Devstral Small 24B on Dell Precision 7960 Rack (2x RTX PRO 6000, 192 GB)
23.572B on 192 GB at 3,584 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
Q8_0
Calculated gguf
Memory required
27.2 GB
Calculated of 169.0 GB usable — 16%
Max practical context
128K
Calculated model supports 128K
Memory budget at 8K context
Model weights ⓘ
23.8 GB Calculated
KV cache ⓘ
1.3 GB Calculated
Runtime overhead ⓘ
2.1 GB Estimated
Total required
27.2 GB Calculated
Usable memory ⓘ
169.0 GB Assumption
Headroom
141.8 GB Calculated
Highest-precision quantization that leaves headroom: uses 16% of usable memory at 8K context.
2 GPUs providing 192 GB aggregate VRAM. Assumes tensor- or layer-parallel sharding; each GPU carries its own context and communication buffers. Interconnect: PCIe 5.0 x16.
Multi-GPU throughput depends heavily on interconnect. Without NVLink or NVSwitch, tensor parallelism over PCIe adds latency that partially offsets the extra bandwidth.
PerformanceEstimated0/10
Decode (generation)
~53.7–77.3 t/s
Estimated calculated, not measured
Prefill (prompt)
~2720–5660 t/s
Estimated
TTFT at 8K
~1.5–3.1 s
Estimated time to first token
Power while generating
920 W
Measured 0.22 tokens/s per watt
These figures are estimated, not measured. Nobody has published a benchmark of Devstral Small 24B 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: 1792 GB/s per card, with each additional card contributing 40% of its bandwidth — 2509 GB/s effective, not the 3584 GB/s aggregate. Cross-GPU collectives use PCIe rather than a dedicated GPU fabric.
- Decode: reading 23.6B active parameters at 8.50 bits/weight takes 13.87 ms at 2509 GB/s x 72% achieved efficiency.
- Prefill: 380 TFLOPS (FP16) x 2 for native FP8 tensor cores x 26% assumed model-FLOPs utilisation, divided by 2 x 23.6B 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 | 12.8 GB | 15.9 GB | 9% | Comfortable | 128K |
Q4_K_M | 13.7 GB | 16.8 GB | 10% | Comfortable | 128K |
Q8_0recommended | 23.8 GB | 27.2 GB | 16% | Comfortable | 128K |
BF16 | 43.9 GB | 47.9 GB | 28% | Comfortable | 128K |
Cost of running Devstral Small 24B on this machineUnited States (federal) · C corporation · 8h/day
Monthly economic cost
$467
Calculated after tax
Codex/Claude Code
$100/mo
≈ $100/mo · local is $367 more
Net cash at purchase
$36,663
Calculated VAT not reclaimable
Total over 5 years
$28,021
Calculated after tax, after resale
Cost per USD/1M tokens
$56.26
Calculated 8.3M 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 31,163 over 5 years, straight-line to a 5,499 residual | $519.39 |
| Electricity 51.4 kWh/month at 0.140/kWh | $7.19 |
| Cost of capital 4.0%/yr on 21,081 average capital employed | $70.27 |
| Monthly cost before tax | $596.85 |
| Electricity tax shield Running costs are deductible business expenses | −$1.51 |
| First-year expensing §179 (100.0%) | −$128.32 |
| Monthly economic cost after tax | $467.02 |
Three different numbers, deliberately
Cash cost
$36,663
Money that leaves the bank account on day one, net of reclaimable VAT.
Accounting depreciation
$519.39/month
$31,163 written down over 5 years to a $5,499 residual.
After-tax economic cost
$467.02/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
$5,499 (15%) Assumption
Two GPU generations later, the card is worth a fraction of its list price.
Electricity
$0.140/kWh Assumption
Average power draw
292 W Calculated
Load 920 W for 20% of powered hours, idle 135 W for the rest — a machine that is on is not generating tokens the whole time.
Investment allowances
§179 100.0% Official spec
Worth $7,699 in total — first-year expensing that replaces later tax depreciation.
Cost of capital
$70.27/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
1.8 t/s
Calculated purchase price only
Decode per $100/month
14.0 t/s
Calculated after-tax ownership cost
Tokens per joule
0.22
Calculated same as tokens/s per watt
USD per 1M output tokens
$56.26
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 | 303.3M/mo | $13 | API cheaper | Different model |
| DeepSeek-V4.1 Flash DeepSeek | $0.60 | 259.5M/mo | $15 | API cheaper | Different model |
| DeepSeek-V4 Pro DeepSeek | $1.98 | 64.3M/mo | $60 | API cheaper | Different model |
| DeepSeek V4 Pro (Together) Together AI | $4.40 | 22.0M/mo | $176 | API cheaper | Different model |
| GLM-5.3 Z.ai | $4.40 | 29.9M/mo | $129 | API cheaper | Different model |
| Claude Haiku 4.5 Anthropic | $5.00 | 35.9M/mo | $108 | API 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 $467/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.
Devstral Small 24B on other hardware
| Hardware | Decode | Price | Fit |
|---|---|---|---|
| NVIDIA DGX H200 (8x H200, 1,128 GB) | ~354–509 t/s | $359,430 * | Comfortable |
| Quad RTX 5090 workstation (4x 32 GB) | ~87–125 t/s | $12,611 * | Comfortable |
| Lenovo ThinkStation PX (4x RTX PRO 6000, 384 GB) | ~76.3–110 t/s | $58,553 * | Comfortable |
| RTX 5090 workstation (1x 32 GB) | ~75.4–109 t/s | $3,332 * | Comfortable |
| Dual RTX 5090 workstation (2x 32 GB) | ~58.5–84.2 t/s | $5,854 * | Comfortable |
| Used RTX 4090 workstation (24 GB) | ~44.3–63.7 t/s | $2,071 * | Comfortable |
Nearest alternatives to the Dell Precision 7960 Rack (2x RTX PRO 6000, 192 GB)
Devstral Small 24B on Lenovo ThinkStation PX (4x RTX PRO 6000, 384 GB)384 GBDevstral Small 24B on CPU-only workstation (Ryzen 9950X, 192 GB DDR5)192 GBDevstral Small 24B on NVIDIA DGX H200 (8x H200, 1,128 GB)1,128 GBDevstral Small 24B on Mac Studio M5 Ultra 256 GB256 GBDevstral Small 24B on Quad RTX 5090 workstation (4x 32 GB)128 GB
Compare Dell Precision 7960 Rack (2x RTX PRO 6000, 192 GB) against Lenovo ThinkStation PX (4x RTX PRO 6000, 384 GB) →
Compare Dell Precision 7960 Rack (2x RTX PRO 6000, 192 GB) against Lenovo ThinkStation PX (4x RTX PRO 6000, 384 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)