Experiments | Super GPU Cluster & Marketplace

FREE TIER ACTIVE Verify your email to claim one free GPU experiment per registration IP.

Want a jump start on the Open Kernel Challenge leaderboard? Clone our open-source Auto-Research Agent on GitHub pre-loaded with Muse challenge prompts and CUDA best practices.

curl -s https://app.getsupers.com/experiments/cli.py | python3 - submit my_kernel.py

Track Your Experiment in the Queue

Check real-time queue position and estimated completion time:

📈 Platform Growth & Experiments Over Time

Live hourly tracking of registered users and GPU workload executions

Total Users

Total Jobs

Add Your GPU to the Marketplace

Monetize idle GPU capacity. Earn 85% revenue share for every experiment execution second (only 15% platform cut).

How Host Payouts Work

Industry-lowest 15% platform cut & 85% host payout.

85% Host Revenue Share (15% Cut)

When a user runs a job on your node at 2 credits/sec ($0.002/s), you earn 1.7 credits ($0.0017/sec = $6.12/hr). Platform fee is reduced to just 15% to cover network gateway routing.

Stripe Connect & 1-Click Credit Swap

Payouts settle directly to your bank account via Stripe Connect, or swap instantly into platform compute credits at zero fees!

Active Marketplace Nodes

Node Name	GPU Model	VRAM	Rate / Sec	Status	Served Seconds	Earned Credits
Loading registered marketplace GPUs...

Prepaid Experiment Credits

Pay-as-you-go micro-pricing. 1 Credit = $0.001 (0.1¢). Each experiment only burns credits for the exact execution seconds it uses.

H200 SXM Compute

5 Credits / sec

$0.005 / sec ($0.30 / min)

Marketplace Community GPUs

2 Credits / sec

$0.002 / sec ($0.12 / min)

Typical 10s Kernel Test

50 Credits

≈ 5 Cents per run!

MICRO DEV PACK

5,000 Credits

~100 Experiments

1,000 H200 execution seconds
2,500 Community GPU seconds
~50 to 100 kernel test runs
Unlocks Give-to-Get Referral Rewards

STARTER PACK

$10

10,000 Credits

~200 Experiments

2,000 H200 execution seconds
5,000 Community GPU seconds
~100 to 200 kernel test runs
Full REST API & CLI access

RESEARCHER PACK

$20

22,000 Credits

+10% Bonus (~440 Experiments)

4,400 H200 execution seconds
11,000 Community GPU seconds
Systematic parameter sweeps

PRO FLEET PACK

$50

60,000 Credits

+20% Bonus (~1,200 Experiments)

12,000 H200 execution seconds
30,000 Community GPU seconds
Multi-worker autonomous runs

CLUSTER PIONEER

$100

130,000 Credits

+30% Bonus (~2,600 Experiments)

26,000 H200 execution seconds
65,000 Community GPU seconds
Highest priority execution scheduler

🎁 Give 100 Experiments, Get 100 Experiments

Share your invite link with other AI researchers. When your friend tops up just $5 to activate their developer account, you BOTH immediately unlock 5,000 bonus credits (~100 free GPU experiments) on NVIDIA H200 silicon!

Active Referrals: 0 • Earned: 0 Free Experiments

Cost Comparison: Super Experiments vs Renting an H200

Typical Cloud Rental (RunPod, Lambda, AWS)

• Cost: $3.50 – $4.50 / hour ($84 – $108 / day)
• Billed continuously while idle (90%+ of developer time is idle reading/debugging)
• Minimum 1-hour or 15-minute billing increments
• Setup friction: Driver config, CUDA compilation, environment downloads
• 20 runs in an afternoon: ~$15.20

Super Experiments API (Microsecond Billing)

• Rate: $0.005 / second (only pay for exact execution time!)
• Average experiment (12–15s): $0.06 – $0.075 per run
• One free experiment after email verification, limited to one trial per IP
• Zero setup: Pre-warmed Hopper H200 with CUDA 13.0, Triton 3.8, SGLang
• 20 runs in an afternoon: ~$1.35 (~91% cheaper!)

Your API Credentials

User ID: usr_guest

API Key: sk_exp_...

Free Experiments Left: 1

Include in requests: Authorization: Bearer <API_KEY> or pass ?api_key=...

Auto-Research Agent: Jump Start the Open Kernel Challenge

Clone the open-source autonomous research agent harness pre-configured for the Open Kernel Challenge (OKC). Hook it up directly to Super Experiments to run 24/7 hillclimbing sweeps on H200 GPUs.

View on GitHub ↗

git clone https://github.com/rohanarun/Auto-research-agent.git
cd Auto-research-agent
# Run autonomous research iterations against the Super Experiments H200 queue:
export SUPER_EXPERIMENTS_KEY=sk_exp_YOUR_KEY
python3 -m auto_research --target muse_decode_layer

Codex / Claude / OpenCode CLI Integration

1-Liner: Run directly without installing

Paste this into any Codex, Claude, or OpenCode session:

curl -s https://app.getsupers.com/experiments/cli.py | python3 - submit my_kernel.py --wait

Login / Register Email

python3 <(curl -fsS https://app.getsupers.com/experiments/cli.py) login you@domain.com

Using python3 <(curl ...) in Bash or Zsh preserves keyboard input when prompted for your verification code.

Check Queue Position of Job

curl -s https://app.getsupers.com/experiments/cli.py | python3 - status JOB_ID

Check Balance & Free Runs

curl -s https://app.getsupers.com/experiments/cli.py | python3 - balance

Opt-In Owner-Only Artifact & Profiler Export

By default, worker scratch and temporary files are wiped immediately upon job completion so no subsequent job can access your code or outputs. Pass --export to securely copy source scripts and profiler artifacts (e.g. profiles/, *.txt, *.json, *.csv) into isolated, private owner-only storage outside the worker's reach before teardown:

curl -s https://app.getsupers.com/experiments/cli.py | python3 - submit my_kernel.py --export --wait

Download or View Exported Artifacts (Owner Only)

View file listings and text previews of your exported artifacts, or download the full .tar.gz archive. Only the authenticated owner who ran the experiment has access (all non-owners are strictly rejected with 403 Forbidden):

# View artifact list & previews via CLI:
curl -s https://app.getsupers.com/experiments/cli.py | python3 - export JOB_ID

# Direct download of .tar.gz bundle via curl:
curl -H "Authorization: Bearer YOUR_API_KEY" "https://app.getsupers.com/experiments/api/export?job=JOB_ID&download=true" -o export_JOB_ID.tar.gz