Nemotron 3 Super 120B-A12B on GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB)
123.611B (12B active) on 128 GB at 256 GB/s. Hardware details · Model details
Yes. Nemotron 3 Super 120B-A12B fits on GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB) at the recommended
Q4_K_M configuration, requiring approximately 75.3 GB at 8K context. Practical context capacity is 1M. Expected decode for the recommended configuration is ~10.1–14.5 t/s. EstimatedCompatibilityComfortable
Calculated fit
Yes
Calculated Uses under 80% of usable memory. Room for a long context and other work.
Recommended quantization
Q4_K_M
Calculated gguf
Memory required
75.3 GB
Calculated of 105.0 GB usable — 72%
Max practical context
1M
Calculated model supports 1M
Memory budget at 8K context
Model weights ⓘ
71.9 GB Calculated
KV cache ⓘ
0.2 GB Calculated
Runtime overhead ⓘ
3.2 GB Estimated
Total required
75.3 GB Calculated
Headroom ⓘ
29.7 GB Calculated
Highest-precision quantization that leaves headroom: uses 72% 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 123.611B parameters must be resident in memory even though only ~12B are active per token. Memory follows total parameters; speed follows active parameters.
Hybrid cache: token-growing KV memory is charged only to 8 attention blocks; 40 Mamba-2 blocks use fixed-size convolution and recurrent state instead.
Performance
Estimated performance · recommended Q4_K_M
Decode, prefill and TTFT below are estimates for Q4_K_M. We do not have a comparable Q4_K_M measurement on this machine.
Estimated decode
~10.1–14.5 t/s
Estimated Q4_K_M; calculated range
Estimated prefill
~89.6–186 t/s
Estimated Q4_K_M; calculated range
Estimated TTFT at 8K
~44.1–91.5 s
Estimated Q4_K_M; calculated range
Hardware load reference
135 W
Measured machine-level load; not this model run
How the estimate is calculated
- Decode: reading 12.0B active parameters at 4.85 bits/weight takes 67.90 ms at 256 GB/s x 42% 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 38.5B parameters per token.
- Prefill uses 38.5B effective parameters, not the 12B 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.
Submit a benchmarkSubmissions stay pending until reviewed.
This model at other quantizations on this machine
| Quantization | Resident | Disk | Total needed | Utilisation | Fit | Max context |
|---|---|---|---|---|---|---|
Q4_K_Mrecommended | 71.9 GB | 71.9 GB | 75.3 GB | 72% | Comfortable | 1M |
Q5_K_M | 84.3 GB | 84.3 GB | 88.1 GB | 84% | Fits | 1M |
Q8_0 | 124.8 GB | 124.8 GB | 129.7 GB | 124% | Does not fit | 0 |
BF16 | 230.2 GB | 230.2 GB | 238.4 GB | 227% | Does not fit | 0 |
Cost of running Nemotron 3 Super 120B-A12B on this machineUnited States (federal) · C corporation · 8h/day
Monthly economic cost
$22
Calculated after tax
Codex/Claude Code
$100/mo
≈ $100/mo · local is $78 less
Net cash at purchase
$1,711
Calculated VAT not reclaimable
Total over 5 years
$1,336
Calculated after tax, after resale
Cost per USD/1M tokens
$14.30
Calculated 1.6M 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: $1,711 in United States (federal), tax/VAT excluded · estimated · GMKtec (converted catalogue estimate). This is the localized purchase input; the after-tax economic cost is calculated separately below.
Monthly breakdown
| Depreciation 1,454 over 5 years, straight-line to a 257 residual | $24.23 |
| Electricity 6.7 kWh/month at 0.140/kWh | $0.94 |
| Cost of capital 4.0%/yr on 984 average capital employed | $3.28 |
| Monthly cost before tax | $28.45 |
| Electricity tax shield Running costs are deductible business expenses | −$0.20 |
| First-year expensing §179 (100.0%) | −$5.99 |
| Monthly economic cost after tax | $22.27 |
Three different numbers, deliberately
Cash cost
$1,711
Money that leaves the bank account on day one, net of reclaimable VAT.
Accounting depreciation
$24.23/month
$1,454 written down over 5 years to a $257 residual.
After-tax economic cost
$22.27/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
$257 (15%) Assumption
Two GPU generations later, the card is worth a fraction of its list price.
Electricity
$0.140/kWh Assumption
Average power draw
38 W Calculated
Load 135 W for 20% of powered hours, idle 14 W for the rest — a machine that is on is not generating tokens the whole time.
Investment allowances
§179 100.0% Official spec
Worth $359 in total — first-year expensing that replaces later tax depreciation.
Cost of capital
$3.28/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
7.2 t/s
Calculated purchase price only
Decode per $100/month
55.2 t/s
Calculated after-tax ownership cost
Tokens per joule
0.32
Calculated same as tokens/s per watt
USD per 1M output tokens
$14.30
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 | 14.5M/mo | $2 | API cheaper | Different model |
| DeepSeek-V4.1 Flash DeepSeek | $0.60 | 12.4M/mo | $3 | API cheaper | Different model |
| DeepSeek-V4 Pro DeepSeek | $1.98 | 3.1M/mo | $11 | API cheaper | Different model |
| DeepSeek V4 Pro (Together) Together AI | $4.40 | 1.1M/mo | $33 | Local cheaper | Different model |
| GLM-5.3 Z.ai | $4.40 | 1.4M/mo | $24 | About equal | Different model |
| Claude Haiku 4.5 Anthropic | $5.00 | 1.7M/mo | $20 | About equal | 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 $22/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.
Nemotron 3 Super 120B-A12B on other hardware
| Hardware | Decode | Price | Fit |
|---|---|---|---|
| NVIDIA DGX H200 (8x H200, 1,128 GB) | ~421–605 t/s | $359,430 * | Comfortable |
| Quad RTX 5090 workstation (4x 32 GB) | ~178–256 t/s | $12,611 * | Comfortable |
| Lenovo ThinkStation PX (4x RTX PRO 6000, 384 GB) | ~117–169 t/s | $58,553 * | Comfortable |
| RTX PRO 6000 Blackwell workstation (96 GB) | ~113–163 t/s | $10,359 * | Fits |
| RTX PRO 6000 Max-Q workstation (96 GB, 300 W) | ~113–163 t/s | $10,809 * | Fits |
| Mac Studio M5 Ultra 96 GB | ~95.9–138 t/s | $5,945 * | Fits |
Nearest alternatives to the GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB)
Nemotron 3 Super 120B-A12B on Framework Desktop (Ryzen AI Max+ 395, 128 GB)128 GBNemotron 3 Super 120B-A12B on NVIDIA DGX Spark 128 GB128 GBNemotron 3 Super 120B-A12B on CPU-only workstation (Ryzen 9950X, 192 GB DDR5)192 GBNemotron 3 Super 120B-A12B on Mac Studio M5 Max 128 GB128 GBNemotron 3 Super 120B-A12B on Radeon AI PRO R9700 workstation (32 GB)32 GB
Compare GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB) against Framework Desktop (Ryzen AI Max+ 395, 128 GB) →
Compare GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB) against Framework Desktop (Ryzen AI Max+ 395, 128 GB) →