Skip to content

TW3 JRGL V2 VRAM requirements

How much VRAM TW3 JRGL V2 needs
4-bit weights: 40.8 GBWith 8K context: 61.8 GBWith 32K context: 122 GB8-bit, 8K context: 92.5 GB
Parameters
72.3B
Longest context
32K
Released
Apr 2024
License
Apache 2.0
Strongest at
Not rated

How much VRAM TW3 JRGL V2 needs

Memory TW3 JRGL V2 needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit135 GB156 GB216 GB
8-bit71.5 GB92.5 GB153 GB
6-bit55.2 GB76.2 GB136 GB
5-bit47.9 GB68.9 GB129 GB
4-bit40.8 GB61.8 GB122 GB
3-bit32.9 GB53.9 GB114 GB
2-bit25.2 GB46.2 GB106 GB
1.6-bit13.9 GB34.9 GB94.9 GB

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

GPUs and Macs that run TW3 JRGL V2

The best way TW3 JRGL V2 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: TW3 JRGL V2 on RTX 4060 report
RTX 5060Too bigPDF: TW3 JRGL V2 on RTX 5060 report
RTX 3070Too bigPDF: TW3 JRGL V2 on RTX 3070 report
RTX 3080 10 GBToo bigPDF: TW3 JRGL V2 on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: TW3 JRGL V2 on RTX 3060 12 GB report
Arc B580Too bigPDF: TW3 JRGL V2 on Arc B580 report
RTX 4070Too bigPDF: TW3 JRGL V2 on RTX 4070 report
RTX 4070 SuperToo bigPDF: TW3 JRGL V2 on RTX 4070 Super report
RTX 5070Too bigPDF: TW3 JRGL V2 on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: TW3 JRGL V2 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: TW3 JRGL V2 on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: TW3 JRGL V2 on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: TW3 JRGL V2 on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: TW3 JRGL V2 on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: TW3 JRGL V2 on RTX 4080 Super report
RTX 5070 TiToo bigPDF: TW3 JRGL V2 on RTX 5070 Ti report
RTX 5080Too bigPDF: TW3 JRGL V2 on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: TW3 JRGL V2 on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: TW3 JRGL V2 on RTX 3090 report
RTX 4090Too bigPDF: TW3 JRGL V2 on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: TW3 JRGL V2 on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: TW3 JRGL V2 on RTX 5090 report
2× RTX 3090Too bigPDF: TW3 JRGL V2 on 2× RTX 3090 report
2× RTX 4090Too bigPDF: TW3 JRGL V2 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: TW3 JRGL V2 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: TW3 JRGL V2 on RTX 6000 Ada report
L40SToo bigPDF: TW3 JRGL V2 on L40S report
2× RTX 5090Too bigPDF: TW3 JRGL V2 on 2× RTX 5090 report
4× RTX 30905-bit16K13 tok/s1.3 sPDF: TW3 JRGL V2 on 4× RTX 3090 report
RTX PRO 6000 Blackwell5-bit16K23 tok/s1.2 sPDF: TW3 JRGL V2 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: TW3 JRGL V2 on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: TW3 JRGL V2 on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: TW3 JRGL V2 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: TW3 JRGL V2 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: TW3 JRGL V2 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: TW3 JRGL V2 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB5-bit16K7.7 tok/s24 sPDF: TW3 JRGL V2 on M4 Max 128 GB report
M5 Max 128 GB5-bit16K8.6 tok/s7.0 sPDF: TW3 JRGL V2 on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: TW3 JRGL V2 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K6.7 tok/s14 sPDF: TW3 JRGL V2 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K6.7 tok/s14 sPDF: TW3 JRGL V2 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K9.7 tok/s4.3 sPDF: TW3 JRGL V2 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: TW3 JRGL V2 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB5-bit16K3.0 tok/s14 sPDF: TW3 JRGL V2 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit16K2.8 tok/s4.9 sPDF: TW3 JRGL V2 on RTX Spark report
DGX Spark8-bit16K2.5 tok/s4.9 sPDF: TW3 JRGL V2 on DGX Spark report
Dual DGX Spark8-bit32K2.5 tok/s2.8 sPDF: TW3 JRGL V2 on Dual DGX Spark report
DGX Station GB3008-bit32K57 tok/s0.4 sPDF: TW3 JRGL V2 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K16 tok/s0.7 sPDF: TW3 JRGL V2 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K16 tok/s0.4 sPDF: TW3 JRGL V2 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K16 tok/s0.2 sPDF: TW3 JRGL V2 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM3-bit16K37 tok/s3.1 sPDF: TW3 JRGL V2 on A100 80 GB SXM report
H100 SXM3-bit16K58 tok/s1.0 sPDF: TW3 JRGL V2 on H100 SXM report
4× H200 NVL8-bit32K40 tok/s0.4 sPDF: TW3 JRGL V2 on 4× H200 NVL report
8× H200 NVL8-bit32K40 tok/s0.2 sPDF: TW3 JRGL V2 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 mradermacher/TW3-JRGL-v2-i1-GGUF
vLLM
vllm serve paloalma/TW3-JRGL-v2

Questions

How much VRAM does TW3 JRGL V2 need?

TW3 JRGL V2 needs 40.8 GB for its weights at 4-bit, 61.8 GB with an 8K context and 122 GB with 32K. At 8-bit, 92.5 GB.

What is the minimum VRAM for TW3 JRGL V2?

About 53.9 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 4× RTX 3090.

How big is TW3 JRGL V2?

72.3B parameters: 135 GB of weights at 16-bit, 40.8 GB at 4-bit.

Can TW3 JRGL V2 run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the 4× RTX 3090.

More models