Granite 4.1 3B · 36 GPUs measured first-party · llama.cpp Q4_K_M · Updated October 2026

What GPU Do You Need for Granite 4.1 3B?

We ran Granite 4.1 3B (Q4_K_M, llama.cpp) on 36 GPUs, from budget cards to datacenter parts, and measured generation speed, prompt processing, power draw and peak VRAM on every one. The fastest was NVIDIA GeForce RTX 5090 at 375.4 tok/s.

Benchmarked weights: ibm-granite/granite-4.1-3b-GGUF

Fastest we measured

NVIDIA GeForce RTX 5090

375.4 tok/s on Granite 4.1 3B, the ceiling. Measured on our bench. 32GB of VRAM, $1,999 at launch.

Pros

• 375.4 tok/s on Granite 4.1 3B
• 32GB, clears the Granite 4.1 3B floor

Cons

• 575W board rating

Best consumer card

NVIDIA GeForce RTX 4090

276.6 tok/s on Granite 4.1 3B, fastest card you can buy at retail. Measured on our bench. 24GB of VRAM, $1,599 at launch.

Pros

• 276.6 tok/s on Granite 4.1 3B
• 24GB, clears the Granite 4.1 3B floor

Cons

• 450W board rating

Cheapest card that runs it

GeForce RTX 4060

100.7 tok/s on Granite 4.1 3B, lowest launch price that still fits. Measured on our bench. 8GB of VRAM, $299 at launch.

Pros

• 100.7 tok/s on Granite 4.1 3B
• 8GB, clears the Granite 4.1 3B floor

Cons

• 115W board rating

Best value

NVIDIA GeForce RTX 3060 Ti

146.2 tok/s on Granite 4.1 3B, most speed per dollar. Measured on our bench. 8GB of VRAM, $399 at launch. That is 366.5 tok/s per $1,000 of launch price.

Pros

• 146.2 tok/s on Granite 4.1 3B
• 8GB, clears the Granite 4.1 3B floor

Cons

• 200W board rating

What the numbers show. Across 36 GPUs measured on our own bench, RTX 5090 is fastest at 375 tok/s. RTX 5080 is the most efficient, 254 tok/s at 121W (2.091 tok/W). Per dollar of launch price, RTX 5060 Ti gives the most (397.2 tok/s per $1,000). The fastest card with 16GB or less is RTX 5080 at 254 tok/s. The measured peak was ~3GB, so it runs on 6GB cards and up. That floor is a property of the model, so it applies to every GPU, measured or not.

About Granite 4.1 3B. Granite 4.1 3B: from ibm-granite, 3.4B parameters, on Hugging Face since April 2026, Apache 2.0 licence. 772,711 downloads in the last 30 days and 1 community quantizations.

How it compares. H100 80GB HBM3: Granite 4.1 3B 332.3 tok/s, Llama 3.2 3B 422.9 (3B), Qwen2.5-3B 395.5 (3B), Qwen2.5-Coder-3B 394.0 (3B), SmolLM3 3B 398.0 (3B). All 4 beat Granite 4.1 3B here.

Cost on a rented GPU. 1M generated tokens of Granite 4.1 3B: $0.11 on a RTX 3080 ($0.082/hr, 82 min), $0.29 on a RTX 5090 ($0.39/hr, 44 min, 2.6x the cost).

Speed tiers for Granite 4.1 3B. 30+ tok/s: 36 (RTX 5090, RTX 4090, RTX 5080). 30 tok/s is roughly where replies outpace reading.

Reading your prompt. Before Granite 4.1 3B writes anything it reads the input: 20663.4 tok/s on the RTX 5090 (0.2s for a 4,000-token prompt), 15825.8 on the RTX 4090 (0.3s), 2350.5 on the T4 (1.7s). Long documents and big code files feel this number more than the generation speed.

VRAM for Granite 4.1 3B. Measured peak 2.3GB, so 8GB is the smallest common card size; smallest card it ran on: RTX 3070 Ti (8GB). With long context: Q4_K_M 3GB (tested), Q2_K 2GB, Q3_K_M 3GB, Q5_K_M 4GB, Q6_K 4GB.

Power on Granite 4.1 3B. Most efficient: RTX 5080, 121W, 0.13 kWh per 1M generated tokens. Hungriest: RTX 3090 Ti, 231W, 0.26 kWh. At $0.15/kWh: $0.020 per 1M generated tokens.

Our verdict

Fastest on Granite 4.1 3B: NVIDIA GeForce RTX 5090, 375.4 tok/s. Cheapest consumer card that ran it: GeForce RTX 4060 ($299, 100.7 tok/s). Cheapest to rent per job: NVIDIA GeForce RTX 3080, $0.11 per 1M generated tokens.

FAQ

What GPU do I need to run Granite 4.1 3B?

About 2GB. Cheapest consumer card that ran it: GeForce RTX 4060 (8GB, 100.7 tok/s).

How fast is Granite 4.1 3B on the NVIDIA GeForce RTX 5090?

375.4 tok/s at 190W, faster than every datacenter card we ran it on.

How much does it cost to run Granite 4.1 3B in the cloud?

$0.11 per 1M generated tokens on a NVIDIA GeForce RTX 3080 at $0.082/hr, cheapest of 31 rentable cards we measured.

Can I run Granite 4.1 3B on a 12GB, 16GB or 24GB card?

It used 2.3GB at the precision we tested. 12GB: yes; 16GB: yes; 24GB: yes.

Is the RTX 4090 or the RTX 3090 faster for Granite 4.1 3B?

The RTX 4090: 276.6 vs 227.5 tok/s, 22% faster on our bench.

Should I buy or rent a RTX 5090 for Granite 4.1 3B?

Its $1,999 launch price buys 5,139 rented hours at $0.39/hr, enough for about 6,944x 1M generated tokens of Granite 4.1 3B. 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.