Gemma 4 31B on NVIDIA DGX H200 (8x H200, 1,128 GB)
31.273B on 1,128 GB at 38,400 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
43.8 GB
Calculated of 992.6 GB usable — 4%
Max practical context
256K
Calculated model supports 256K
Memory budget at 8K context
Model weights ⓘ
31.6 GB Calculated
KV cache ⓘ
7.5 GB Calculated
Runtime overhead ⓘ
4.7 GB Estimated
Total required
43.8 GB Calculated
Usable memory ⓘ
992.6 GB Assumption
Headroom
948.8 GB Calculated
Highest-precision quantization that leaves headroom: uses 4% of usable memory at 8K context.
8 GPUs providing 1128 GB aggregate VRAM. Assumes tensor- or layer-parallel sharding; each GPU carries its own context and communication buffers. Interconnect: NVSwitch / 4th-generation NVLink, 900 GB/s per GPU.
The high-speed GPU fabric substantially reduces collective-communication overhead, although aggregate VRAM is still physically distributed.
PerformanceEstimated0/10
Decode (generation)
~313–451 t/s
Estimated calculated, not measured
Prefill (prompt)
~85560–177690 t/s
Estimated
TTFT at 8K
~0.1–0.2 s
Estimated time to first token
Power while generating
8500 W
Measured 0.13 tokens/s per watt
These figures are estimated, not measured. Nobody has published a benchmark of Gemma 4 31B 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: 4800 GB/s per card, with each additional card contributing 90% of its bandwidth — 35040 GB/s effective, not the 38400 GB/s aggregate. NVSwitch / 4th-generation NVLink, 900 GB/s per GPU carries cross-GPU collectives.
- Decode: reading 31.3B active parameters at 8.50 bits/weight takes 1.22 ms at 35040 GB/s x 78% achieved efficiency.
- Per-token overhead of 1.4 ms (kernel launches, attention bookkeeping, sampling) is significant here — this model is not purely bandwidth-bound on this hardware.
- Prefill: 15832 TFLOPS (FP16) x 2 for native FP8 tensor cores x 26% assumed model-FLOPs utilisation, divided by 2 x 31.3B parameters per token.
- With few active parameters on fast memory, fixed per-token overhead rather than bandwidth sets the ceiling. Real engines vary widely in how well they hide it.
- 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 | 17.0 GB | 28.8 GB | 3% | Comfortable | 256K |
Q4_K_M | 18.2 GB | 30.0 GB | 3% | Comfortable | 256K |
Q8_0recommended | 31.6 GB | 43.8 GB | 4% | Comfortable | 256K |
BF16 | 58.3 GB | 71.3 GB | 7% | Comfortable | 256K |
Cost of running Gemma 4 31B on this machineUnited States (federal) · C corporation · 8h/day
Monthly economic cost
$4,743
Calculated after tax
Codex/Claude Code
$100/mo
≈ $100/mo · local is $4,643 more
Net cash at purchase
$359,430
Calculated VAT not reclaimable
Total over 5 years
$284,555
Calculated after tax, after resale
Cost per USD/1M tokens
$97.90
Calculated 48.4M 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 316,298 over 5 years, straight-line to a 43,132 residual | $5,271.64 |
| Electricity 524.5 kWh/month at 0.140/kWh | $73.43 |
| Cost of capital 4.0%/yr on 201,281 average capital employed | $670.94 |
| Monthly cost before tax | $6,016.00 |
| Electricity tax shield Running costs are deductible business expenses | −$15.42 |
| First-year expensing §179 (100.0%) | −$1,258.00 |
| Monthly economic cost after tax | $4,742.58 |
Three different numbers, deliberately
Cash cost
$359,430
Money that leaves the bank account on day one, net of reclaimable VAT.
Accounting depreciation
$5,271.64/month
$316,298 written down over 5 years to a $43,132 residual.
After-tax economic cost
$4,742.58/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
$43,132 (12%) Assumption
Extrapolated.
Electricity
$0.140/kWh Assumption
Average power draw
2980 W Calculated
Load 8500 W for 20% of powered hours, idle 1600 W for the rest — a machine that is on is not generating tokens the whole time.
Investment allowances
§179 100.0% Official spec
Worth $75,480 in total — first-year expensing that replaces later tax depreciation.
Cost of capital
$670.94/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
1.1 t/s
Calculated purchase price only
Decode per $100/month
8.1 t/s
Calculated after-tax ownership cost
Tokens per joule
0.13
Calculated same as tokens/s per watt
USD per 1M output tokens
$97.90
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 | 3,079.6M/mo | $75 | API cheaper | Different model |
| DeepSeek-V4.1 Flash DeepSeek | $0.60 | 2,634.8M/mo | $87 | API cheaper | Different model |
| DeepSeek-V4 Pro DeepSeek | $1.98 | 653.2M/mo | $352 | API cheaper | Different model |
| DeepSeek V4 Pro (Together) Together AI | $4.40 | 223.7M/mo | $1,027 | API cheaper | Different model |
| GLM-5.3 Z.ai | $4.40 | 304.0M/mo | $756 | API cheaper | Different model |
| Claude Haiku 4.5 Anthropic | $5.00 | 364.8M/mo | $630 | 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 $4,743/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 4 31B on other hardware
| Hardware | Decode | Price | Fit |
|---|---|---|---|
| Quad RTX 5090 workstation (4x 32 GB) | ~68.1–98 t/s | $12,611 * | Comfortable |
| RTX 5090 workstation (1x 32 GB) | ~62.1–89.4 t/s | $3,332 * | Fits |
| Lenovo ThinkStation PX (4x RTX PRO 6000, 384 GB) | ~59.4–85.5 t/s | $58,553 * | Comfortable |
| Dual RTX 5090 workstation (2x 32 GB) | ~45.2–65.1 t/s | $5,854 * | Comfortable |
| Dual used RTX 4090 workstation (48 GB) | ~44–63.4 t/s | $3,692 * | Comfortable |
| Dell Precision 7960 Rack (2x RTX PRO 6000, 192 GB) | ~41.4–59.6 t/s | $36,663 * | Comfortable |
Nearest alternatives to the NVIDIA DGX H200 (8x H200, 1,128 GB)
Gemma 4 31B on NVIDIA DGX Station GB300 (748 GB)748 GBGemma 4 31B on Lenovo ThinkStation PX (4x RTX PRO 6000, 384 GB)384 GBGemma 4 31B on Dell Precision 7960 Rack (2x RTX PRO 6000, 192 GB)192 GBGemma 4 31B on Quad RTX 5090 workstation (4x 32 GB)128 GBGemma 4 31B on NVIDIA DGX Spark 128 GB128 GB
Compare NVIDIA DGX H200 (8x H200, 1,128 GB) against NVIDIA DGX Station GB300 (748 GB) →
Compare NVIDIA DGX H200 (8x H200, 1,128 GB) against NVIDIA DGX Station GB300 (748 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)