NVIDIA A100 PCIe 40 GB vs Dual used RTX 4090 workstation (48 GB)
A100 PCIe has 40 GB memory, 1,555 GB/s bandwidth, 250 W nominal power; Dual RTX 4090 has 48 GB memory, 2,016 GB/s bandwidth, 1,150 W nominal power. Compare model compatibility, like-for-like tokens/sec and ownership cost below.
NVIDIA A100 PCIe 40 GB · Dual used RTX 4090 workstation (48 GB)
Verdict
- throughputNo valid like-for-like performance comparison is available. The practical configurations below differ in quantization, runtime, context, settings, or evidence type, so their speed difference must not be attributed to hardware alone.
- 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 A100 PCIe 40 GB has approximately 97% lower total monthly cost ($1 against $48), for a C corporation in United States (federal).
- powerThe Dual used RTX 4090 workstation (48 GB) draws 980 W under load against 250 W for the NVIDIA A100 PCIe 40 GB — a 3.9× difference that shows up in the electricity line and in how loud and hot the room gets.
- evidenceThere are no comparable throughput rows, so this page makes no hardware-speed ranking. Practical deployment figures remain useful for fit planning but answer a different question.
- recommendationThroughput is close enough to be a wash, so this comes down to cost and capacity: the NVIDIA A100 PCIe 40 GB is cheaper to own, and capacity differences above decide which models you can run at all.
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 A100 PCIe 40 GB | Dual used RTX 4090 workstation (48 GB) | |
|---|---|---|
| Purchase price estimate | — | $3,692 * |
| Memory | 40 GB | 48 GB |
| Theoretical bandwidth (aggregate on multi-GPU) | 1,555 GB/s | 2,016 GB/s |
| Usable bandwidth indicator | 1,213 GB/s (78%) | interconnect-adjusted per workload |
| Nominal power | 250 W | 1150 W |
| Measured load power | — | 980 W |
| Idle power | — | 95 W |
| Architecture | discrete gpu | 2× GPU |
Like-for-like performance0 comparable models
Ternary Bonsai 2 27BMixtral 8x7BPhi-3.5 MoE 42B-A6.6BNemotron 3.5 Lightning 30B-A3BQwen3-Coder 30B-A3BQwen3 30B-A3BGemma 4 26B-A4Bgpt-oss-20b
| Model | NVIDIA A100 PCIe 40 GB | Dual used RTX 4090 workstation (48 GB) | Winner |
|---|---|---|---|
| Ternary Bonsai 2 27B 27.36B | — no matched configuration | — no matched configuration | Not comparable |
| Mixtral 8x7B 46.703B (13B active) | — no matched configuration | — no matched configuration | Not comparable |
| Phi-3.5 MoE 42B-A6.6B 41.873B (6.6B active) | — no matched configuration | — no matched configuration | Not comparable |
| Nemotron 3.5 Lightning 30B-A3B 31.578B (3B active) | — no matched configuration | — no matched configuration | Not comparable |
| Qwen3-Coder 30B-A3B 30.532B (3.3B active) | — no matched configuration | — no matched configuration | Not comparable |
| Qwen3 30B-A3B 30.532B (3.3B active) | — no matched configuration | — no matched configuration | Not comparable |
| Gemma 4 26B-A4B 25.806B (3.8B active) | — no matched configuration | — no matched configuration | Not comparable |
| gpt-oss-20b 20.915B (3.6B active) | — no matched configuration | — no matched configuration | Not comparable |
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 A100 PCIe 40 GB | Dual used RTX 4090 workstation (48 GB) | Practical result |
|---|---|---|---|
| Ternary Bonsai 2 27B | — PTQ1_0 · estimated | — PTQ1_0 · estimated | Both fit; configurations may differ |
| Mixtral 8x7B | — Q4_K_M · estimated | — Q4_K_M · estimated | Both fit; configurations may differ |
| Phi-3.5 MoE 42B-A6.6B | — Q4_K_M · estimated | — Q5_K_M · estimated | Both fit; configurations may differ |
| Nemotron 3.5 Lightning 30B-A3B | — Q5_K_M · estimated | ~223–321 t/s Q5_K_M · estimated | Both fit; configurations may differ |
| Qwen3-Coder 30B-A3B | — Q5_K_M · estimated | ~210–302 t/s Q5_K_M · estimated | Both fit; configurations may differ |
| Qwen3 30B-A3B | — Q5_K_M · estimated | ~210–302 t/s Q5_K_M · estimated | Both fit; configurations may differ |
| Gemma 4 26B-A4B | — Q5_K_M · estimated | ~143–206 t/s Q8_0 · estimated | Both fit; configurations may differ |
| gpt-oss-20b | — MXFP4 · estimated | ~238–343 t/s MXFP4 · 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 A100 PCIe 40 GB | Dual used RTX 4090 workstation (48 GB) | |
|---|---|---|
| Purchase price estimate | — | $3,692 |
| Reclaimable VAT | $0 | $0 |
| Net cash at purchase | $0 | $3,692 |
| Monthly depreciation | $0.00 | $48.00 |
| Monthly electricity | $1.82 (13 kWh) | $6.70 (48 kWh) |
| Tax programme | §179 100.0% | §179 100.0% |
| Tax-programme benefit | $0 | $775 |
| Monthly economic cost | $1.44 | $47.88 |
| Total over 5 years | $86 | $2,873 |
Purchase prices are localized inputs in USD: tax/VAT excluded for NVIDIA A100 PCIe 40 GB and tax/VAT excluded for Dual used RTX 4090 workstation (48 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.