MiniCPM5 2B · 36 GPUs measured first-party · llama.cpp · Updated October 2026
What GPU Do You Need for MiniCPM5 2B?
MiniCPM5 2B on 36 GPUs, measured first-party: NVIDIA GeForce RTX 5090 leads at 507.3 tok/s, T4 trails at 109 tok/s, and it peaked at 2GB of VRAM.
Benchmarked weights: openbmb/MiniCPM5-2B-GGUF
NVIDIA GeForce RTX 5090
507.3 tok/s on MiniCPM5 2B, the ceiling. Measured on our bench. 32GB of VRAM, $1,999 at launch.
- 507.3 tok/s on MiniCPM5 2B
- 32GB, clears the MiniCPM5 2B floor
- 575W board rating
NVIDIA GeForce RTX 4090
372.7 tok/s on MiniCPM5 2B, fastest card you can buy at retail. Measured on our bench. 24GB of VRAM, $1,599 at launch.
- 372.7 tok/s on MiniCPM5 2B
- 24GB, clears the MiniCPM5 2B floor
- 450W board rating
GeForce RTX 4060
148.3 tok/s on MiniCPM5 2B, lowest launch price that still fits. Measured on our bench. 8GB of VRAM, $299 at launch.
- 148.3 tok/s on MiniCPM5 2B
- 8GB, clears the MiniCPM5 2B floor
- 115W board rating
NVIDIA GeForce RTX 3060 Ti
205.9 tok/s on MiniCPM5 2B, most speed per dollar. Measured on our bench. 8GB of VRAM, $399 at launch. That is 515.9 tok/s per $1,000 of launch price.
- 205.9 tok/s on MiniCPM5 2B
- 8GB, clears the MiniCPM5 2B floor
- 200W board rating
What GPU Do You Need for MiniCPM5 2B?, tok/s by GPU
Top 15 shown; 21 more cards in the full table below.
Efficiency: tok/s per 100W drawn
Top 15 shown; 20 more cards in the full table below.
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
Top 15 shown; 21 more cards in the full table below.
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
MiniCPM5 2B. Measured tokens per second by GPU
| GPU | tok/s | Prompt t/s | tok/W | Avg power |
|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | 507.3 | 28723 | 2.91 | 174.6 W |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 443.7 | 22287.3 | 3.29 | 135.0 W |
| NVIDIA H100 80GB HBM3 | 428.7 | 16175.6 | 2.62 | 163.8 W |
| NVIDIA GeForce RTX 4090 | 372.7 | 23142.6 | 3.54 | 105.3 W |
| GeForce RTX 5080 | 367.1 | 19505.8 | 2.91 | 126.1 W |
| GeForce RTX 5070 Ti | 337 | 16315.4 | 3.57 | 94.3 W |
| NVIDIA GeForce RTX 3090 Ti | 327.2 | 13152.9 | 1.71 | 191.4 W |
| NVIDIA GeForce RTX 4080 | 312.8 | 17084.5 | 2.89 | 108.3 W |
| GeForce RTX 4080 Super | 311.9 | 18153.9 | 2.57 | 121.5 W |
| NVIDIA L40S | 310.5 | 18164.7 | 2.41 | 128.9 W |
| NVIDIA GeForce RTX 3080 Ti | 304.6 | 11168.5 | 1.69 | 180.2 W |
| NVIDIA GeForce RTX 3090 | 296.8 | 11304.6 | 1.46 | 202.9 W |
| NVIDIA GeForce RTX 4070 Ti Super | 295.9 | 17776.5 | 2.68 | 110.3 W |
| GeForce RTX 5070 | 275.7 | 12943.3 | 3.35 | 82.4 W |
| NVIDIA GeForce RTX 3080 | 272.2 | 11508.7 | 1.51 | 179.8 W |
| NVIDIA RTX PRO 4000 Blackwell | 271.2 | 13094.2 | 3.34 | 81.3 W |
| NVIDIA RTX A6000 | 270.6 | 10673.9 | 1.71 | 158.5 W |
| NVIDIA RTX A5000 | 257.4 | 11268 | 1.99 | 129.3 W |
| NVIDIA A100 80GB SXM4 | 255.1 | 9702 | 1.69 | 151.4 W |
| NVIDIA GeForce RTX 4070 | 244.7 | 11473.3 | 2.75 | 89.0 W |
| NVIDIA GeForce RTX 3070 Ti | 244.3 | 8910.3 | 1.4 | 174.0 W |
| NVIDIA GeForce RTX 4070 Ti | 237.5 | 13488.5 | 2.31 | 103.0 W |
| NVIDIA Titan RTX | 225.2 | 7464.5 | 1.19 | 188.5 W |
| NVIDIA GeForce RTX 3060 Ti | 205.9 | 6692.8 | 1.87 | 109.8 W |
| NVIDIA GeForce RTX 3070 Founders Edition | 204.4 | 6963.9 | 1.79 | 113.9 W |
| NVIDIA A10G | 201.2 | 8058.9 | 1.93 | 104.5 W |
| GeForce RTX 5060 Ti | 200.6 | 8589.9 | 3.62 | 55.4 W |
| NVIDIA TITAN V | 192.1 | 6599.5 | 1.81 | 106.3 W |
| NVIDIA RTX A4000 | 188.8 | 7268.5 | 1.96 | 96.5 W |
| NVIDIA RTX 4000 (Ada Generation) | 183.4 | 11032.9 | 2.52 | 72.8 W |
| NVIDIA Quadro RTX 8000 | 181.2 | 6053 | 1.39 | 130.8 W |
| NVIDIA GeForce RTX 3060 | 167.6 | 5556.4 | 1.6 | 105.1 W |
| NVIDIA GeForce RTX 4060 Ti 16GB | 158.1 | 9286.2 | 2.17 | 73.0 W |
| GeForce RTX 4060 | 148.3 | 6917.7 | — | — |
| NVIDIA L4 | 138.2 | 8062.6 | 2.64 | 52.4 W |
| NVIDIA T4 | 109.1 | 3325.3 | 1.85 | 59.0 W |
What the numbers show. Across 36 GPUs measured on our own bench, RTX 5090 is fastest at 507 tok/s. The slowest, T4, manages 109, so the spread is 4.7x from top to bottom. RTX 5060 Ti is the most efficient, 201 tok/s at 55W. Per dollar of launch price, RTX 5060 Ti gives the most (609.7 tok/s per $1,000). The fastest card with 16GB or less is RTX 5080 at 367 tok/s.
About MiniCPM5 2B. MiniCPM5 2B: from openbmb, 2.5B parameters, on Hugging Face since September 2026, Apache 2.0 licence. 1,525,437 downloads in the last 30 days and 1 community quantizations.
How it compares. H100 80GB HBM3: MiniCPM5 2B 428.7 tok/s, gemma-2-2b 392.3 (3B), phi-2 340.1 (3B), Qwen3 1.7B 556.5 (2B), SmolLM3 3B 398.0 (3B). 1 of 4 beat MiniCPM5 2B here.
Cost on a rented GPU. 1M generated tokens of MiniCPM5 2B: $0.060 on a RTX 3060 ($0.036/hr, 99 min), $0.21 on a RTX 5090 ($0.39/hr, 33 min, 3.6x the cost).
MiniCPM5 2B: cost per 1M generated tokens on rented GPUs
| GPU | Cheapest rate | Speed (tok/s) | Cost per 1M generated tokens |
|---|---|---|---|
| NVIDIA GeForce RTX 3060 | $0.036/hr | 167.6 | $0.060 |
| NVIDIA GeForce RTX 3080 | $0.082/hr | 272.2 | $0.084 |
| NVIDIA GeForce RTX 3070 Founders Edition | $0.075/hr | 204.4 | $0.10 |
| NVIDIA GeForce RTX 3090 | $0.12/hr | 296.8 | $0.11 |
| NVIDIA RTX A4000 | $0.078/hr | 188.8 | $0.11 |
| GeForce RTX 5070 Ti | $0.15/hr | 337 | $0.12 |
| NVIDIA GeForce RTX 4070 | $0.12/hr | 244.7 | $0.14 |
| NVIDIA GeForce RTX 4070 Ti Super | $0.16/hr | 295.9 | $0.15 |
| GeForce RTX 5080 | $0.21/hr | 367.1 | $0.16 |
| NVIDIA GeForce RTX 3080 Ti | $0.18/hr | 304.6 | $0.16 |
| NVIDIA RTX A5000 | $0.16/hr | 257.4 | $0.17 |
| NVIDIA GeForce RTX 4080 | $0.20/hr | 312.8 | $0.18 |
| NVIDIA Titan RTX | $0.15/hr | 225.2 | $0.18 |
| GeForce RTX 5060 Ti | $0.14/hr | 200.6 | $0.19 |
| GeForce RTX 5070 | $0.19/hr | 275.7 | $0.19 |
| NVIDIA TITAN V | $0.14/hr | 192.1 | $0.20 |
| GeForce RTX 4080 Super | $0.22/hr | 311.9 | $0.20 |
| NVIDIA GeForce RTX 4070 Ti | $0.18/hr | 237.5 | $0.21 |
| NVIDIA RTX PRO 4000 Blackwell | $0.20/hr | 271.2 | $0.21 |
| NVIDIA GeForce RTX 5090 | $0.39/hr | 507.3 | $0.21 |
| NVIDIA GeForce RTX 3090 Ti | $0.27/hr | 327.2 | $0.23 |
| NVIDIA GeForce RTX 4090 | $0.34/hr | 372.7 | $0.25 |
| NVIDIA RTX 4000 (Ada Generation) | $0.20/hr | 183.4 | $0.30 |
| NVIDIA RTX A6000 | $0.33/hr | 270.6 | $0.34 |
| NVIDIA T4 | $0.14/hr | 109.1 | $0.35 |
| NVIDIA Quadro RTX 8000 | $0.26/hr | 181.2 | $0.39 |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | $1.08/hr | 443.7 | $0.67 |
| NVIDIA L40S | $0.79/hr | 310.5 | $0.71 |
| NVIDIA L4 | $0.44/hr | 138.2 | $0.88 |
| NVIDIA A100 80GB SXM4 | $0.95/hr | 255.1 | $1.03 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 428.7 | $1.38 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for MiniCPM5 2B. 30+ tok/s: 36 (RTX 5090, RTX 4090, RTX 5080). 30 tok/s is roughly where replies outpace reading.
Reading your prompt. Before MiniCPM5 2B writes anything it reads the input: 28723.0 tok/s on the RTX 5090 (0.1s for a 4,000-token prompt), 23142.6 on the RTX 4090 (0.2s), 3325.3 on the T4 (1.2s). Long documents and big code files feel this number more than the generation speed.
VRAM for MiniCPM5 2B. Measured peak 1.5GB, so 8GB is the smallest common card size; smallest card it ran on: RTX 3070 Ti (8GB). With long context: Q4_K_M 2GB (tested), Q8_0 4GB, BF16 7GB.
Power on MiniCPM5 2B. Most efficient: RTX 5060 Ti, 55W, 76.7 Wh per 1M generated tokens. Hungriest: RTX 3090, 203W, 0.19 kWh. At $0.15/kWh: $0.012 per 1M generated tokens.
Our verdict
Fastest on MiniCPM5 2B: NVIDIA GeForce RTX 5090, 507.3 tok/s. Cheapest consumer card that ran it: GeForce RTX 4060 ($299, 148.3 tok/s). Cheapest to rent per job: NVIDIA GeForce RTX 3060, $0.060 per 1M generated tokens.
FAQ
- What GPU do I need to run MiniCPM5 2B?
- About 2GB. Cheapest consumer card that ran it: GeForce RTX 4060 (8GB, 148.3 tok/s).
- How fast is MiniCPM5 2B on the NVIDIA GeForce RTX 5090?
- 507.3 tok/s at 175W, faster than every datacenter card we ran it on.
- How much does it cost to run MiniCPM5 2B in the cloud?
- $0.060 per 1M generated tokens on a NVIDIA GeForce RTX 3060 at $0.036/hr, cheapest of 31 rentable cards we measured.
- Can I run MiniCPM5 2B on a 12GB, 16GB or 24GB card?
- It used 1.5GB at the precision we tested. 12GB: yes; 16GB: yes; 24GB: yes.
- Is the RTX 4090 or the RTX 3090 faster for MiniCPM5 2B?
- The RTX 4090: 372.7 vs 296.8 tok/s, 26% faster on our bench.
- Should I buy or rent a RTX 5090 for MiniCPM5 2B?
- Its $1,999 launch price buys 5,139 rented hours at $0.39/hr, enough for about 9,385x 1M generated tokens of MiniCPM5 2B. Buy only if you'll run more than that.
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.