TextDense32B params41K max context

Qwen 3 32B VRAM Requirements: How Many GPUs & What It Costs

Qwen 3 32B needs 32.0 GB of VRAM for its weights at FP8 (E4M3) — 33.1 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 (FP8 (E4M3))
32.0 GB
33.1 GB with KV cache
Fits on 1 GPU?
Yes
Mac M3 Max (64GB) or larger
Cheapest config
$0.0294
8× RTX 3080 10GB / 1M tokens
Fastest config
24863
tok/s on 2× 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
FP1664.0 GB65.1 GBH100 SXM5 80GB (80 GB)
FP8 (E4M3)recommended32.0 GB33.1 GBA100 40GB SXM4 (40 GB)
AWQ 4-bit17.6 GB18.7 GBA10G 24GB (24 GB)
Which GPU should you use?
Reference workload: 8K context, 8 concurrent users, continuous batching, FP8 (E4M3). Ranked by fitting first, then by tokens per dollar-hour.
GPUGPUsTotal VRAMDecode tok/sAggregate tok/sCost / 1M tokens
RTX 3080 10GB8×80 GB157.515116$0.0294
RTX 3090 24GB4×96 GB96.964450$0.0624
RTX 4090 24GB4×96 GB104.45012$0.0887
RTX 4080 16GB4×64 GB74.283565$0.0935
RTX 5090 32GB2×64 GB92.822228$0.1247
A10G 24GB4×96 GB62.162984$0.1341
RTX A4000 16GB4×64 GB46.412228$0.1496
T4 16GB4×64 GB33.151013$0.2195
V100 SXM2 32GB2×64 GB46.621119$0.2483
A100 40GB SXM42×80 GB80.291927$0.2595
L4 24GB4×96 GB31.081492$0.2979
AMD MI300X 192GB1×192 GB161.51938$0.3584

RTX 3080 10GB has no NVLink, so 8-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 8× RTX 3080 10GB at FP8 (E4M3). 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?
4K441.07 GB✓ fits
8K222.15 GB✓ fits
33K58.59 GB✓ fits
66K217.18 GB✓ fits
131K134.36 GB✓ fits
Frequently asked questions

Can Qwen 3 32B run on a single GPU?

Yes. At FP8 (E4M3) the model needs 33.1 GB of VRAM, so it fits on a single Mac M3 Max (64GB) or anything with more VRAM. At full FP16 it needs 65.1 GB, which fits on one H100 SXM5 80GB or larger.

How many GPUs does Qwen 3 32B need?

At FP8 (E4M3) with 8,192-token context and 8 concurrent users, the cheapest configuration that fits is 8× RTX 3080 10GB (80 GB total VRAM). $0.0294 per 1M output tokens.

How fast can Qwen 3 32B be served?

On 2× Cerebras CS-3 (WSE-3) at FP8 (E4M3), the aggregate throughput is 24863 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 Qwen 3 32B need?

Weights alone: 64.0 GB at FP16, 32.0 GB at FP8 (E4M3). 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.

Should I run Qwen 3 32B in FP8 or FP16?

FP8 halves the weight footprint (64 GB → 32 GB) and is faster than FP16 on Hopper and Blackwell, where it has native kernels. FP16 remains the reference quality bar; FP8 loses a small amount of accuracy in exchange for roughly doubling the batch size you can serve per GPU.

Is it cheaper to self-host Qwen 3 32B than to call an API?

At 8× RTX 3080 10GB the marginal cost is $0.0294 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 Qwen 3 32B 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

Qwen 3 32B on specific GPUs

Qwen 3 32B on RTX 5090 32GBQwen 3 32B on AMD MI300X 192GBQwen 3 32B on L40S 48GBQwen 3 32B on AMD MI325X 288GBQwen 3 32B on A100 80GB SXM4Qwen 3 32B on Intel Gaudi 3 128GBQwen 3 32B on H100 SXM5 80GBQwen 3 32B on B200 SXM 192GB
Model architecture
The numbers behind the KV-cache and throughput math.
Total parameters32.00B
Active parameters per token32.00B
ArchitectureDense
Attention layoutFull attention (GQA)
Transformer layers64
Hidden dimension5120
Attention heads64 query / 8 KV (GQA)
Head dimension128
Vocabulary size151,936
Max context (catalog)41K
KV cache per token @ 8K256.0 KB

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