Skip to content

Ornith 1.0 397B VRAM requirements

How much VRAM Ornith 1.0 397B needs
4-bit weights: 224 GBWith 8K context: 226 GBWith 32K context: 226 GB8-bit, 8K context: 394 GB
Parameters
397B
Active per token
17.8B
Longest context
256K
Released
Jun 2026
License
MIT
Strongest at
Not rated

How much VRAM Ornith 1.0 397B needs

Memory Ornith 1.0 397B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit739 GB741 GB741 GB744 GB
8-bit393 GB394 GB395 GB398 GB
6-bit303 GB304 GB305 GB308 GB
5-bit263 GB264 GB265 GB268 GB
4-bit224 GB226 GB226 GB229 GB
3-bit181 GB182 GB183 GB186 GB
2-bit139 GB140 GB141 GB144 GB
1.6-bit76.2 GB77.7 GB78.4 GB81.2 GB

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

GPUs and Macs that run Ornith 1.0 397B

The best way Ornith 1.0 397B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Ornith 1.0 397B on RTX 4060 report
RTX 5060Too bigPDF: Ornith 1.0 397B on RTX 5060 report
RTX 3070Too bigPDF: Ornith 1.0 397B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Ornith 1.0 397B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Ornith 1.0 397B on RTX 3060 12 GB report
Arc B580Too bigPDF: Ornith 1.0 397B on Arc B580 report
RTX 4070Too bigPDF: Ornith 1.0 397B on RTX 4070 report
RTX 4070 SuperToo bigPDF: Ornith 1.0 397B on RTX 4070 Super report
RTX 5070Too bigPDF: Ornith 1.0 397B on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Ornith 1.0 397B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Ornith 1.0 397B on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Ornith 1.0 397B on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Ornith 1.0 397B on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Ornith 1.0 397B on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Ornith 1.0 397B on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Ornith 1.0 397B on RTX 5070 Ti report
RTX 5080Too bigPDF: Ornith 1.0 397B on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Ornith 1.0 397B on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Ornith 1.0 397B on RTX 3090 report
RTX 4090Too bigPDF: Ornith 1.0 397B on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Ornith 1.0 397B on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Ornith 1.0 397B on RTX 5090 report
2× RTX 3090Too bigPDF: Ornith 1.0 397B on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Ornith 1.0 397B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Ornith 1.0 397B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Ornith 1.0 397B on RTX 6000 Ada report
L40SToo bigPDF: Ornith 1.0 397B on L40S report
2× RTX 5090Too bigPDF: Ornith 1.0 397B on 2× RTX 5090 report
4× RTX 3090Too bigPDF: Ornith 1.0 397B on 4× RTX 3090 report
RTX PRO 6000 BlackwellToo bigPDF: Ornith 1.0 397B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Ornith 1.0 397B on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Ornith 1.0 397B on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Ornith 1.0 397B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Ornith 1.0 397B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Ornith 1.0 397B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Ornith 1.0 397B on Mac mini M4 Pro 64 GB report
M4 Max 128 GBToo bigPDF: Ornith 1.0 397B on M4 Max 128 GB report
M5 Max 128 GBToo bigPDF: Ornith 1.0 397B on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Ornith 1.0 397B on M3 Ultra 96 GB report
M3 Ultra 256 GB3-bit24K40 tok/s10 sPDF: Ornith 1.0 397B on M3 Ultra 256 GB report
M3 Ultra 512 GB6-bit256K28 tok/s10 sPDF: Ornith 1.0 397B on M3 Ultra 512 GB report
M5 Ultra 512 GB6-bit256K37 tok/s3.3 sPDF: Ornith 1.0 397B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Ornith 1.0 397B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBToo bigPDF: Ornith 1.0 397B on Ryzen AI Max+ 395 128 GB report
RTX SparkToo bigPDF: Ornith 1.0 397B on RTX Spark report
DGX SparkToo bigPDF: Ornith 1.0 397B on DGX Spark report
Dual DGX Spark4-bit96K16 tok/s2.9 sPDF: Ornith 1.0 397B on Dual DGX Spark report
DGX Station GB3004-bit256K172 tok/s0.3 sPDF: Ornith 1.0 397B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell2-bit256K110 tok/s0.5 sPDF: Ornith 1.0 397B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell6-bit256K65 tok/s0.3 sPDF: Ornith 1.0 397B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K53 tok/s0.2 sPDF: Ornith 1.0 397B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Ornith 1.0 397B on A100 80 GB SXM report
H100 SXMToo bigPDF: Ornith 1.0 397B on H100 SXM report
4× H200 NVL8-bit256K108 tok/s0.3 sPDF: Ornith 1.0 397B on 4× H200 NVL report
8× H200 NVL8-bit256K108 tok/s0.1 sPDF: Ornith 1.0 397B 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 unsloth/Ornith-1.0-397B-GGUF
vLLM
vllm serve ornith-ai/Ornith-1.0-397B

Questions

How much VRAM does Ornith 1.0 397B need?

Ornith 1.0 397B needs 224 GB for its weights at 4-bit, 226 GB with an 8K context and 226 GB with 32K. At 8-bit, 394 GB.

What is the minimum VRAM for Ornith 1.0 397B?

About 140 GB at 2-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 M3 Ultra 256 GB.

How big is Ornith 1.0 397B?

397B parameters (17.8B active per token): 739 GB of weights at 16-bit, 224 GB at 4-bit.

Can Ornith 1.0 397B run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the M3 Ultra 256 GB.

More models