Skip to content

Granite 4.2 8B VRAM requirements

How much VRAM Granite 4.2 8B needs
4-bit weights: 5.0 GBWith 8K context: 7.0 GBWith 32K context: 10.7 GB8-bit, 8K context: 10.7 GB
Parameters
8.8B
Longest context
128K
Released
Aug 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM Granite 4.2 8B needs

Memory Granite 4.2 8B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit16.4 GB18.4 GB22.1 GB37.1 GB
8-bit8.7 GB10.7 GB14.5 GB29.5 GB
6-bit6.7 GB8.7 GB12.5 GB27.5 GB
5-bit5.8 GB7.8 GB11.6 GB26.6 GB
4-bit5.0 GB7.0 GB10.7 GB25.7 GB
3-bit4.0 GB6.0 GB9.8 GB24.8 GB
2-bit3.1 GB5.1 GB8.8 GB23.8 GB
1.6-bit1.7 GB3.7 GB7.5 GB22.5 GB

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

GPUs and Macs that run Granite 4.2 8B

The best way Granite 4.2 8B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40603-bit16K44 tok/s1.2 sPDF: Granite 4.2 8B on RTX 4060 report
RTX 50603-bit16K70 tok/s0.9 sPDF: Granite 4.2 8B on RTX 5060 report
RTX 30703-bit16K70 tok/s0.9 sPDF: Granite 4.2 8B on RTX 3070 report
RTX 3080 10 GB5-bit16K81 tok/s0.6 sPDF: Granite 4.2 8B on RTX 3080 10 GB report
RTX 3060 12 GB6-bit24K35 tok/s1.4 sPDF: Granite 4.2 8B on RTX 3060 12 GB report
Arc B5806-bit24K31 tok/s1.0 sPDF: Granite 4.2 8B on Arc B580 report
RTX 40706-bit24K48 tok/s0.6 sPDF: Granite 4.2 8B on RTX 4070 report
RTX 4070 Super6-bit24K48 tok/s0.5 sPDF: Granite 4.2 8B on RTX 4070 Super report
RTX 50706-bit24K63 tok/s0.6 sPDF: Granite 4.2 8B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit36K22 tok/s0.8 sPDF: Granite 4.2 8B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit36K34 tok/s0.8 sPDF: Granite 4.2 8B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit36K29 tok/s0.9 sPDF: Granite 4.2 8B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit36K40 tok/s0.5 sPDF: Granite 4.2 8B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit36K50 tok/s0.4 sPDF: Granite 4.2 8B on RTX 4070 Ti Super report
RTX 4080 Super8-bit36K54 tok/s0.3 sPDF: Granite 4.2 8B on RTX 4080 Super report
RTX 5070 Ti8-bit36K65 tok/s0.4 sPDF: Granite 4.2 8B on RTX 5070 Ti report
RTX 50808-bit36K69 tok/s0.3 sPDF: Granite 4.2 8B on RTX 5080 report
Radeon RX 7900 XTX8-bit80K57 tok/s0.8 sPDF: Granite 4.2 8B on Radeon RX 7900 XTX report
RTX 30908-bit80K68 tok/s0.5 sPDF: Granite 4.2 8B on RTX 3090 report
RTX 40908-bit80K72 tok/s0.2 sPDF: Granite 4.2 8B on RTX 4090 report
Radeon AI PRO R97008-bit128K40 tok/s0.5 sPDF: Granite 4.2 8B on Radeon AI PRO R9700 report
RTX 50908-bit128K121 tok/s0.2 sPDF: Granite 4.2 8B on RTX 5090 report
2× RTX 30908-bit128K68 tok/s0.3 sPDF: Granite 4.2 8B on 2× RTX 3090 report
2× RTX 40908-bit128K72 tok/s0.1 sPDF: Granite 4.2 8B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit128K94 tok/s0.3 sPDF: Granite 4.2 8B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit128K69 tok/s0.2 sPDF: Granite 4.2 8B on RTX 6000 Ada report
L40S8-bit128K63 tok/s0.3 sPDF: Granite 4.2 8B on L40S report
2× RTX 50908-bit128K121 tok/s0.1 sPDF: Granite 4.2 8B on 2× RTX 5090 report
4× RTX 30908-bit128K68 tok/s0.2 sPDF: Granite 4.2 8B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit128K121 tok/s0.1 sPDF: Granite 4.2 8B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB6-bit16K14 tok/s11 sPDF: Granite 4.2 8B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit36K11 tok/s11 sPDF: Granite 4.2 8B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit64K11 tok/s11 sPDF: Granite 4.2 8B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit36K23 tok/s5.7 sPDF: Granite 4.2 8B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit128K23 tok/s5.7 sPDF: Granite 4.2 8B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit128K23 tok/s5.7 sPDF: Granite 4.2 8B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit128K40 tok/s2.8 sPDF: Granite 4.2 8B on M4 Max 128 GB report
M5 Max 128 GB8-bit128K44 tok/s0.8 sPDF: Granite 4.2 8B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit128K50 tok/s1.6 sPDF: Granite 4.2 8B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit128K50 tok/s1.6 sPDF: Granite 4.2 8B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit128K50 tok/s1.6 sPDF: Granite 4.2 8B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit128K68 tok/s0.5 sPDF: Granite 4.2 8B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit128K17 tok/s1.6 sPDF: Granite 4.2 8B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit128K17 tok/s1.6 sPDF: Granite 4.2 8B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit128K23 tok/s0.6 sPDF: Granite 4.2 8B on RTX Spark report
DGX Spark8-bit128K21 tok/s0.6 sPDF: Granite 4.2 8B on DGX Spark report
Dual DGX Spark8-bit128K21 tok/s0.3 sPDF: Granite 4.2 8B on Dual DGX Spark report
DGX Station GB3008-bit128K334 tok/s0.1 sPDF: Granite 4.2 8B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit128K121 tok/s0.1 sPDF: Granite 4.2 8B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit128K121 tok/s0.1 sPDF: Granite 4.2 8B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit128K121 tok/s0.1 sPDF: Granite 4.2 8B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit128K134 tok/s0.4 sPDF: Granite 4.2 8B on A100 80 GB SXM report
H100 SXM8-bit128K200 tok/s0.1 sPDF: Granite 4.2 8B on H100 SXM report
4× H200 NVL8-bit128K260 tok/s0.1 sPDF: Granite 4.2 8B on 4× H200 NVL report
8× H200 NVL8-bit128K260 tok/s0.1 sPDF: Granite 4.2 8B 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-4.2-8b-GGUF
vLLM
vllm serve ibm-granite/granite-4.2-8b

Questions

How much VRAM does Granite 4.2 8B need?

Granite 4.2 8B needs 5.0 GB for its weights at 4-bit, 7.0 GB with an 8K context and 10.7 GB with 32K. At 8-bit, 10.7 GB.

What is the minimum VRAM for Granite 4.2 8B?

About 6.0 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 4.2 8B?

8.8B parameters: 16.4 GB of weights at 16-bit, 5.0 GB at 4-bit.

Can Granite 4.2 8B run on an RTX 4090?

Yes. At 8-bit it holds 80K of context and writes about 72 tokens a second.

More models