Mistral Small 24B (Q3_K_M) · 11 GPUs measured first-party · llama.cpp · Updated October 2026
What GPU Do You Need for Mistral Small 24B (Q3_K_M)?
Mistral Small 24B (Q3_K_M) on 11 GPUs, measured first-party: NVIDIA RTX PRO 6000 Blackwell Workstation Edition leads at 108.9 tok/s, T4 trails at 10.1 tok/s, and it peaked at 12GB of VRAM.
Benchmarked weights: bartowski/mistralai_Mistral-Small-3.1-24B-Instruct-2503-GGUF
NVIDIA RTX PRO 6000 Blackwell Workstation Edition
108.9 tok/s on Mistral Small 24B (Q3_K_M), the ceiling. Measured on our bench. 96GB of VRAM, $8,565 at launch.
- 108.9 tok/s on Mistral Small 24B (Q3_K_M)
- 96GB, clears the Mistral Small 24B (Q3_K_M) 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 Mistral Small 24B (Q3_K_M)?, 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.
Mistral Small 24B (Q3_K_M). Measured tokens per second by GPU
| GPU | tok/s | Prompt t/s | tok/W | Avg power |
|---|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 108.9 | 4460.9 | 0.33 | 334.1 W |
| NVIDIA B300 | 108.4 | 1564.7 | 0.24 | 460.3 W |
| NVIDIA B200 | 87.72 | 3517.5 | 0.18 | 483.3 W |
| NVIDIA H200 | 85.1 | 3128.9 | 0.24 | 359.0 W |
| NVIDIA H100 80GB HBM3 | 84 | 3176.9 | 0.25 | 341.0 W |
| NVIDIA L40S | 58.21 | 3177.8 | 0.22 | 263.1 W |
| NVIDIA A100 80GB SXM4 | 46.63 | 1260.3 | 0.18 | 259.4 W |
| NVIDIA A100 40GB SXM4 | 45.34 | 1241.5 | 0.23 | 199.6 W |
| NVIDIA A10G | 26.8 | 1111.9 | 0.2 | 131.3 W |
| NVIDIA L4 | 16.62 | 914.3 | 0.25 | 67.1 W |
| NVIDIA T4 | 10.13 | 381.4 | 0.15 | 65.6 W |
What the numbers show. Across 11 GPUs measured on our own bench, RTX PRO 6000 Blackwell Workstation Edition is fastest at 109 tok/s. The slowest, T4, manages 10.1, so the spread is 10.7x from top to bottom. The fastest card with 16GB or less is T4 at 10.1 tok/s.
How it compares. H100 80GB HBM3: Mistral Small 24B (Q3_K_M) 84.0 tok/s, Gemma 3 27B 84.77, Codestral 22B (Q3_K_M) 87.21, Qwen3.8 27B 80.5 (28B), Qwen3.6 27B 79.74 (28B). 2 of 4 beat Mistral Small 24B (Q3_K_M) here.
Cost on a rented GPU. 1M generated tokens of Mistral Small 24B (Q3_K_M): $2.75 on a RTX PRO 6000 Blackwell Workstation Edition ($1.08/hr, 2.6 hours).
Mistral Small 24B (Q3_K_M): cost per 1M generated tokens on rented GPUs
| GPU | Cheapest rate | Speed (tok/s) | Cost per 1M generated tokens |
|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | $1.08/hr | 108.9 | $2.75 |
| NVIDIA A100 40GB SXM4 | $0.47/hr | 45.34 | $2.89 |
| NVIDIA T4 | $0.14/hr | 10.13 | $3.73 |
| NVIDIA L40S | $0.79/hr | 58.21 | $3.77 |
| NVIDIA A100 80GB SXM4 | $0.95/hr | 46.63 | $5.64 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 84 | $7.06 |
| NVIDIA L4 | $0.44/hr | 16.62 | $7.35 |
| NVIDIA H200 | $3.59/hr | 85.1 | $11.72 |
| NVIDIA B300 | $6.94/hr | 108.4 | $17.79 |
| NVIDIA B200 | $5.98/hr | 87.72 | $18.94 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for Mistral Small 24B (Q3_K_M). 30+ tok/s: 8 (RTX PRO 6000 Blackwell Workstation Edition, B300, B200); 10-30 tok/s: 3 (A10G, L4, T4). 30 tok/s is roughly where replies outpace reading.
Reading your prompt. Before Mistral Small 24B (Q3_K_M) writes anything it reads the input: 4460.9 tok/s on the RTX PRO 6000 Blackwell Workstation Edition (0.9s for a 4,000-token prompt), 381.4 on the T4 (10.5s). Long documents and big code files feel this number more than the generation speed.
VRAM for Mistral Small 24B (Q3_K_M). Measured peak 11.3GB, so 12GB is the smallest common card size; smallest card it ran on: T4 (16GB).
Power on Mistral Small 24B (Q3_K_M). Most efficient: RTX PRO 6000 Blackwell Workstation Edition, 334W, 0.85 kWh per 1M generated tokens. Hungriest: B200, 483W, 1.53 kWh. At $0.15/kWh: $0.13 per 1M generated tokens.
Our verdict
Fastest on Mistral Small 24B (Q3_K_M): NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 108.9 tok/s. Cheapest to rent per job: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, $2.75 per 1M generated tokens.
FAQ
- What GPU do I need to run Mistral Small 24B (Q3_K_M)?
- About 12GB. Smallest card that ran it: NVIDIA T4 (16GB).
- How much does it cost to run Mistral Small 24B (Q3_K_M) in the cloud?
- $2.75 per 1M generated tokens on a NVIDIA RTX PRO 6000 Blackwell Workstation Edition at $1.08/hr, cheapest of 10 rentable cards we measured.
- Can I run Mistral Small 24B (Q3_K_M) on a 12GB, 16GB or 24GB card?
- It used 11.3GB at the precision we tested. 12GB: yes; 16GB: yes; 24GB: yes.
- Is the H100 80GB HBM3 or the A100 80GB SXM4 faster for Mistral Small 24B (Q3_K_M)?
- The H100 80GB HBM3: 84.0 vs 46.63 tok/s, 80% faster on our bench.
How we test
llama.cpp llama-bench at Q3_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.