Qwen3-Coder 30B-A3B on Framework Desktop (Ryzen AI Max+ 395, 128 GB)
30.532B (3.3B active) on 128 GB at 256 GB/s. Hardware details · Model details
Yes. Qwen3-Coder 30B-A3B fits on Framework Desktop (Ryzen AI Max+ 395, 128 GB) at the recommended
Q8_0 configuration, requiring approximately 33.5 GB at 8K context. Practical context capacity is 256K. Measured alternative: 52.00 t/s median across 1 comparable run at Q4_K_M and 4K context. MeasuredCompatibilityComfortable
Calculated fit
Yes
Calculated Uses under 80% of usable memory. Room for a long context and other work.
Recommended quantization
Q8_0
Calculated gguf
Memory required
33.5 GB
Calculated of 105.0 GB usable — 32%
Max practical context
256K
Calculated model supports 256K
Observed resident fit: Q4_K_M with eng_llama_cpp at 4K context. This is measured evidence for those exact artifacts and settings; it does not change the standard memory calculation above.
Memory budget at 8K context
Model weights ⓘ
30.8 GB Calculated
KV cache ⓘ
0.8 GB Calculated
Runtime overhead ⓘ
1.9 GB Estimated
Total required
33.5 GB Calculated
Headroom ⓘ
71.5 GB Calculated
Highest-precision quantization that leaves headroom: uses 32% of usable memory at 8K context.
Unified LPDDR: the accelerator allocates from the same 128 GB pool as the OS. We assume 82% is available to the inference engine.
Mixture of experts: all 30.532B parameters must be resident in memory even though only ~3.3B are active per token. Memory follows total parameters; speed follows active parameters.
Performance
Estimated performance · recommended Q8_0
Decode, prefill and TTFT below are estimates for Q8_0. We do not have a comparable Q8_0 measurement on this machine.
Estimated decode
~23.5–33.8 t/s
Estimated Q8_0; calculated range
Estimated prefill
~344–714 t/s
Estimated Q8_0; calculated range
Estimated TTFT at 8K
~11.5–23.9 s
Estimated Q8_0; calculated range
Hardware load reference
130 W
Measured machine-level load; not this model run
Measured results · Q4_K_M
1 measured runThese are real Q4_K_M measurements. The recommendation above uses Q8_0, so they are not used to produce its estimated KPIs.
| Quant | Engine | Decode mode | Context | Runs | Median decode | Median prefill | Latest | Sources |
|---|---|---|---|---|---|---|---|---|
Q4_K_M | llama.cpp | standard | 4K | 1 | 52.00 t/s | — | 10 Feb 2026 | mindstudio.ai |
How the estimate is calculated
- Decode: reading 3.3B active parameters at 8.50 bits/weight takes 28.40 ms at 256 GB/s x 48% achieved efficiency.
- MoE routing penalty of 15% applied: expert gathers are less bandwidth-efficient than a dense sweep.
- Prefill: 59 TFLOPS (FP16) x 18% assumed model-FLOPs utilisation, divided by 2 x 10.0B parameters per token.
- Prefill uses 10.0B effective parameters, not the 3.3B active in decode: a batch of hundreds of tokens routes across most of the expert pool.
- MoE decode depends on how well the engine batches expert gathers; real results vary more than for dense models.
- Prefill throughput is highly engine-dependent. Flash attention, batch size and quantized KV all move this number substantially, and for sparse mixture-of-experts models it is the least reliable figure we produce.
- This is a calculated estimate, not a measurement. It assumes a single request, a short prompt, no speculative decoding and a warm model already resident in memory.
Estimator version estimator_v1. Stored with every estimated row so old estimates can be regenerated when the model improves.
This model at other quantizations on this machine
| Quantization | Resident | Disk | Total needed | Utilisation | Fit | Max context |
|---|---|---|---|---|---|---|
Q4_K_M | 17.8 GB | 17.8 GB | 20.0 GB | 19% | ComfortableResident run observed | 256K |
Q5_K_M | 20.8 GB | 20.8 GB | 23.2 GB | 22% | Comfortable | 256K |
Q8_0recommended | 30.8 GB | 30.8 GB | 33.5 GB | 32% | Comfortable | 256K |
BF16 | 56.9 GB | 56.9 GB | 60.3 GB | 57% | Comfortable | 256K |
Cost of running Qwen3-Coder 30B-A3B on this machineUnited States (federal) · C corporation · 8h/day
Monthly economic cost
$28
Calculated after tax
Codex/Claude Code
$100/mo
≈ $100/mo · local is $72 less
Net cash at purchase
$2,161
Calculated VAT not reclaimable
Total over 5 years
$1,673
Calculated after tax, after resale
Cost per USD/1M tokens
$7.68
Calculated 3.6M tokens/month
The $100 comparison uses the Codex Pro 5x / Claude Max 5x plans. A fixed planning conversion is used for USD. This compares monthly spend only: subscriptions have usage limits, local hardware has different capabilities and constraints, and taxes or regional pricing may change the charged amount. Prices checked 7 September 2026.
Purchase-price input: $2,161 in United States (federal), tax/VAT excluded · estimated · Framework (converted catalogue estimate). This is the localized purchase input; the after-tax economic cost is calculated separately below.
Monthly breakdown
| Depreciation 1,837 over 5 years, straight-line to a 324 residual | $30.62 |
| Electricity 6.3 kWh/month at 0.140/kWh | $0.88 |
| Cost of capital 4.0%/yr on 1,243 average capital employed | $4.14 |
| Monthly cost before tax | $35.63 |
| Electricity tax shield Running costs are deductible business expenses | −$0.18 |
| First-year expensing §179 (100.0%) | −$7.56 |
| Monthly economic cost after tax | $27.89 |
Three different numbers, deliberately
Cash cost
$2,161
Money that leaves the bank account on day one, net of reclaimable VAT.
Accounting depreciation
$30.62/month
$1,837 written down over 5 years to a $324 residual.
After-tax economic cost
$27.89/month
Depreciation plus running costs plus cost of capital, less the tax those deductions save. This is the figure to compare between machines.
Assumptions and sources (verified 2026-09-07)
VAT rate
0% Official spec
VAT recoverable
0% Official spec
Federal C-corporation estimate at the flat 21% rate. Pass-through entities should use the sole-proprietor estimate as a rougher proxy.
Effective deduction rate
21.00% Calculated
Headline marginal rate 21.00%.
Depreciation
5 years, straight-line Assumption
Residual value
$324 (15%) Assumption
Two GPU generations later, the card is worth a fraction of its list price.
Electricity
$0.140/kWh Assumption
Average power draw
36 W Calculated
Load 130 W for 20% of powered hours, idle 12 W for the rest — a machine that is on is not generating tokens the whole time.
Investment allowances
§179 100.0% Official spec
Worth $454 in total — first-year expensing that replaces later tax depreciation.
Cost of capital
$4.14/month Assumption
- VAT rate: No US federal VAT · verified 2026-09-07
- Marginal tax rate: IRS Publication 542 — Corporations · verified 2026-09-07
- VAT recoverable fraction: site assumption · verified 2026-09-07
- Useful life: IRS Publication 946 — How To Depreciate Property · verified 2026-09-07
- Electricity price: Site assumption · verified 2026-09-07
Assumes the machine generates tokens 20% of its 176 powered hours per month. Cost per token scales inversely with this number — halve the utilisation and the cost per token doubles.
Federal planning estimate only, not tax advice. State and local income tax, sales/use tax and incentives are excluded. It assumes 100% business use, a Section 179 election, enough business income to use it, and that the full annual limit remains available.
Efficiency metrics
Decode per $1,000 spent
13.3 t/s
Calculated purchase price only
Decode per $100/month
102.7 t/s
Calculated after-tax ownership cost
Tokens per joule
0.80
Calculated same as tokens/s per watt
USD per 1M output tokens
$7.68
Calculated at 20% utilisation
Local versus hosted APIs
| Hosted model | USD/1M output | Break-even | API at your volume | Verdict | Comparison type |
|---|---|---|---|---|---|
| Qwen3.8 Flash Alibaba Cloud | $0.42 | 18.1M/mo | $6 | API cheaper | Different model |
| DeepSeek-V4.1 Flash DeepSeek | $0.60 | 15.5M/mo | $7 | API cheaper | Different model |
| DeepSeek-V4 Pro DeepSeek | $1.98 | 3.8M/mo | $26 | About equal | Different model |
| DeepSeek V4 Pro (Together) Together AI | $4.40 | 1.3M/mo | $77 | Local cheaper | Different model |
| GLM-5.3 Z.ai | $4.40 | 1.8M/mo | $57 | Local cheaper | Different model |
| Claude Haiku 4.5 Anthropic | $5.00 | 2.1M/mo | $47 | Local cheaper | Different model |
Different-model comparison: the hosted model is not the model you would run locally. Treat this as a workload-quality trade-off, not a direct economic equivalence — the frontier model may complete a task in fewer tokens, or complete tasks the local model cannot. Break-even is the monthly output volume at which API spend equals the $28/month economic cost of owning this machine, assuming 8 input tokens per output token. Hosted prices are published in USD. This table is the canonical US default.
Qwen3-Coder 30B-A3B on other hardware
| Hardware | Decode | Price | Fit |
|---|---|---|---|
| NVIDIA DGX H200 (8x H200, 1,128 GB) | ~529–761 t/s | $359,430 * | Comfortable |
| Quad RTX 5090 workstation (4x 32 GB) | ~301–434 t/s | $12,611 * | Comfortable |
| RTX 5090 workstation (1x 32 GB) | ~299–430 t/s | $3,332 * | Comfortable |
| Lenovo ThinkStation PX (4x RTX PRO 6000, 384 GB) | ~279–402 t/s | $58,553 * | Comfortable |
| Dual RTX 5090 workstation (2x 32 GB) | ~236–339 t/s | $5,854 * | Comfortable |
| Dell Precision 7960 Rack (2x RTX PRO 6000, 192 GB) | ~223–320 t/s | $36,663 * | Comfortable |
Nearest alternatives to the Framework Desktop (Ryzen AI Max+ 395, 128 GB)
Qwen3-Coder 30B-A3B on GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB)128 GBQwen3-Coder 30B-A3B on Framework Desktop (Ryzen AI Max+ 395, 64 GB)64 GBQwen3-Coder 30B-A3B on NVIDIA DGX Spark 128 GB128 GBQwen3-Coder 30B-A3B on Radeon AI PRO R9700 workstation (32 GB)32 GBQwen3-Coder 30B-A3B on Mac Studio M5 Max 128 GB128 GB
Compare Framework Desktop (Ryzen AI Max+ 395, 128 GB) against GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB) →
Compare Framework Desktop (Ryzen AI Max+ 395, 128 GB) against GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB) →