NVIDIA DGX H200 (8x H200, 1,128 GB) vs NVIDIA DGX Station GB300 (748 GB)
DGX H200 has 1,128 GB memory, 38,400 GB/s bandwidth, 10,200 W nominal power; DGX Station GB300 has 748 GB memory, 7,100 GB/s bandwidth, 1,600 W nominal power. Compare model compatibility, like-for-like tokens/sec and ownership cost below.
NVIDIA DGX H200 (8x H200, 1,128 GB) · NVIDIA DGX Station GB300 (748 GB)
Verdict
- throughputThe NVIDIA DGX H200 (8x H200, 1,128 GB) delivers approximately 902% higher aggregate decode throughput across 5 like-for-like model configurations (same quantization, runtime, context and inference settings).
- capacityBoth machines run all 8 compared models in their best practical configurations. Compare their quantizations and evidence labels before treating any throughput difference as a hardware result.
- costAt 8 hours a day, the NVIDIA DGX Station GB300 (748 GB) has approximately 75% lower total monthly cost ($1,182 against $4,743), for a C corporation in United States (federal).
- powerThe NVIDIA DGX H200 (8x H200, 1,128 GB) draws 8500 W under load against 1350 W for the NVIDIA DGX Station GB300 (748 GB) — a 6.3× difference that shows up in the electricity line and in how loud and hot the room gets.
- evidenceAll 5 like-for-like throughput rows are estimates. Treat the ranking as indicative and the margins as uncertain.
- recommendationThere is no single answer: the NVIDIA DGX H200 (8x H200, 1,128 GB) is faster, the NVIDIA DGX Station GB300 (748 GB) is cheaper to own, and the NVIDIA DGX H200 (8x H200, 1,128 GB) gives more throughput per unit of purchase currency. Pick on whichever constraint actually binds you — interactive latency, monthly budget, or capacity headroom for larger models later.
Every statement above is generated from the calculations on this page by fixed rules — no prose is written for individual comparisons. If the numbers change, the verdict changes with them.
Hardware
| NVIDIA DGX H200 (8x H200, 1,128 GB) | NVIDIA DGX Station GB300 (748 GB) | |
|---|---|---|
| Purchase price estimate | $359,430 * | $90,082 * |
| Memory | 1,128 GB | 748 GB |
| Theoretical bandwidth (aggregate on multi-GPU) | 38,400 GB/s | 7,100 GB/s |
| Usable bandwidth indicator | interconnect-adjusted per workload | 1,065 GB/s (15%) |
| Nominal power | 10200 W | 1600 W |
| Measured load power | 8500 W | 1350 W |
| Idle power | 1600 W | 180 W |
| Architecture | 8× GPU | unified memory |
Like-for-like performance5 comparable models
Kimi K2.6GLM-5.3MiniMax M3GLM-5.3 FlashTernary Bonsai 2 27BKimi K2 1T-A32BMiMo-V2.6 Pro 1.02T-A42BDeepSeek-V4.1 Flash
| Model | NVIDIA DGX H200 (8x H200, 1,128 GB) | NVIDIA DGX Station GB300 (748 GB) | Winner |
|---|---|---|---|
| Kimi K2.6 1.0T (32B active) | ~367–528 t/s Q4_K_M | ~36.6–52.6 t/s Q4_K_M | NVIDIA DGX H200 (8x H200, 1,128 GB) +904% estimated · estimator_v1 · 8K context |
| GLM-5.3 753.33B (40B active) | ~312–450 t/s Q5_K_M | ~25.3–36.4 t/s Q5_K_M | NVIDIA DGX H200 (8x H200, 1,128 GB) +1136% estimated · estimator_v1 · 8K context |
| MiniMax M3 427.04B (23B active) | ~334–481 t/s Q8_0 | ~29.3–42.1 t/s Q8_0 | NVIDIA DGX H200 (8x H200, 1,128 GB) +1042% estimated · estimator_v1 · 8K context |
| GLM-5.3 Flash 321.323B (18B active) | ~369–531 t/s Q8_0 | ~37.1–53.3 t/s Q8_0 | NVIDIA DGX H200 (8x H200, 1,128 GB) +895% estimated · estimator_v1 · 8K context |
| Ternary Bonsai 2 27B 27.36B | — no matched configuration | — no matched configuration | Not comparable |
| Kimi K2 1T-A32B 1.0T (32B active) | — no matched configuration | — no matched configuration | Not comparable |
| MiMo-V2.6 Pro 1.02T-A42B 1.0T (42B active) | — no matched configuration | — no matched configuration | Not comparable |
| DeepSeek-V4.1 Flash 763.205B (16B active) | ~434–624 t/s Q5_K_M | ~60.4–86.9 t/s Q5_K_M | NVIDIA DGX H200 (8x H200, 1,128 GB) +618% estimated · estimator_v1 · 8K context |
Hardware deltas are calculated only when quantization, runtime, exact context, execution/decode mode and relevant inference settings match. Measured and estimated results are never compared to each other. A difference under 10% is reported as a tie.
Best practical configuration
| Model | NVIDIA DGX H200 (8x H200, 1,128 GB) | NVIDIA DGX Station GB300 (748 GB) | Practical result |
|---|---|---|---|
| Kimi K2.6 | ~345–496 t/s Q5_K_M · estimated | ~36.6–52.6 t/s Q4_K_M · estimated | Both fit; configurations may differ |
| GLM-5.3 | ~254–365 t/s Q8_0 · estimated | ~29.5–42.5 t/s Q4_K_M · estimated | Both fit; configurations may differ |
| MiniMax M3 | ~334–481 t/s Q8_0 · estimated | ~29.3–42.1 t/s Q8_0 · estimated | Both fit; configurations may differ |
| GLM-5.3 Flash | ~369–531 t/s Q8_0 · estimated | ~37.1–53.3 t/s Q8_0 · estimated | Both fit; configurations may differ |
| Ternary Bonsai 2 27B | — PTQ1_0 · estimated | — PTQ1_0 · estimated | Both fit; configurations may differ |
| Kimi K2 1T-A32B | — Q5_K_M · estimated | — Q4_K_M · estimated | Both fit; configurations may differ |
| MiMo-V2.6 Pro 1.02T-A42B | — Q5_K_M · estimated | — Q4_K_M · estimated | Both fit; configurations may differ |
| DeepSeek-V4.1 Flash | ~434–624 t/s Q5_K_M · estimated | ~70–101 t/s Q4_K_M · estimated | Both fit; configurations may differ |
This is a deployment comparison: each machine uses its own recommended configuration. Different quantizations can change speed and quality, so these figures are not raw hardware benchmarks.
Economics — United States (federal), C corporation, 8h/day
| NVIDIA DGX H200 (8x H200, 1,128 GB) | NVIDIA DGX Station GB300 (748 GB) | |
|---|---|---|
| Purchase price estimate | $359,430 | $90,082 |
| Reclaimable VAT | $0 | $0 |
| Net cash at purchase | $359,430 | $90,082 |
| Monthly depreciation | $5,271.64 | $1,321.20 |
| Monthly electricity | $73.43 (524 kWh) | $10.20 (73 kWh) |
| Tax programme | §179 100.0% | §179 100.0% |
| Tax-programme benefit | $75,480 | $18,917 |
| Monthly economic cost | $4,742.58 | $1,182.12 |
| Total over 5 years | $284,555 | $70,927 |
Purchase prices are localized inputs in USD: tax/VAT excluded for NVIDIA DGX H200 (8x H200, 1,128 GB) and tax/VAT excluded for NVIDIA DGX Station GB300 (748 GB). Net cash and monthly economic cost then apply the selected United States (federal) / C corporation tax treatment. Tax assumptions last verified 2026-09-07. 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.