How much VRAM Qwen3 30B-A3B (MoE) needs, and what it costs to run
Some links on this page are referral links: we may earn a commission, at no cost to you.
Trying to decide what to buy? The best GPU for Qwen3 30B-A3B (MoE) ↓ · Got a card already? See what your PC can run →
Cheapest way to rent it right now
$0.04/hr
NVIDIA L4 · Q5_K_M · spot · ≈ $9/mo at 8h/day
Live price captured Thu, 19 Nov 2026 21:20:10 GMT.It never changes which card is cheapest.
Memory needed, by quantisation
At 16k context. Weights + KV cache + ~0.8 GB overhead. Memory is in binary GB (1,024³ bytes), the unit GPU memory is sold in; Hugging Face and Ollama list downloads in decimal GB, about 7% more.
| Quant | Weights | KV cache | Total | Quality |
|---|---|---|---|---|
| F16 | 56.9 GB | 1.5 GB | 59.2 GB | Full precision. Reference quality, twice the size of Q8 for no practical gain in most chat use. |
| Q8_0 | 30.2 GB | 1.5 GB | 32.5 GB | Effectively lossless. Use when VRAM is not the constraint. |
| Q6_K | 23.3 GB | 1.5 GB | 25.6 GB | Very close to Q8 at meaningfully less memory. A safe high-quality pick. |
| Q5_K_M | 20.2 GB | 1.5 GB | 22.5 GB | Small, hard-to-notice quality loss. Good balance. |
| Q5_0 | 19.7 GB | 1.5 GB | 22.0 GB | Older-style 5-bit. Q5_K_M is usually the better pick at the same size. |
| Q4_K_M | 17.2 GB | 1.5 GB | 19.5 GB | The community default. Best quality-per-gigabyte for most people. |
| Q4_0 | 16.2 GB | 1.5 GB | 18.5 GB | Older-style 4-bit, measurably worse than Q4_K_M at a similar size. Avoid unless required. |
| IQ4_XS | 15.1 GB | 1.5 GB | 17.4 GB | Newer 4-bit, smaller than Q4_K_M with comparable quality. Needs a recent llama.cpp. |
| Q3_K_M | 13.9 GB | 1.5 GB | 16.2 GB | Noticeable degradation. Use to fit a larger model that would otherwise not run. |
| Q2_K | 11.9 GB | 1.5 GB | 14.2 GB | Heavy degradation. Almost always better to run a smaller model at Q4_K_M instead. |
| IQ3_XS | 11.7 GB | 1.5 GB | 14.0 GB | Aggressive. Usually better than a smaller model at Q4, but test before trusting it. |
Cards that can run it
Best quantisation each card fits at 16k context, with live rental price where we track one.
The best GPU for Qwen3 30B-A3B (MoE)
Ranked from live rental prices and computed VRAM fit, not opinion. Every card is judged at Q4_K_M with a 8K context, so the comparison is fair; a card that could only run a crushed version isn't listed.
Best value
NVIDIA RTX A5000
Best throughput per dollar among cards that fit.
Cheapest that runs it
NVIDIA RTX A5000
Lowest hourly rental that fits it — $0.076/hr, Q4_K_M.
No compromise
—
Most headroom and highest throughput.
Every card that runs it, ranked
Sorted by tokens/sec per rental dollar. Throughput is a bandwidth-derived estimate.
| GPU | VRAM | VRAM used | ~tok/s | Cheapest rental | tok/s per $ |
|---|---|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 96 GB | 19 GB20% | — | $0.400 | — |
| NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | 96 GB | 19 GB20% | — | $0.387 | — |
| NVIDIA RTX 6000 Ada Generation | 48 GB | 19 GB39% | — | $0.207 | — |
| NVIDIA RTX PRO 5000 Blackwell | 48 GB | 19 GB39% | — | $0.189 | — |
| NVIDIA RTX A6000 | 48 GB | 19 GB39% | — | $0.200 | — |
| AMD Radeon PRO W7900 | 48 GB | 19 GB39% | — | — | — |
| NVIDIA RTX PRO 4500 Blackwell | 32 GB | 19 GB59% | — | $0.267 | — |
| NVIDIA GeForce RTX 5090 | 32 GB | 19 GB59% | — | $0.231 | — |
| NVIDIA RTX 5000 Ada Generation | 32 GB | 19 GB59% | — | $0.265 | — |
| NVIDIA GeForce RTX 3090 Ti | 24 GB | 19 GB78% | — | $0.173 | — |
| AMD Radeon RX 7900 XTX | 24 GB | 19 GB78% | — | — | — |
| NVIDIA RTX PRO 4000 Blackwell | 24 GB | 19 GB78% | — | $0.107 | — |
The value column ranks renting, because that's what we can price precisely (prices captured Thu, 19 Nov 2026 21:20:10 UTC). Buying a card? The fit and throughput columns are what you need; we don't publish a purchase-price ranking because we don't hold verified street prices. Tokens/sec is an estimate, not a benchmark. Got a card already? See everything your PC can run.
qwen3-30b-a3bHow these numbers are produced
- Model shape comes from the model's own config.json — 48 layers, 32 attention heads, 4 KV heads.
- Weight size is parameters × effective bits-per-weight. Those constants are checked against real published quantised file sizes — median error 0.7% across 40 measurements.
- KV cache is 2 × layers × KV-heads × head-dim × context × 2 bytes. Using KV-heads rather than attention heads is what makes this correct for grouped-query attention; treating a GQA model as multi-head overstates the cache by up to 8×.
- Tokens/sec is an estimate, not a benchmark. Generation is memory-bandwidth-bound, so this is bandwidth ÷ weight-bytes derated to 75%. Real throughput depends on your runtime, batch size and kernels. We don't run our own hardware tests — see our editorial policy.
- This is a mixture-of-experts model. Generation reads only the experts active for each token (8 of 128), so throughput is scored on ~3.3B active parameters, not the full 31B. Memory is the opposite: every expert must still be resident in VRAM, so the figures above use the full weight set.Active = 3.3B, vendor figure (Qwen model card, huggingface.co/Qwen/Qwen3-30B-A3B, read 29 Oct 2026: "30.5B in total and 3.3B activated"). Also set in workers/prices/src/models.ts seeds so the daily ingest keeps it.
- Some memory-bandwidth figures are not yet independently verified.Where that's the case we show no tokens/sec at all rather than a number we can't stand behind.
- Rental prices are pulled hourly from provider APIs (last updated Thu, 19 Nov 2026 21:20:10 GMT). Spot/interruptible pricing can change or vanish without notice.
Try it interactively
Drag the context slider and watch the KV cache fill the card
Running Qwen3 30B-A3B (MoE): common questions
How much VRAM do you need to run Qwen3 30B-A3B (MoE)?
Can an RTX 4090 run Qwen3 30B-A3B (MoE)?
Can an RTX 3090 run Qwen3 30B-A3B (MoE)?
Can an RTX 4060 Ti 16GB run Qwen3 30B-A3B (MoE)?
What's the cheapest way to run Qwen3 30B-A3B (MoE)?
Worth reading before you buy
Llama 4 or Qwen3 for local use? Here's the honest comparison
The quiet truth: for most home users, Qwen3 is the better default than Llama 4 Scout. Where each actually wins, the LMArena controversy, and the specific cases where Scout is worth it.
Should you run gpt-oss, Qwen3 or DeepSeek as your open model in 2026?
Three of the best open models, three different strengths. gpt-oss is the cost-effective all-rounder, Qwen3 owns coding, DeepSeek owns hard reasoning. Here's how to pick for your work.
Cheapest RTX 4090 rental
Live per-hour 4090 prices on Vast and Runpod, the spot-vs-on-demand lever, and what a 24 GB card actually runs — the rent-a-card answer for everything in the 14B–32B class.
What --n-cpu-moe actually does
The trick that makes MoE models 5× faster — except it usually makes them slower, and the famous speedup only happens when the model didn't fit in the first place.