NVIDIA DGX Spark · ARM64 · GB10 · introductory offer

One DGX Spark for a week.
€99.

A whole dedicated DGX Spark for 7 days, paid once. GB10 Grace Blackwell, 128 GB unified memory, 1 TB NVMe. Sign in, pay, and the Spark starts in your console. No hourly bill during the week, running or stopped.

One pass per account. Your Spark starts as soon as a Spark is free. If none starts within 48 hours of your payment, I refund it in full. The 7 days count from when it starts.

You choose the location before you pay: Europe (EEA), or any free Spark (Europe, North America or East Asia). The console shows the machine's country.

Price excludes VAT, which Stripe adds at checkout where it applies.

374instances started by 94 developers since September 2026 (as of 9 November 2026)
303.6tok/sGemma 4 26B-A4B on one DGX Spark, 8 requests in parallel (vLLM, 4-bit, MTP). Public benchmark.

How the week works.

1. Sign in

Google, GitHub or email. That creates your account.

2. Pay once

€99 plus VAT where it applies, on a Stripe page. Business customers can add a VAT ID.

3. Your Spark starts

It appears under Instances as soon as a Spark is free: within seconds if one is free now, otherwise when the next one frees up. Open the browser terminal, or add an SSH key.

4. Seven days

The week runs from when the Spark is deployed and can't be ended early. You can stop and start it as you like. I email you a day before it ends.

After the 7 days the Spark is billed at the normal hourly rate from your credit. If you have no credit, it is stopped and its workspace kept on the machine for a short time, then deleted. Terminate it when you’re done and billing ends. Nothing is backed up, so keep your own copy of anything you need. If no Spark has started for you within 48 hours of your payment, I refund it in full.

NVIDIA DGX Spark · per node

The hardware.

One whole machine for the week. Nothing is shared, virtualized, or oversubscribed.

SuperchipGB10 Grace Blackwell
AI performance1 PFLOP (sparse FP4, NVIDIA peak figure)
Memory128 GB unified LPDDR5x, 273 GB/s
CPU20-core Arm (ARM64)
Storage1 TB NVMe
OSDGX OS 7 · Ubuntu 24.04
LocationYou choose the location before you pay: Europe (EEA), or any free Spark (Europe, North America or East Asia). The console shows the machine's country.

New platform · real company

Who runs this.

GPUwerk is new, so here is what you would want to check before paying. I'm Samuel Seidel, and I run it from Prague as PRINT IT! SE. My own Sparks are in a datacenter in Hradec Králové, Czech Republic. The others belong to partners who own a DGX Spark and host it in their own country, some of them outside the EU.

One thing to know first: a Spark is an ARM64 machine with a GB10 chip, so check that your stack runs on ARM. Questions before you pay? Write to me; I answer myself. See the Terms.

Seven days on a Spark
for €99.

Introductory offer, one per account, while passes are available.

Start my week