Skip to content

Granite 3.3 8B Instruct VRAM requirements

How much VRAM Granite 3.3 8B Instruct needs
4-bit weights: 4.6 GBWith 8K context: 6.5 GBWith 32K context: 10.3 GB8-bit, 8K context: 10.0 GB
Parameters
8.2B
Longest context
128K
Released
Apr 2025
License
Apache 2.0
Strongest at
Not rated

How much VRAM Granite 3.3 8B Instruct needs

Memory Granite 3.3 8B Instruct needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit15.2 GB17.1 GB20.9 GB35.9 GB
8-bit8.1 GB10.0 GB13.8 GB28.8 GB
6-bit6.2 GB8.2 GB11.9 GB26.9 GB
5-bit5.4 GB7.3 GB11.1 GB26.1 GB
4-bit4.6 GB6.5 GB10.3 GB25.3 GB
3-bit3.7 GB5.6 GB9.4 GB24.4 GB
2-bit2.9 GB4.8 GB8.5 GB23.5 GB
1.6-bit1.6 GB3.5 GB7.2 GB22.2 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Granite 3.3 8B Instruct

The best way Granite 3.3 8B Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40603-bit20K45 tok/s1.2 sPDF: Granite 3.3 8B Instruct on RTX 4060 report
RTX 50603-bit20K72 tok/s0.9 sPDF: Granite 3.3 8B Instruct on RTX 5060 report
RTX 30703-bit20K72 tok/s0.9 sPDF: Granite 3.3 8B Instruct on RTX 3070 report
RTX 3080 10 GB6-bit16K72 tok/s0.6 sPDF: Granite 3.3 8B Instruct on RTX 3080 10 GB report
RTX 3060 12 GB8-bit16K28 tok/s1.4 sPDF: Granite 3.3 8B Instruct on RTX 3060 12 GB report
Arc B5808-bit16K25 tok/s1.0 sPDF: Granite 3.3 8B Instruct on Arc B580 report
RTX 40708-bit16K39 tok/s0.6 sPDF: Granite 3.3 8B Instruct on RTX 4070 report
RTX 4070 Super8-bit16K39 tok/s0.5 sPDF: Granite 3.3 8B Instruct on RTX 4070 Super report
RTX 50708-bit16K51 tok/s0.6 sPDF: Granite 3.3 8B Instruct on RTX 5070 report
RTX 4060 Ti 16 GB8-bit40K23 tok/s0.8 sPDF: Granite 3.3 8B Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit40K35 tok/s0.8 sPDF: Granite 3.3 8B Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit40K30 tok/s0.9 sPDF: Granite 3.3 8B Instruct on Arc A770 16 GB report
Radeon RX 9070 XT8-bit40K41 tok/s0.5 sPDF: Granite 3.3 8B Instruct on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit40K51 tok/s0.4 sPDF: Granite 3.3 8B Instruct on RTX 4070 Ti Super report
RTX 4080 Super8-bit40K55 tok/s0.3 sPDF: Granite 3.3 8B Instruct on RTX 4080 Super report
RTX 5070 Ti8-bit40K66 tok/s0.4 sPDF: Granite 3.3 8B Instruct on RTX 5070 Ti report
RTX 50808-bit40K71 tok/s0.3 sPDF: Granite 3.3 8B Instruct on RTX 5080 report
Radeon RX 7900 XTX8-bit88K59 tok/s0.8 sPDF: Granite 3.3 8B Instruct on Radeon RX 7900 XTX report
RTX 30908-bit88K69 tok/s0.5 sPDF: Granite 3.3 8B Instruct on RTX 3090 report
RTX 40908-bit88K74 tok/s0.2 sPDF: Granite 3.3 8B Instruct on RTX 4090 report
Radeon AI PRO R97008-bit128K41 tok/s0.5 sPDF: Granite 3.3 8B Instruct on Radeon AI PRO R9700 report
RTX 50908-bit128K123 tok/s0.2 sPDF: Granite 3.3 8B Instruct on RTX 5090 report
2× RTX 30908-bit128K69 tok/s0.3 sPDF: Granite 3.3 8B Instruct on 2× RTX 3090 report
2× RTX 40908-bit128K74 tok/s0.1 sPDF: Granite 3.3 8B Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit128K96 tok/s0.3 sPDF: Granite 3.3 8B Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit128K71 tok/s0.2 sPDF: Granite 3.3 8B Instruct on RTX 6000 Ada report
L40S8-bit128K64 tok/s0.3 sPDF: Granite 3.3 8B Instruct on L40S report
2× RTX 50908-bit128K123 tok/s0.1 sPDF: Granite 3.3 8B Instruct on 2× RTX 5090 report
4× RTX 30908-bit128K69 tok/s0.2 sPDF: Granite 3.3 8B Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit128K123 tok/s0.1 sPDF: Granite 3.3 8B Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB6-bit20K14 tok/s11 sPDF: Granite 3.3 8B Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit40K11 tok/s11 sPDF: Granite 3.3 8B Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit72K11 tok/s11 sPDF: Granite 3.3 8B Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit40K24 tok/s5.7 sPDF: Granite 3.3 8B Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit128K24 tok/s5.7 sPDF: Granite 3.3 8B Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit128K24 tok/s5.7 sPDF: Granite 3.3 8B Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit128K41 tok/s2.8 sPDF: Granite 3.3 8B Instruct on M4 Max 128 GB report
M5 Max 128 GB8-bit128K45 tok/s0.8 sPDF: Granite 3.3 8B Instruct on M5 Max 128 GB report
M3 Ultra 96 GB8-bit128K51 tok/s1.6 sPDF: Granite 3.3 8B Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit128K51 tok/s1.6 sPDF: Granite 3.3 8B Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit128K51 tok/s1.6 sPDF: Granite 3.3 8B Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit128K70 tok/s0.5 sPDF: Granite 3.3 8B Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit128K17 tok/s1.6 sPDF: Granite 3.3 8B Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit128K17 tok/s1.6 sPDF: Granite 3.3 8B Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit128K24 tok/s0.6 sPDF: Granite 3.3 8B Instruct on RTX Spark report
DGX Spark8-bit128K21 tok/s0.6 sPDF: Granite 3.3 8B Instruct on DGX Spark report
Dual DGX Spark8-bit128K21 tok/s0.3 sPDF: Granite 3.3 8B Instruct on Dual DGX Spark report
DGX Station GB3008-bit128K339 tok/s0.1 sPDF: Granite 3.3 8B Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit128K123 tok/s0.1 sPDF: Granite 3.3 8B Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit128K123 tok/s0.1 sPDF: Granite 3.3 8B Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit128K123 tok/s0.1 sPDF: Granite 3.3 8B Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit128K137 tok/s0.4 sPDF: Granite 3.3 8B Instruct on A100 80 GB SXM report
H100 SXM8-bit128K204 tok/s0.1 sPDF: Granite 3.3 8B Instruct on H100 SXM report
4× H200 NVL8-bit128K264 tok/s0.1 sPDF: Granite 3.3 8B Instruct on 4× H200 NVL report
8× H200 NVL8-bit128K264 tok/s0.1 sPDF: Granite 3.3 8B Instruct on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf ibm-granite/granite-3.3-8b-instruct-GGUF
vLLM
vllm serve ibm-granite/granite-3.3-8b-instruct
MLX (Mac)
mlx_lm.chat --model mlx-community/granite-3.3-8b-instruct-4bit

Questions

How much VRAM does Granite 3.3 8B Instruct need?

Granite 3.3 8B Instruct needs 4.6 GB for its weights at 4-bit, 6.5 GB with an 8K context and 10.3 GB with 32K. At 8-bit, 10.0 GB.

What is the minimum VRAM for Granite 3.3 8B Instruct?

About 5.6 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is Granite 3.3 8B Instruct?

8.2B parameters: 15.2 GB of weights at 16-bit, 4.6 GB at 4-bit.

Can Granite 3.3 8B Instruct run on an RTX 4090?

Yes. At 8-bit it holds 88K of context and writes about 74 tokens a second.

More models