Skip to content

Granite 3.1 3B A800m Instruct VRAM requirements

How much VRAM Granite 3.1 3B A800m Instruct needs
4-bit weights: 1.9 GBWith 8K context: 3.0 GBWith 32K context: 4.5 GB8-bit, 8K context: 4.4 GB
Parameters
3.3B
Active per token
883M
Longest context
128K
Released
Dec 2024
License
Apache 2.0
Strongest at
Not rated

How much VRAM Granite 3.1 3B A800m Instruct needs

Memory Granite 3.1 3B A800m Instruct needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit6.1 GB7.2 GB8.7 GB14.7 GB
8-bit3.3 GB4.4 GB5.9 GB11.9 GB
6-bit2.5 GB3.6 GB5.1 GB11.1 GB
5-bit2.2 GB3.3 GB4.8 GB10.8 GB
4-bit1.9 GB3.0 GB4.5 GB10.5 GB
3-bit1.5 GB2.6 GB4.1 GB10.1 GB
2-bit1.2 GB2.2 GB3.7 GB9.7 GB
1.6-bit649 MB1.7 GB3.2 GB9.2 GB

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

GPUs and Macs that run Granite 3.1 3B A800m Instruct

The best way Granite 3.1 3B A800m Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit56K146 tok/s0.4 sPDF: Granite 3.1 3B A800m Instruct on RTX 4060 report
RTX 50608-bit56K204 tok/s0.3 sPDF: Granite 3.1 3B A800m Instruct on RTX 5060 report
RTX 30708-bit56K204 tok/s0.3 sPDF: Granite 3.1 3B A800m Instruct on RTX 3070 report
RTX 3080 10 GB8-bit88K271 tok/s0.2 sPDF: Granite 3.1 3B A800m Instruct on RTX 3080 10 GB report
RTX 3060 12 GB8-bit120K177 tok/s0.5 sPDF: Granite 3.1 3B A800m Instruct on RTX 3060 12 GB report
Arc B5808-bit120K137 tok/s0.4 sPDF: Granite 3.1 3B A800m Instruct on Arc B580 report
RTX 40708-bit120K218 tok/s0.2 sPDF: Granite 3.1 3B A800m Instruct on RTX 4070 report
RTX 4070 Super8-bit120K218 tok/s0.2 sPDF: Granite 3.1 3B A800m Instruct on RTX 4070 Super report
RTX 50708-bit120K255 tok/s0.2 sPDF: Granite 3.1 3B A800m Instruct on RTX 5070 report
RTX 4060 Ti 16 GB8-bit128K152 tok/s0.3 sPDF: Granite 3.1 3B A800m Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit128K204 tok/s0.3 sPDF: Granite 3.1 3B A800m Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit128K153 tok/s0.3 sPDF: Granite 3.1 3B A800m Instruct on Arc A770 16 GB report
Radeon RX 9070 XT8-bit128K188 tok/s0.2 sPDF: Granite 3.1 3B A800m Instruct on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit128K255 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on RTX 4070 Ti Super report
RTX 4080 Super8-bit128K267 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on RTX 4080 Super report
RTX 5070 Ti8-bit128K293 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on RTX 5070 Ti report
RTX 50808-bit128K302 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on RTX 5080 report
Radeon RX 7900 XTX8-bit128K222 tok/s0.3 sPDF: Granite 3.1 3B A800m Instruct on Radeon RX 7900 XTX report
RTX 30908-bit128K298 tok/s0.2 sPDF: Granite 3.1 3B A800m Instruct on RTX 3090 report
RTX 40908-bit128K308 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on RTX 4090 report
Radeon AI PRO R97008-bit128K188 tok/s0.2 sPDF: Granite 3.1 3B A800m Instruct on Radeon AI PRO R9700 report
RTX 50908-bit128K375 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on RTX 5090 report
2× RTX 30908-bit128K298 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on 2× RTX 3090 report
2× RTX 40908-bit128K308 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit128K343 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit128K302 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on RTX 6000 Ada report
L40S8-bit128K288 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on L40S report
2× RTX 50908-bit128K375 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on 2× RTX 5090 report
4× RTX 30908-bit128K298 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit128K375 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit96K71 tok/s4.1 sPDF: Granite 3.1 3B A800m Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit128K71 tok/s4.1 sPDF: Granite 3.1 3B A800m Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit128K71 tok/s4.1 sPDF: Granite 3.1 3B A800m Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit128K115 tok/s2.1 sPDF: Granite 3.1 3B A800m Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit128K115 tok/s2.1 sPDF: Granite 3.1 3B A800m Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit128K115 tok/s2.1 sPDF: Granite 3.1 3B A800m Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit128K148 tok/s1.0 sPDF: Granite 3.1 3B A800m Instruct on M4 Max 128 GB report
M5 Max 128 GB8-bit128K153 tok/s0.3 sPDF: Granite 3.1 3B A800m Instruct on M5 Max 128 GB report
M3 Ultra 96 GB8-bit128K160 tok/s0.6 sPDF: Granite 3.1 3B A800m Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit128K160 tok/s0.6 sPDF: Granite 3.1 3B A800m Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit128K160 tok/s0.6 sPDF: Granite 3.1 3B A800m Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit128K176 tok/s0.2 sPDF: Granite 3.1 3B A800m Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit128K111 tok/s0.6 sPDF: Granite 3.1 3B A800m Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit128K111 tok/s0.6 sPDF: Granite 3.1 3B A800m Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit128K157 tok/s0.2 sPDF: Granite 3.1 3B A800m Instruct on RTX Spark report
DGX Spark8-bit128K146 tok/s0.2 sPDF: Granite 3.1 3B A800m Instruct on DGX Spark report
Dual DGX Spark8-bit128K146 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on Dual DGX Spark report
DGX Station GB3008-bit128K474 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit128K375 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit128K375 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit128K375 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit128K388 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on A100 80 GB SXM report
H100 SXM8-bit128K431 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on H100 SXM report
4× H200 NVL8-bit128K455 tok/s0.1 sPDF: Granite 3.1 3B A800m Instruct on 4× H200 NVL report
8× H200 NVL8-bit128K455 tok/s0.1 sPDF: Granite 3.1 3B A800m 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-3.1-3b-a800m-instruct-GGUF
vLLM
vllm serve ibm-granite/granite-3.1-3b-a800m-instruct

Questions

How much VRAM does Granite 3.1 3B A800m Instruct need?

Granite 3.1 3B A800m Instruct needs 1.9 GB for its weights at 4-bit, 3.0 GB with an 8K context and 4.5 GB with 32K. At 8-bit, 4.4 GB.

What is the minimum VRAM for Granite 3.1 3B A800m Instruct?

About 2.6 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 3.1 3B A800m Instruct?

3.3B parameters (883M active per token): 6.1 GB of weights at 16-bit, 1.9 GB at 4-bit.

Can Granite 3.1 3B A800m Instruct run on an RTX 4090?

Yes. At 8-bit it holds 128K of context and writes about 308 tokens a second.

More models