Quad RTX 5090 workstation (4x 32 GB)
The fastest consumer memory subsystem you can buy: 1,792 GB/s of GDDR7. Only 32 GB of it, which is the whole story of this card for local AI — extremely fast on anything that fits, useless on anything that does not.
NVIDIAcomponent18 models fit0 measured
Memory
128 GB
Official spec GDDR7
Memory bandwidth
7,168 GB/s
Official spec ~79% achieved in practice
Specifications
CPU
AMD Threadripper 7960X Official spec
GPU
4× 4x GeForce RTX 5090 32 GB Official spec
Memory
128 GB GDDR7 Official spec
Memory bandwidth
7,168 GB/s Official spec source
Achieved bandwidth ⓘ
~79% (5,675 GB/s) Assumption
FP16 compute ⓘ
~838 TFLOPS Official spec
System RAM
256 GB Official spec
Architecture
Discrete GPU with dedicated VRAM Official spec
Nominal power
2600 W Official spec
Measured load power
2250 W Measured
Idle power
190 W Measured
Released
30 Jan 2025 Official spec
128 GB of GDDR7 at the cost of a 2.6 kW power budget that exceeds a standard Dutch 16 A wall circuit shared with anything else. Included as the upper bound of the GPU-workstation approach; the electricity line in the economics is not a rounding error here. Achieved-bandwidth coefficient INHERITED from rtx-5090-workstation, which shares the same silicon and software stack.
Standout models on this machine
Fastest
gpt-oss-20b — ~387–556 t/s
MXFP4 · estimated confidence
Largest that fits
Related hardware
Dual RTX 5090 workstation (2x 32 GB)64 GB · 3,584 GB/sRTX 5090 workstation (1x 32 GB)32 GB · 1,792 GB/sRTX PRO 6000 Max-Q workstation (96 GB, 300 W)96 GB · 1,792 GB/sRTX PRO 6000 Blackwell workstation (96 GB)96 GB · 1,792 GB/sDual used RTX 4090 workstation (48 GB)48 GB · 2,016 GB/sNVIDIA DGX Spark 128 GB128 GB · 273 GB/s
Model performance on Quad RTX 5090 workstation (4x 32 GB)0 measured, 18 estimated
18 of 18 rows
| Model↕ | Quant | Memory↕ | Decode▼ | Prefill↕ | Context | Fit | Confidence↕ |
|---|---|---|---|---|---|---|---|
| gpt-oss-20b OpenAI · 20.915B (3.6B active) | MXFP4 | 14.2 GB | ~387–556 t/s | ~23090–47960 t/s | 128K | Comfortable | Estimated |
| gpt-oss-120b OpenAI · 116.829B (5.1B active) | MXFP4 | 67.7 GB | ~339–487 t/s | ~8210–17050 t/s | 128K | Comfortable | Estimated |
| Qwen3.8 Flash Next Alibaba Qwen · 180B (6B active) | MLX 4-bit | 104.0 GB | ~307–442 t/s | ~6100–12660 t/s | — | Borderline | Estimated |
| Qwen3-Coder 30B-A3B Alibaba Qwen · 30.532B (3.3B active) | Q8_0 | 34.7 GB | ~301–434 t/s | ~9980–20730 t/s | 256K | Comfortable | Estimated |
| Qwen3 30B-A3B Alibaba Qwen · 30.532B (3.3B active) | Q8_0 | 34.7 GB | ~301–434 t/s | ~9980–20730 t/s | 40K | Comfortable | Estimated |
| Gemma 4 26B-A4B Google DeepMind · 25.806B (3.8B active) | Q8_0 | 30.9 GB | ~281–404 t/s | ~10120–21010 t/s | 256K | Comfortable | Estimated |
| Qwen3.5 122B-A10B Alibaba Qwen · 125.086B (10B active) | Q4_K_M | 77.9 GB | ~223–320 t/s | ~2830–5880 t/s | 256K | Comfortable | Estimated |
| Qwen3 8B Alibaba Qwen · 8.191B | Q8_0 | 11.8 GB | ~196–282 t/s | ~12230–25400 t/s | 40K | Comfortable | Estimated |
| Phi-4 14B Microsoft · 14.66B | Q8_0 | 19.0 GB | ~128–185 t/s | ~6830–14190 t/s | 16K | Comfortable | Estimated |
| Devstral Small 24B Mistral AI · 23.572B | Q8_0 | 28.0 GB | ~87–125 t/s | ~4250–8830 t/s | 128K | Comfortable | Estimated |
| Mistral Small 3.2 24B Mistral AI · 24.011B | Q8_0 | 28.4 GB | ~85.7–123 t/s | ~4170–8670 t/s | 128K | Comfortable | Estimated |
| Gemma 3 27B Google DeepMind · 27.432B | Q8_0 | 34.6 GB | ~76.4–110 t/s | ~3650–7590 t/s | 128K | Comfortable | Estimated |
| Qwen3.8 27B Alibaba Qwen · 27.781B | Q8_0 | 33.1 GB | ~75.5–109 t/s | ~3610–7490 t/s | 256K | Comfortable | Estimated |
| Qwen3.6 27B Alibaba Qwen · 27.781B | Q8_0 | 33.1 GB | ~75.5–109 t/s | ~3610–7490 t/s | 256K | Comfortable | Estimated |
| Gemma 4 31B Google DeepMind · 31.273B | Q8_0 | 42.2 GB | ~68.1–98 t/s | ~3200–6650 t/s | 64K | Comfortable | Estimated |
| Qwen3 32B Alibaba Qwen · 32.762B | Q8_0 | 38.3 GB | ~65.3–94 t/s | ~3060–6350 t/s | 40K | Comfortable | Estimated |
| DeepSeek-R1-Distill 32B DeepSeek · 32.764B | Q8_0 | 38.3 GB | ~65.3–94 t/s | ~3060–6350 t/s | 128K | Comfortable | Estimated |
| Llama 3.3 70B Meta · 70.554B | Q8_0 | 78.0 GB | ~32.3–46.4 t/s | ~1420–2950 t/s | 64K | Comfortable | Estimated |
Rows in grey are estimates from our bandwidth model, not measurements — they are shown as a range and never as a precise figure. Use the “measured only” filter to see just the 0 pairings on this machine that a real benchmark backs.
Ownership economicsUnited States (federal) · C corporation · 8h/day
Monthly economic cost
$170
Calculated after tax
Codex/Claude Code
$100/mo
≈ $100/mo · local is $70 more
Net cash at purchase
$12,611
Calculated VAT not reclaimable
Total over 5 years
$10,224
Calculated after tax, after resale
Cost per USD/1M tokens
$2.85
Calculated 59.8M 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 10,719 over 5 years, straight-line to a 1,892 residual | $178.65 |
| Electricity 106.0 kWh/month at 0.140/kWh | $14.83 |
| Cost of capital 4.0%/yr on 7,251 average capital employed | $24.17 |
| Monthly cost before tax | $217.65 |
| Electricity tax shield Running costs are deductible business expenses | −$3.11 |
| First-year expensing §179 (100.0%) | −$44.14 |
| Monthly economic cost after tax | $170.40 |
Three different numbers, deliberately
Cash cost
$12,611
Money that leaves the bank account on day one, net of reclaimable VAT.
Accounting depreciation
$178.65/month
$10,719 written down over 5 years to a $1,892 residual.
After-tax economic cost
$170.40/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
$1,892 (15%) Assumption
Two GPU generations later, the card is worth a fraction of its list price.
Electricity
$0.140/kWh Assumption
Average power draw
602 W Calculated
Load 2250 W for 20% of powered hours, idle 190 W for the rest — a machine that is on is not generating tokens the whole time.
Investment allowances
§179 100.0% Official spec
Worth $2,648 in total — first-year expensing that replaces later tax depreciation.
Cost of capital
$24.17/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.