Gemma 3 27B on Dual used RTX 4090 workstation (48 GB)
27.432B on 48 GB at 2,016 GB/s. Hardware details · Model details
CompatibilityComfortable
Fits
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
21.7 GB
Calculated of 42.2 GB usable — 51%
Max practical context
32K
Calculated model supports 128K
Memory budget at 8K context
Model weights ⓘ
16.0 GB Calculated
KV cache ⓘ
3.9 GB Calculated
Runtime overhead ⓘ
1.9 GB Estimated
Total required
21.7 GB Calculated
Usable memory ⓘ
42.2 GB Assumption
Headroom
20.5 GB Calculated
Highest-precision quantization that leaves headroom: uses 51% of usable memory at 8K context.
2 GPUs pooling 48 GB total VRAM. Assumes tensor- or layer-parallel sharding; each GPU carries its own context and communication buffers.
Multi-GPU throughput depends heavily on interconnect. Without NVLink, tensor parallelism over PCIe adds latency that partially offsets the extra bandwidth.
PerformanceEstimated0/10
Decode (generation)
~49.7–71.5 t/s
Estimated calculated, not measured
Prefill (prompt)
~2030–4220 t/s
Estimated
TTFT at 8K
~2.0–4.2 s
Estimated time to first token
Power while generating
980 W
Measured 0.22 tokens/s per watt
These figures are estimated, not measured. Nobody has published a benchmark of Gemma 3 27B on this machine that we know of, so we calculate throughput from memory bandwidth and active parameter count. Estimates are always shown as a range. Expand the working below to see exactly how the number was produced.
How the estimate is calculated
- Multi-GPU: 1008 GB/s per card, with each additional card contributing 40% of its bandwidth — 1411 GB/s effective, not the 2016 GB/s aggregate. Consumer Blackwell has no NVLink, so every layer's all-reduce crosses PCIe.
- Decode: reading 27.4B active parameters at 4.85 bits/weight takes 15.11 ms at 1411 GB/s x 78% achieved efficiency.
- Prefill: 330 TFLOPS (FP16) x 2 for native FP8 tensor cores x 26% assumed model-FLOPs utilisation, divided by 2 x 27.4B parameters per token.
- 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 | Weights | Total needed | Utilisation | Fit | Max context |
|---|---|---|---|---|---|
MLX 4-bit | 14.9 GB | 20.7 GB | 49% | Comfortable | 32K |
Q4_K_Mrecommended | 16.0 GB | 21.7 GB | 51% | Comfortable | 32K |
Q8_0 | 27.7 GB | 33.8 GB | 80% | Fits | 16K |
BF16 | 51.1 GB | 57.9 GB | 137% | Does not fit | 0 |
Cost of running Gemma 3 27B on this machineUnited States (federal) · C corporation · 8h/day
Monthly economic cost
$48
Calculated after tax
Codex/Claude Code
$100/mo
≈ $100/mo · local is $52 less
Net cash at purchase
$3,692
Calculated VAT not reclaimable
Total over 5 years
$2,873
Calculated after tax, after resale
Cost per USD/1M tokens
$6.24
Calculated 7.7M 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.
Monthly breakdown
| Depreciation 2,880 over 5 years, straight-line to a 812 residual | $48.00 |
| Electricity 47.9 kWh/month at 0.140/kWh | $6.70 |
| Cost of capital 4.0%/yr on 2,252 average capital employed | $7.51 |
| Monthly cost before tax | $62.21 |
| Electricity tax shield Running costs are deductible business expenses | −$1.41 |
| First-year expensing §179 (100.0%) | −$12.92 |
| Monthly economic cost after tax | $47.88 |
Three different numbers, deliberately
Cash cost
$3,692
Money that leaves the bank account on day one, net of reclaimable VAT.
Accounting depreciation
$48.00/month
$2,880 written down over 5 years to a $812 residual.
After-tax economic cost
$47.88/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
$812 (22%) Assumption
Extrapolated.
Electricity
$0.140/kWh Assumption
Average power draw
272 W Calculated
Load 980 W for 20% of powered hours, idle 95 W for the rest — a machine that is on is not generating tokens the whole time.
Investment allowances
§179 100.0% Official spec
Worth $775 in total — first-year expensing that replaces later tax depreciation.
Cost of capital
$7.51/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
16.4 t/s
Calculated purchase price only
Decode per $100/month
126.5 t/s
Calculated after-tax ownership cost
Tokens per joule
0.22
Calculated same as tokens/s per watt
USD per 1M output tokens
$6.24
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 | 31.1M/mo | $12 | API cheaper | Different model |
| DeepSeek-V4 Flash DeepSeek | $0.66 | 19.8M/mo | $19 | API cheaper | Different model |
| DeepSeek-V4 Pro DeepSeek | $1.98 | 6.6M/mo | $56 | Local cheaper | Different model |
| DeepSeek V4 Pro (Together) Together AI | $4.40 | 2.3M/mo | $163 | Local cheaper | Different model |
| GLM-5.3 Z.ai | $4.40 | 3.1M/mo | $120 | Local cheaper | Different model |
| Claude Haiku 4.5 Anthropic | $5.00 | 3.7M/mo | $100 | 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 $48/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.
Gemma 3 27B on other hardware
| Hardware | Decode | Price | Fit |
|---|---|---|---|
| Quad RTX 5090 workstation (4x 32 GB) | ~76.4–110 t/s | €13.999 | Comfortable |
| RTX 5090 workstation (1x 32 GB) | ~65.5–94.2 t/s | €3.699 | Comfortable |
| Dual RTX 5090 workstation (2x 32 GB) | ~51–73.4 t/s | €6.499 | Comfortable |
| Used RTX 4090 workstation (24 GB) | ~38.3–55.1 t/s | €2.299 | Borderline |
| RTX PRO 6000 Blackwell workstation (96 GB) | ~36.1–52 t/s | €11.499 | Comfortable |
| RTX PRO 6000 Max-Q workstation (96 GB, 300 W) | ~36.1–52 t/s | €11.999 | Comfortable |
Nearest alternatives to the Dual used RTX 4090 workstation (48 GB)
Gemma 3 27B on Used RTX 4090 workstation (24 GB)24 GBGemma 3 27B on RTX 5090 workstation (1x 32 GB)32 GBGemma 3 27B on Dual RTX 5090 workstation (2x 32 GB)64 GBGemma 3 27B on RTX PRO 6000 Blackwell workstation (96 GB)96 GBGemma 3 27B on RTX PRO 6000 Max-Q workstation (96 GB, 300 W)96 GB
Compare Dual used RTX 4090 workstation (48 GB) against Used RTX 4090 workstation (24 GB) →
Compare Dual used RTX 4090 workstation (48 GB) against Used RTX 4090 workstation (24 GB) →
Page quality score (why this page is or is not indexed)
10/13 — indexed. Generated pages are gated so we do not ask a search engine to rank a page with nothing computed to say. The directive is emitted in the page head via the metadata API, not in the body, so it is authoritative.
- Has a memory-fit calculation (+3)
- Has a performance estimate (+2)
- Has a price (+2)
- Has an ownership economics calculation (+2)
- Well connected (18 internal links) (+1)