Skip to content

Granite 3.0 8B Instruct VRAM requirements

How much VRAM Granite 3.0 8B Instruct needs
4-bit weights: 4.6 GBWith 4K context: 5.9 GB8-bit, 4K context: 9.4 GB
Parameters
8.2B
Longest context
4K
Released
Oct 2024
License
Apache 2.0
Strongest at
Not rated

How much VRAM Granite 3.0 8B Instruct needs

Memory Granite 3.0 8B Instruct needs at each precision and context length
PrecisionWeightsWith 4K
16-bit15.2 GB16.5 GB
8-bit8.1 GB9.4 GB
6-bit6.2 GB7.5 GB
5-bit5.4 GB6.7 GB
4-bit4.6 GB5.9 GB
3-bit3.7 GB5.0 GB
2-bit2.9 GB4.1 GB
1.6-bit1.6 GB2.9 GB

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

GPUs and Macs that run Granite 3.0 8B Instruct

The best way Granite 3.0 8B Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40606-bit4K27 tok/s1.2 sPDF: Granite 3.0 8B Instruct on RTX 4060 report
RTX 50606-bit4K44 tok/s0.9 sPDF: Granite 3.0 8B Instruct on RTX 5060 report
RTX 30706-bit4K44 tok/s0.9 sPDF: Granite 3.0 8B Instruct on RTX 3070 report
RTX 3080 10 GB8-bit4K57 tok/s0.6 sPDF: Granite 3.0 8B Instruct on RTX 3080 10 GB report
RTX 3060 12 GB8-bit4K28 tok/s1.4 sPDF: Granite 3.0 8B Instruct on RTX 3060 12 GB report
Arc B5808-bit4K25 tok/s1.0 sPDF: Granite 3.0 8B Instruct on Arc B580 report
RTX 40708-bit4K39 tok/s0.6 sPDF: Granite 3.0 8B Instruct on RTX 4070 report
RTX 4070 Super8-bit4K39 tok/s0.5 sPDF: Granite 3.0 8B Instruct on RTX 4070 Super report
RTX 50708-bit4K51 tok/s0.6 sPDF: Granite 3.0 8B Instruct on RTX 5070 report
RTX 4060 Ti 16 GB8-bit4K23 tok/s0.8 sPDF: Granite 3.0 8B Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit4K35 tok/s0.8 sPDF: Granite 3.0 8B Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit4K30 tok/s0.9 sPDF: Granite 3.0 8B Instruct on Arc A770 16 GB report
Radeon RX 9070 XT8-bit4K41 tok/s0.5 sPDF: Granite 3.0 8B Instruct on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit4K51 tok/s0.4 sPDF: Granite 3.0 8B Instruct on RTX 4070 Ti Super report
RTX 4080 Super8-bit4K55 tok/s0.3 sPDF: Granite 3.0 8B Instruct on RTX 4080 Super report
RTX 5070 Ti8-bit4K66 tok/s0.4 sPDF: Granite 3.0 8B Instruct on RTX 5070 Ti report
RTX 50808-bit4K71 tok/s0.3 sPDF: Granite 3.0 8B Instruct on RTX 5080 report
Radeon RX 7900 XTX8-bit4K59 tok/s0.8 sPDF: Granite 3.0 8B Instruct on Radeon RX 7900 XTX report
RTX 30908-bit4K69 tok/s0.5 sPDF: Granite 3.0 8B Instruct on RTX 3090 report
RTX 40908-bit4K74 tok/s0.2 sPDF: Granite 3.0 8B Instruct on RTX 4090 report
Radeon AI PRO R97008-bit4K41 tok/s0.5 sPDF: Granite 3.0 8B Instruct on Radeon AI PRO R9700 report
RTX 50908-bit4K123 tok/s0.2 sPDF: Granite 3.0 8B Instruct on RTX 5090 report
2× RTX 30908-bit4K69 tok/s0.3 sPDF: Granite 3.0 8B Instruct on 2× RTX 3090 report
2× RTX 40908-bit4K74 tok/s0.1 sPDF: Granite 3.0 8B Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit4K96 tok/s0.3 sPDF: Granite 3.0 8B Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit4K71 tok/s0.2 sPDF: Granite 3.0 8B Instruct on RTX 6000 Ada report
L40S8-bit4K64 tok/s0.3 sPDF: Granite 3.0 8B Instruct on L40S report
2× RTX 50908-bit4K123 tok/s0.1 sPDF: Granite 3.0 8B Instruct on 2× RTX 5090 report
4× RTX 30908-bit4K69 tok/s0.2 sPDF: Granite 3.0 8B Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit4K123 tok/s0.1 sPDF: Granite 3.0 8B Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit4K11 tok/s11 sPDF: Granite 3.0 8B Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit4K11 tok/s11 sPDF: Granite 3.0 8B Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit4K11 tok/s11 sPDF: Granite 3.0 8B Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit4K24 tok/s5.7 sPDF: Granite 3.0 8B Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit4K24 tok/s5.7 sPDF: Granite 3.0 8B Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit4K24 tok/s5.7 sPDF: Granite 3.0 8B Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit4K41 tok/s2.8 sPDF: Granite 3.0 8B Instruct on M4 Max 128 GB report
M5 Max 128 GB8-bit4K45 tok/s0.8 sPDF: Granite 3.0 8B Instruct on M5 Max 128 GB report
M3 Ultra 96 GB8-bit4K51 tok/s1.6 sPDF: Granite 3.0 8B Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit4K51 tok/s1.6 sPDF: Granite 3.0 8B Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit4K51 tok/s1.6 sPDF: Granite 3.0 8B Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit4K70 tok/s0.5 sPDF: Granite 3.0 8B Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit4K17 tok/s1.6 sPDF: Granite 3.0 8B Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit4K17 tok/s1.6 sPDF: Granite 3.0 8B Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit4K24 tok/s0.6 sPDF: Granite 3.0 8B Instruct on RTX Spark report
DGX Spark8-bit4K21 tok/s0.6 sPDF: Granite 3.0 8B Instruct on DGX Spark report
Dual DGX Spark8-bit4K21 tok/s0.3 sPDF: Granite 3.0 8B Instruct on Dual DGX Spark report
DGX Station GB3008-bit4K339 tok/s0.1 sPDF: Granite 3.0 8B Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit4K123 tok/s0.1 sPDF: Granite 3.0 8B Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit4K123 tok/s0.1 sPDF: Granite 3.0 8B Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit4K123 tok/s0.1 sPDF: Granite 3.0 8B Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit4K137 tok/s0.4 sPDF: Granite 3.0 8B Instruct on A100 80 GB SXM report
H100 SXM8-bit4K204 tok/s0.1 sPDF: Granite 3.0 8B Instruct on H100 SXM report
4× H200 NVL8-bit4K264 tok/s0.1 sPDF: Granite 3.0 8B Instruct on 4× H200 NVL report
8× H200 NVL8-bit4K264 tok/s0.1 sPDF: Granite 3.0 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 bartowski/granite-3.0-8b-instruct-GGUF
vLLM
vllm serve ibm-granite/granite-3.0-8b-instruct

Questions

How much VRAM does Granite 3.0 8B Instruct need?

Granite 3.0 8B Instruct needs 4.6 GB for its weights at 4-bit, 5.9 GB with an 4K context and 5.9 GB with 4K. At 8-bit, 9.4 GB.

What is the minimum VRAM for Granite 3.0 8B Instruct?

About 5.0 GB at 3-bit with an 4K 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 3.0 8B Instruct?

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

Can Granite 3.0 8B Instruct run on an RTX 4090?

Yes. At 8-bit it holds 4K of context and writes about 74 tokens a second.

More models