Skip to content

Llama3 ChatQA 1.5 8B VRAM requirements

How much VRAM Llama3 ChatQA 1.5 8B needs
4-bit weights: 4.5 GBWith 8K context: 6.4 GB8-bit, 8K context: 9.8 GB
Parameters
8B
Longest context
8K
Released
Apr 2024
License
Llama 3 Community
Strongest at
Not rated

Its licence (Llama 3 Community) is not a standard open-source one: check the terms before commercial use.

How much VRAM Llama3 ChatQA 1.5 8B needs

Memory Llama3 ChatQA 1.5 8B needs at each precision and context length
PrecisionWeightsWith 8K
16-bit15.0 GB16.8 GB
8-bit7.9 GB9.8 GB
6-bit6.1 GB8.0 GB
5-bit5.3 GB7.1 GB
4-bit4.5 GB6.4 GB
3-bit3.7 GB5.5 GB
2-bit2.8 GB4.6 GB
1.6-bit1.5 GB3.4 GB

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

GPUs and Macs that run Llama3 ChatQA 1.5 8B

The best way Llama3 ChatQA 1.5 8B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40605-bit8K34 tok/s1.0 sPDF: Llama3 ChatQA 1.5 8B on RTX 4060 report
RTX 50605-bit8K56 tok/s0.8 sPDF: Llama3 ChatQA 1.5 8B on RTX 5060 report
RTX 30705-bit8K56 tok/s0.8 sPDF: Llama3 ChatQA 1.5 8B on RTX 3070 report
RTX 3080 10 GB6-bit8K80 tok/s0.5 sPDF: Llama3 ChatQA 1.5 8B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit8K31 tok/s1.2 sPDF: Llama3 ChatQA 1.5 8B on RTX 3060 12 GB report
Arc B5808-bit8K27 tok/s0.8 sPDF: Llama3 ChatQA 1.5 8B on Arc B580 report
RTX 40708-bit8K42 tok/s0.5 sPDF: Llama3 ChatQA 1.5 8B on RTX 4070 report
RTX 4070 Super8-bit8K42 tok/s0.4 sPDF: Llama3 ChatQA 1.5 8B on RTX 4070 Super report
RTX 50708-bit8K56 tok/s0.5 sPDF: Llama3 ChatQA 1.5 8B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit8K25 tok/s0.7 sPDF: Llama3 ChatQA 1.5 8B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit8K38 tok/s0.7 sPDF: Llama3 ChatQA 1.5 8B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit8K33 tok/s0.8 sPDF: Llama3 ChatQA 1.5 8B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit8K45 tok/s0.4 sPDF: Llama3 ChatQA 1.5 8B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit8K56 tok/s0.4 sPDF: Llama3 ChatQA 1.5 8B on RTX 4070 Ti Super report
RTX 4080 Super8-bit8K61 tok/s0.3 sPDF: Llama3 ChatQA 1.5 8B on RTX 4080 Super report
RTX 5070 Ti8-bit8K73 tok/s0.4 sPDF: Llama3 ChatQA 1.5 8B on RTX 5070 Ti report
RTX 50808-bit8K78 tok/s0.3 sPDF: Llama3 ChatQA 1.5 8B on RTX 5080 report
Radeon RX 7900 XTX8-bit8K65 tok/s0.7 sPDF: Llama3 ChatQA 1.5 8B on Radeon RX 7900 XTX report
RTX 30908-bit8K76 tok/s0.4 sPDF: Llama3 ChatQA 1.5 8B on RTX 3090 report
RTX 40908-bit8K82 tok/s0.2 sPDF: Llama3 ChatQA 1.5 8B on RTX 4090 report
Radeon AI PRO R97008-bit8K45 tok/s0.4 sPDF: Llama3 ChatQA 1.5 8B on Radeon AI PRO R9700 report
RTX 50908-bit8K137 tok/s0.1 sPDF: Llama3 ChatQA 1.5 8B on RTX 5090 report
2× RTX 30908-bit8K76 tok/s0.3 sPDF: Llama3 ChatQA 1.5 8B on 2× RTX 3090 report
2× RTX 40908-bit8K82 tok/s0.1 sPDF: Llama3 ChatQA 1.5 8B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit8K106 tok/s0.2 sPDF: Llama3 ChatQA 1.5 8B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit8K78 tok/s0.2 sPDF: Llama3 ChatQA 1.5 8B on RTX 6000 Ada report
L40S8-bit8K71 tok/s0.3 sPDF: Llama3 ChatQA 1.5 8B on L40S report
2× RTX 50908-bit8K137 tok/s0.1 sPDF: Llama3 ChatQA 1.5 8B on 2× RTX 5090 report
4× RTX 30908-bit8K76 tok/s0.1 sPDF: Llama3 ChatQA 1.5 8B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit8K137 tok/s0.1 sPDF: Llama3 ChatQA 1.5 8B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit8K12 tok/s9.9 sPDF: Llama3 ChatQA 1.5 8B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit8K12 tok/s9.9 sPDF: Llama3 ChatQA 1.5 8B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit8K12 tok/s9.9 sPDF: Llama3 ChatQA 1.5 8B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit8K26 tok/s5.0 sPDF: Llama3 ChatQA 1.5 8B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit8K26 tok/s5.0 sPDF: Llama3 ChatQA 1.5 8B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit8K26 tok/s5.0 sPDF: Llama3 ChatQA 1.5 8B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit8K45 tok/s2.5 sPDF: Llama3 ChatQA 1.5 8B on M4 Max 128 GB report
M5 Max 128 GB8-bit8K50 tok/s0.7 sPDF: Llama3 ChatQA 1.5 8B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit8K57 tok/s1.4 sPDF: Llama3 ChatQA 1.5 8B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit8K57 tok/s1.4 sPDF: Llama3 ChatQA 1.5 8B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit8K57 tok/s1.4 sPDF: Llama3 ChatQA 1.5 8B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit8K78 tok/s0.4 sPDF: Llama3 ChatQA 1.5 8B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit8K19 tok/s1.4 sPDF: Llama3 ChatQA 1.5 8B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit8K19 tok/s1.4 sPDF: Llama3 ChatQA 1.5 8B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit8K26 tok/s0.5 sPDF: Llama3 ChatQA 1.5 8B on RTX Spark report
DGX Spark8-bit8K23 tok/s0.5 sPDF: Llama3 ChatQA 1.5 8B on DGX Spark report
Dual DGX Spark8-bit8K23 tok/s0.3 sPDF: Llama3 ChatQA 1.5 8B on Dual DGX Spark report
DGX Station GB3008-bit8K390 tok/s0.1 sPDF: Llama3 ChatQA 1.5 8B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit8K137 tok/s0.1 sPDF: Llama3 ChatQA 1.5 8B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit8K137 tok/s0.1 sPDF: Llama3 ChatQA 1.5 8B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit8K137 tok/s0.1 sPDF: Llama3 ChatQA 1.5 8B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit8K153 tok/s0.3 sPDF: Llama3 ChatQA 1.5 8B on A100 80 GB SXM report
H100 SXM8-bit8K229 tok/s0.1 sPDF: Llama3 ChatQA 1.5 8B on H100 SXM report
4× H200 NVL8-bit8K300 tok/s0.1 sPDF: Llama3 ChatQA 1.5 8B on 4× H200 NVL report
8× H200 NVL8-bit8K300 tok/s0.1 sPDF: Llama3 ChatQA 1.5 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 legraphista/Llama3-ChatQA-1.5-8B-IMat-GGUF
vLLM
vllm serve nvidia/Llama3-ChatQA-1.5-8B

Questions

How much VRAM does Llama3 ChatQA 1.5 8B need?

Llama3 ChatQA 1.5 8B needs 4.5 GB for its weights at 4-bit, 6.4 GB with an 8K context and 6.4 GB with 8K. At 8-bit, 9.8 GB.

What is the minimum VRAM for Llama3 ChatQA 1.5 8B?

About 5.5 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 Llama3 ChatQA 1.5 8B?

8B parameters: 15.0 GB of weights at 16-bit, 4.5 GB at 4-bit.

Can Llama3 ChatQA 1.5 8B run on an RTX 4090?

Yes. At 8-bit it holds 8K of context and writes about 82 tokens a second.

More models