Skip to content
GPUCalc
v0.5 · redesign
QUICK ESTIMATE · ~30 SECONDS

Size your LLM deployment.

Pick a model, tell us your load. The estimate fills in on the right — tap any tile to flip and see the math.

01

What model are you serving?

✓ SET

Sliding gallery of common models, or paste any HuggingFace model ID.

Google
Gemma 2 2B
2B
Google
Gemma 2 9B
9B
✓
Google
Gemma 2 27B
27B
Google
Gemma 3 12B
12BNEW
Google
Gemma 3 27B
27BNEW
Google
Gemma 4 2B
2BNEW
Google
Gemma 4 9B
9BNEW
Google
Gemma 4 27B
27BNEW
Meta
Llama 3 8B
8B
Meta
Llama 3 70B
70B
Meta
Llama 3.1 8B
8B
Meta
Llama 3.1 70B
70B
Mistral
Mistral 7B
7B
Mistral
Mixtral 8x7B
47B
Qwen
Qwen 2.5 7B
7B
RedHat
RedHat Model 7B
7B
NVIDIA
Nemotron 340B
340B
✓Selected: Gemma 2 27B · 27B params
OR PASTE A HUGGINGFACE MODEL ID
Browse on HF →
02

Load profile

✓ SET

Tap the pencil to adjust either value.

CONCURRENT USERS
10
custom
CONVERSATION LENGTH
8K
short
DEPLOYMENT TYPE
Ready · estimate is live on the right
LIVE ESTIMATEtap a tile to see the math
ESTIMATED MONTHLYON-PREM
$3,672
4 × H100 · $1.26/hr · 730 hrs/mo
↺ SEE MATH
MONTHLY COST MATH
4 × $1.26/gpu-hr × 730 hrs
= $4k / month
On-prem pricing · amortized 36 months + electricity
↺ FLIP BACK
GPU COUNT
4× H100
42% utilized
↺ SEE MATH
GPU COUNT MATH
[134 ÷ 80GB]
= 4 GPUs
↺ FLIP BACK
TOTAL VRAM
134GB
model + KV cache
↺ SEE MATH
VRAM MATH
Weights: 54.0 GB
KV: 69.5 GB
= 134.3 GB
↺ FLIP BACK
MEMORY LAYOUT4 × H100 80GB
WeightsKV cacheFree
↺ SEE MATH
PER-GPU SPLIT
13.5 GB weights
17.4 GB KV cache
46.4 GB free headroom
Each user adds ~1.7 GB of KV cache.
↺ FLIP BACK
GPU