About Flopper.io

Flopper.io is the searchable database for AI GPU performance. We collect, normalize, and present specifications from NVIDIA, AMD, and Intel accelerators — everything from FLOPS at every precision (FP64, FP32, FP16, BF16, FP8, TF32) to power efficiency, memory bandwidth, and system-level configurations like DGX and HGX.

Our goal is simple: make it easy to understand and compare GPU performance for AI training and inference without digging through dozens of PDFs and marketing slides.

Why We Built It

  • 🚀
    The landscape is confusing. Specs are scattered across datasheets, press releases, and benchmarks.
  • Performance means more than FLOPS. Power, efficiency, and memory bandwidth matter just as much.
  • 📊
    Comparisons should be simple. Engineers and researchers shouldn't waste hours collecting numbers.

Flopper exists to bring clarity to the AI GPU world, just like instances.vantage.sh did for cloud VMs.

What You'll Find Here

GPU Specs

FLOPS, TDP, memory, bandwidth, efficiency.

GPU Systems

DGX, HGX, OAM, and multi-GPU setups.

Comparisons

Side-by-side performance across vendors and generations.

Educational Content

Blog posts explaining FLOPS, precision formats, perf/W, and more.

Who This Is For

Flopper.io is built for:

🧑‍🔬 AI Researchers building training infrastructure.
👩‍💻 ML Engineers optimizing AI training and inference.
💼 Procurement teams evaluating cost, power, and performance trade-offs.
📈 Analysts & journalists writing about AI hardware.

Understanding GPU Metrics

What is FLOPS?

Floating-point operations per second. This measures the computational throughput of a GPU for different numerical precision formats. Higher FLOPS = more calculations per second.

Precision Formats Explained

FP64: Double precision (rarely used in AI)
FP32: Standard precision for AI model training
FP16/BF16: Half precision for faster mixed precision training
FP8: Ultra-efficient format for inference and large models

Memory Bandwidth

Measured in GB/s, this indicates how fast data can move between the GPU's compute cores and memory. Critical for training large language models and serving high-throughput inference.

TDP & Power Efficiency

Thermal Design Power (TDP) in watts indicates power consumption. Compare FLOPS/W to find the most efficient GPU for your AI infrastructure and datacenter power budget.

Our Approach

We rely on:

  • 🔍 Official datasheets from NVIDIA, AMD, and Intel.
  • 📑 Benchmark sources like MLPerf and LINPACK.
  • ⚙️ Calculated aggregates for system-level numbers (e.g., 8× H200 = 32 PFLOPS).
We're optimists: for FLOPS we use the most ideal numbers given from the vendors themselves (including sparsity acceleration and peak theoretical performance). Your mileage may vary.

Stay in the Loop

We're constantly updating the database as new GPUs are announced.

Loading verification...

Get in Touch

Got feedback, spotted an error, or want to work with us?