Nemotron-3-Nano-30B-A3B · 8 GPUs measured first-party · llama.cpp · Updated October 2026
What GPU Do You Need for Nemotron-3-Nano-30B-A3B?
Nemotron-3-Nano-30B-A3B on 8 GPUs, measured first-party: NVIDIA B200 leads at 345.9 tok/s, L40S trails at 191 tok/s, and it peaked at 26GB of VRAM.
Benchmarked weights: unsloth/Nemotron-3-Nano-30B-A3B-GGUF
NVIDIA B200
345.9 tok/s on Nemotron-3-Nano-30B-A3B, the ceiling. Measured on our bench. 192GB of VRAM, $40,000 at launch.
- 345.9 tok/s on Nemotron-3-Nano-30B-A3B
- 192GB, clears the Nemotron-3-Nano-30B-A3B floor
- Rentable by the hour rather than bought
- 1000W board rating
- Datacenter or workstation hardware, not a retail purchase
NVIDIA RTX PRO 6000 Blackwell Workstation Edition
325.4 tok/s on Nemotron-3-Nano-30B-A3B, lowest launch price that still fits. Measured on our bench. 96GB of VRAM, $8,565 at launch.
- 325.4 tok/s on Nemotron-3-Nano-30B-A3B
- 96GB, clears the Nemotron-3-Nano-30B-A3B floor
- Rentable by the hour rather than bought
- 600W board rating
- Datacenter or workstation hardware, not a retail purchase
What GPU Do You Need for Nemotron-3-Nano-30B-A3B?, tok/s by GPU
Efficiency: tok/s per 100W drawn
Power is the average pulled during the run, sampled at 1Hz. The fastest card is often not the one here, and for anything left running this is the number that shows up on the bill.
Value: tok/s per $1,000 of MSRP
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
Nemotron-3-Nano-30B-A3B. Measured tokens per second by GPU
| GPU | tok/s | Prompt t/s | tok/W | Avg power |
|---|---|---|---|---|
| NVIDIA B200 | 345.9 | 7825.2 | 1.13 | 305.8 W |
| NVIDIA B300 | 345.3 | 4555.3 | 1.24 | 279.6 W |
| NVIDIA H200 | 327.5 | 6921.8 | 2.1 | 156.1 W |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 325.4 | 10029.1 | 2.36 | 137.9 W |
| NVIDIA H100 80GB HBM3 | 316.5 | 6861 | 2 | 158.1 W |
| NVIDIA A100 80GB SXM4 | 204.1 | 3297.8 | 1.54 | 132.4 W |
| NVIDIA A100 40GB SXM4 | 192.4 | 3200.9 | 1.87 | 103.0 W |
| NVIDIA L40S | 191 | 6930.2 | 1.38 | 138.7 W |
What the numbers show. Across 8 GPUs measured on our own bench, B200 is fastest at 346 tok/s. The slowest, L40S, manages 191, so the spread is 1.8x from top to bottom. RTX PRO 6000 Blackwell Workstation Edition is the most efficient, 325 tok/s at 138W. Per dollar of launch price, RTX PRO 6000 Blackwell Workstation Edition gives the most (38.0 tok/s per $1,000).
About Nemotron-3-Nano-30B-A3B. Nemotron-3-Nano-30B-A3B: from nvidia, 32B parameters, on Hugging Face since December 2025. 2,639,734 downloads in the last 30 days and 2 community quantizations.
How it compares. H100 80GB HBM3: Nemotron-3-Nano-30B-A3B 316.5 tok/s, Nemotron 3.5 Lightning 30B A3B 323.6 (32B), Gemma 4 31B 76.59 (31B), GLM-4.7-Flash 186.1 (31B), Qwen3 30B A3B 283.8 (31B). 1 of 4 beat Nemotron-3-Nano-30B-A3B here.
Cost on a rented GPU. 1M generated tokens of Nemotron-3-Nano-30B-A3B: $0.68 on a A100 40GB SXM4 ($0.47/hr, 87 min), $4.80 on a B200 ($5.98/hr, 48 min, 7.0x the cost).
Nemotron-3-Nano-30B-A3B: cost per 1M generated tokens on rented GPUs
| GPU | Cheapest rate | Speed (tok/s) | Cost per 1M generated tokens |
|---|---|---|---|
| NVIDIA A100 40GB SXM4 | $0.47/hr | 192.4 | $0.68 |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | $1.08/hr | 325.4 | $0.92 |
| NVIDIA L40S | $0.79/hr | 191 | $1.15 |
| NVIDIA A100 80GB SXM4 | $0.95/hr | 204.1 | $1.29 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 316.5 | $1.87 |
| NVIDIA H200 | $3.59/hr | 327.5 | $3.05 |
| NVIDIA B200 | $5.98/hr | 345.9 | $4.80 |
| NVIDIA B300 | $6.94/hr | 345.3 | $5.58 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for Nemotron-3-Nano-30B-A3B. 30+ tok/s: 8 (B200, B300, H200). 30 tok/s is roughly where replies outpace reading.
Reading your prompt. Before Nemotron-3-Nano-30B-A3B writes anything it reads the input: 10029.1 tok/s on the RTX PRO 6000 Blackwell Workstation Edition (0.4s for a 4,000-token prompt), 3200.9 on the A100 40GB SXM4 (1.2s). Long documents and big code files feel this number more than the generation speed.
VRAM for Nemotron-3-Nano-30B-A3B. Measured peak 24.0GB, so 32GB is the smallest common card size; smallest card it ran on: A100 40GB SXM4 (40GB). With long context: Q4_K_M 31GB (tested), Q3_K_M 26GB, Q5_K_M 33GB, Q6_K 43GB, Q8_0 43GB.
Power on Nemotron-3-Nano-30B-A3B. Most efficient: RTX PRO 6000 Blackwell Workstation Edition, 138W, 0.12 kWh per 1M generated tokens. Hungriest: B200, 306W, 0.25 kWh. At $0.15/kWh: $0.018 per 1M generated tokens.
Our verdict
Fastest on Nemotron-3-Nano-30B-A3B: NVIDIA B200, 345.9 tok/s. Cheapest to rent per job: NVIDIA A100 40GB SXM4, $0.68 per 1M generated tokens.
FAQ
- What GPU do I need to run Nemotron-3-Nano-30B-A3B?
- About 25GB. Smallest card that ran it: NVIDIA A100 40GB SXM4 (40GB).
- How much does it cost to run Nemotron-3-Nano-30B-A3B in the cloud?
- $0.68 per 1M generated tokens on a NVIDIA A100 40GB SXM4 at $0.47/hr, cheapest of 8 rentable cards we measured.
- Can I run Nemotron-3-Nano-30B-A3B on a 12GB, 16GB or 24GB card?
- It used 24.0GB at the precision we tested. 12GB: no; 16GB: no; 24GB: no.
- Is the H100 80GB HBM3 or the A100 80GB SXM4 faster for Nemotron-3-Nano-30B-A3B?
- The H100 80GB HBM3: 316.5 vs 204.1 tok/s, 55% faster on our bench.
How we test
llama.cpp llama-bench at Q4_K_M, 512-token prompt and 128 generated tokens, three runs after a warmup, full GPU offload, with power and VRAM sampled throughout. Token generation is memory-bandwidth-bound, so the ranking tracks bandwidth closely, which makes it a fair guide to cards we haven't run yet.