SmolLM2-135M · 11 GPUs measured first-party · llama.cpp · Updated October 2026
What GPU Do You Need for SmolLM2-135M?
SmolLM2-135M on 11 GPUs, measured first-party: NVIDIA RTX PRO 6000 Blackwell Workstation Edition leads at 1252.8 tok/s, T4 trails at 416 tok/s, and it peaked at 1GB of VRAM.
Benchmarked weights: bartowski/SmolLM2-135M-Instruct-GGUF
NVIDIA RTX PRO 6000 Blackwell Workstation Edition
1252.8 tok/s on SmolLM2-135M, the ceiling. Measured on our bench. 96GB of VRAM, $8,565 at launch.
- 1252.8 tok/s on SmolLM2-135M
- 96GB, clears the SmolLM2-135M 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 SmolLM2-135M?, 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.
SmolLM2-135M. Measured tokens per second by GPU
| GPU | tok/s | Prompt t/s | tok/W | Avg power |
|---|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 1252.8 | 59353.3 | 13.41 | 93.4 W |
| NVIDIA L40S | 916.6 | 56033.4 | 8.94 | 102.5 W |
| NVIDIA H200 | 905.4 | 45619.4 | 7.02 | 128.9 W |
| NVIDIA H100 80GB HBM3 | 898.5 | 43485.5 | 7.21 | 124.7 W |
| NVIDIA B300 | 855.9 | 43943.1 | 3.51 | 243.5 W |
| NVIDIA B200 | 815.3 | 50922.1 | 3.24 | 252.0 W |
| NVIDIA A10G | 680.4 | 33442.5 | 9.61 | 70.8 W |
| NVIDIA L4 | 652.1 | 39454.5 | 18.12 | 36.0 W |
| NVIDIA A100 80GB SXM4 | 554.4 | 25841.1 | 5.35 | 103.7 W |
| NVIDIA A100 40GB SXM4 | 544.1 | 23946.6 | 7.93 | 68.6 W |
| NVIDIA T4 | 416.2 | 10221.7 | 11.69 | 35.6 W |
What the numbers show. Across 11 GPUs measured on our own bench, RTX PRO 6000 Blackwell Workstation Edition is fastest at 1253 tok/s. The slowest, T4, manages 416, so the spread is 3.0x from top to bottom. L4 is the most efficient, 652 tok/s at 36W. Per dollar of launch price, L4 gives the most (260.9 tok/s per $1,000). The fastest card with 16GB or less is T4 at 416 tok/s.
How it compares. H100 80GB HBM3: SmolLM2-135M 898.5 tok/s, Qwen2.5-Coder-0.5B 892.8, Qwen2.5-0.5B 890.0, Qwen2 0.5B 881.9, Llama 3.2 1B 880.6 (1B). SmolLM2-135M beats all 4 here.
Cost on a rented GPU. 1M generated tokens of SmolLM2-135M: $0.091 on a T4 ($0.14/hr, 40 min), $0.24 on a RTX PRO 6000 Blackwell Workstation Edition ($1.08/hr, 13 min, 2.6x the cost).
SmolLM2-135M: cost per 1M generated tokens on rented GPUs
| GPU | Cheapest rate | Speed (tok/s) | Cost per 1M generated tokens |
|---|---|---|---|
| NVIDIA T4 | $0.14/hr | 416.2 | $0.091 |
| NVIDIA L4 | $0.44/hr | 652.1 | $0.19 |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | $1.08/hr | 1252.8 | $0.24 |
| NVIDIA L40S | $0.79/hr | 916.6 | $0.24 |
| NVIDIA A100 40GB SXM4 | $0.47/hr | 544.1 | $0.24 |
| NVIDIA A100 80GB SXM4 | $0.95/hr | 554.4 | $0.47 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 898.5 | $0.66 |
| NVIDIA H200 | $3.59/hr | 905.4 | $1.10 |
| NVIDIA B200 | $5.98/hr | 815.3 | $2.04 |
| NVIDIA B300 | $6.94/hr | 855.9 | $2.25 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for SmolLM2-135M. 30+ tok/s: 11 (RTX PRO 6000 Blackwell Workstation Edition, L40S, H200). 30 tok/s is roughly where replies outpace reading.
Reading your prompt. Before SmolLM2-135M writes anything it reads the input: 59353.3 tok/s on the RTX PRO 6000 Blackwell Workstation Edition (0.1s for a 4,000-token prompt), 10221.7 on the T4 (0.4s). Long documents and big code files feel this number more than the generation speed.
VRAM for SmolLM2-135M. Measured peak 0.8GB, so 8GB is the smallest common card size; smallest card it ran on: T4 (16GB). With long context: Q4_K_M 1GB (tested), Q2_K 1GB, Q3_K_M 1GB, Q5_K_M 1GB, Q6_K 1GB.
Power on SmolLM2-135M. Most efficient: L4, 36W, 15.3 Wh per 1M generated tokens. Hungriest: B200, 252W, 85.9 Wh.
Our verdict
Fastest on SmolLM2-135M: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 1252.8 tok/s. Cheapest to rent per job: NVIDIA T4, $0.091 per 1M generated tokens.
FAQ
- What GPU do I need to run SmolLM2-135M?
- About 1GB. Smallest card that ran it: NVIDIA T4 (16GB).
- How much does it cost to run SmolLM2-135M in the cloud?
- $0.091 per 1M generated tokens on a NVIDIA T4 at $0.14/hr, cheapest of 10 rentable cards we measured.
- Can I run SmolLM2-135M on a 12GB, 16GB or 24GB card?
- It used 0.8GB at the precision we tested. 12GB: yes; 16GB: yes; 24GB: yes.
- Is the H100 80GB HBM3 or the A100 80GB SXM4 faster for SmolLM2-135M?
- The H100 80GB HBM3: 898.5 vs 554.4 tok/s, 62% 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.