | 8 × 141 GB HBM3e | $27.92 / hour | Deploy now for 8× H200 server |
| 8× B200 serverCheck availability · Monthly terms on request for 8× B200 server | 8 × 180 GB HBM3e | $43.92 / hour | Deploy now for 8× B200 server |
| 8× B300 serverCheck availability · Monthly terms on request for 8× B300 server | 8 × 288 GB HBM3e | $51.92 / hour | Deploy now for 8× B300 server |
Certified, on term
Monthly to 5-year terms
L4, L40S, H100, B200 and B300 machines, 8-GPU servers and GPU VMs from our certified provider (ISO/IEC 27001:2022, SOC 2 Type 2) in the EU, on monthly, 1-, 2- or 5-year terms. Available on request and quoted per term.
Guides
Compare and decide.
Verdict-first comparisons for whichever tier you are weighing.
DGX Spark cloud providers compared
Every hourly Spark rental I found, with price, location and billing granularity, including GPUwerk's.
Rent or buy a DGX Spark
Hours per week, purchase price, and the break-even math, with a calculator for your own numbers.
DGX Spark vs RTX 5090
32 GB of fast VRAM vs 128 GB of capacity. Not a close fight on speed: the whole question is whether your model fits.
DGX Spark fleet benchmarks
Measured single-stream and concurrent tok/s on the models staged on every node, with raw logs.
FAQ.
What does "Join the waitlist" mean for an x86 GPU?
For the RTX 5090 it means its one machine is in use; the L40S, RTX 6000 Ada, RTX PRO 6000 and H100 are planned and not rentable yet. Joining the waitlist queues your account for that tier, and GPUwerk emails people in line one at a time, oldest first, when they can deploy. No machine is held for anyone: whoever deploys first gets it.
Is a DGX Spark shared with other customers?
No. Every GPUwerk instance, on any tier, is one dedicated node for the duration of your rental, never a slice of a shared GPU.
How is billing different between tiers?
Hourly tiers bill per minute against prepaid credit, in USD, excluding applicable tax, with no egress fees. Monthly reservations are invoiced monthly, and so is a quoted term on an 8-GPU server.
Start with the hardware that fits your model.
Every tier deploys when capacity is free; otherwise join its waitlist. First top-up: pay $20, get $30 in credit.
Deploy now