TextDense1.70B params8K max context

SmolLM2 1.7B Instruct VRAM Requirements: How Many GPUs & What It Costs

SmolLM2 1.7B Instruct needs 3.4 GB of VRAM for its weights at FP16 — 4.2 GB once a KV cache is loaded. Below is the full breakdown: which GPUs fit, how many you need, what it serves, and what it costs per million tokens. Every figure comes from tokcalc's open-source capacity model.

Weights (FP16)
3.4 GB
4.2 GB with KV cache
Fits on 1 GPU?
Yes
A10G 24GB or larger
Cheapest config
$0.0187
2× RTX 3080 10GB / 1M tokens
Fastest config
45882
tok/s on 1× Cerebras CS-3 (WSE-3)
VRAM requirement by quantization
Weight footprint = params × bytes per param. The “with KV cache” column adds one request at 4K context.
QuantizationWeightsWith KV cacheSmallest single GPU
FP16recommended3.4 GB4.2 GBRTX 3080 10GB (10 GB)
FP8 (E4M3)1.7 GB2.5 GBRTX 3080 10GB (10 GB)
AWQ 4-bit0.9 GB1.7 GBRTX 3080 10GB (10 GB)
Which GPU should you use?
Reference workload: 8K context, 8 concurrent users, continuous batching, FP16. Ranked by fitting first, then by tokens per dollar-hour.
GPUGPUsTotal VRAMDecode tok/sAggregate tok/sCost / 1M tokens
RTX 3080 10GB2×20 GB247.05928$0.0187
RTX 4080 16GB2×32 GB233.05593$0.0298
RTX 3090 24GB1×24 GB178.92147$0.0323
RTX 5090 32GB1×32 GB342.64111$0.0338
T4 16GB2×32 GB104.02496$0.0445
RTX A4000 16GB2×32 GB145.63494$0.0477
RTX 4090 24GB1×24 GB192.72312$0.0480
AMD MI300X 192GB1×192 GB101312159$0.0571
V100 SXM2 32GB1×32 GB172.12065$0.0673
A100 40GB SXM41×40 GB296.33556$0.0703
AMD MI325X 288GB1×288 GB114713765$0.0706
A100 80GB SXM41×80 GB382.44588$0.0726

RTX 3080 10GB has no NVLink, so 2-way tensor parallelism runs over PCIe. tokcalc applies a flat 0.85× multi-GPU efficiency factor for this, which is realistic for NVLink-equipped datacenter cards but optimistic for PCIe scale-out — treat the throughput and cost figures for this configuration as an upper bound, and measure before committing.

How concurrency changes with context length
On 2× RTX 3080 10GB at FP16. KV cache scales linearly with context, so long contexts cut the number of users you can serve from the same GPUs.
ContextConcurrent usersKV per requestFits?
4K20824.6 MB✓ fits
8K101.61 GB✓ fits
33K26.44 GB✓ fits
66K112.88 GB✓ fits
131K025.77 GB✗ does not fit
Frequently asked questions

Can SmolLM2 1.7B Instruct run on a single GPU?

Yes. At FP16 the model needs 4.2 GB of VRAM, so it fits on a single A10G 24GB or anything with more VRAM. At full FP16 it needs 4.2 GB, which fits on one RTX 3080 10GB or larger.

How many GPUs does SmolLM2 1.7B Instruct need?

At FP16 with 8,192-token context and 8 concurrent users, the cheapest configuration that fits is 2× RTX 3080 10GB (20 GB total VRAM). $0.0187 per 1M output tokens.

How fast can SmolLM2 1.7B Instruct be served?

On 1× Cerebras CS-3 (WSE-3) at FP16, the aggregate throughput is 45882 tokens/sec with continuous batching. That hardware has no published rental price, so its cost per million tokens is not comparable.

How much VRAM does SmolLM2 1.7B Instruct need?

Weights alone: 3.4 GB at FP16, 3.4 GB at FP16. Add KV cache for your context length and concurrency — every concurrent request at 8,192 tokens costs extra VRAM, so budget headroom above the weight figure.

Is it cheaper to self-host SmolLM2 1.7B Instruct than to call an API?

At 2× RTX 3080 10GB the marginal cost is $0.0187 per 1M output tokens, before you add idle capacity, ops staff, or egress. That is typically cheaper than frontier API pricing at sustained volume, but API calls win when your load is bursty or too small to keep a GPU busy. The Build vs Buy tab on tokcalc prices this against a specific provider for you.

Try SmolLM2 1.7B Instruct with your own context and concurrency

The calculator uses the exact same formulas as this page — change the workload and watch every number update.

Open in calculator

SmolLM2 1.7B Instruct on specific GPUs

SmolLM2 1.7B Instruct on RTX 3090 24GBSmolLM2 1.7B Instruct on RTX 5090 32GBSmolLM2 1.7B Instruct on RTX 4090 24GBSmolLM2 1.7B Instruct on AMD MI300X 192GBSmolLM2 1.7B Instruct on V100 SXM2 32GBSmolLM2 1.7B Instruct on A100 40GB SXM4SmolLM2 1.7B Instruct on AMD MI325X 288GBSmolLM2 1.7B Instruct on A100 80GB SXM4
Model architecture
The numbers behind the KV-cache and throughput math.
Total parameters1.70B
Active parameters per token1.70B
ArchitectureDense
Attention layoutFull attention (GQA)
Transformer layers24
Hidden dimension2048
Attention heads32 query / 32 KV (GQA)
Head dimension64
Vocabulary size128,256
Max context (catalog)8K
KV cache per token @ 8K192.0 KB

Every layer caches keys and values for the whole context, so the KV cache grows linearly with conversation length.