Runs easilyfp16 · 4K ctx · batch 1

Run Gemma 2 9B on H100

At fp16 precision, Gemma 2 9B (9.0 B parameters) needs roughly 20.9 GB of VRAM. H100 has 80 GB, leaving 59.1 GB of headroom — plenty of room for longer contexts and larger batch sizes. Expected throughput ≈ 121 tokens/sec on a single card.

26%
VRAM
Runs easily
Estimated VRAM (fp16)
20.9 GB
of 80 GB on H100· 59.1 GB free
Tok/sec
121
TFTT
552ms
Power
270W
Rent on GPUniq marketplace
H100
Live prices from verified providers · billed hourly, no commitment.
Rent now

Memory breakdown (fp16)

Model Weights
18.00 GB
KV Cache
0.37 GB
Activations
1.48 GB
Framework Overhead
1.04 GB

Quantization comparison

Lower precision = less VRAM with a small quality trade-off. Quality order: FP16 > INT8 > INT4.

PrecisionVRAMUtilisationFits on H100?
FP16 (full precision)20.9 GB26%Runs easily
INT8 (8-bit)11.1 GB14%Runs easily
INT4 / Q4 (4-bit)6.2 GB8%Runs easily

Frequently asked

Can H100 run Gemma 2 9B?
Yes. Gemma 2 9B needs ≈ 20.9 GB VRAM at fp16 and H100 provides 80 GB. Expected throughput is 121 tokens/sec per GPU.
How much VRAM does Gemma 2 9B use?
About 20.9 GB at fp16, 6.2 GB at INT4 (for a 4K context, batch size 1). Longer contexts add to KV cache size; larger batches increase activations.
What's the fastest way to run Gemma 2 9B on H100?
Use a production inference engine like vLLM, SGLang, or TensorRT-LLM with paged attention — they cut KV cache 2–3× vs the naive estimate and batch multiple requests efficiently. For H100, enable Flash Attention 2.
Where can I rent a H100?
GPUniq aggregates live H100 offers from verified providers. You can deploy an instance in about a minute and pay hourly with no commitment.
Try it live

Chat with Gemma 2 9B — no setup

Send a prompt and see how Gemma 2 9B responds directly in our chat. No installation, no GPU required to test.

Open in chat
Want to tweak sequence length, batch size, or fine-tuning? Open the full calculator →