Skip to content

Yi 34B 200K VRAM requirements

How much VRAM Yi 34B 200K needs
4-bit weights: 19.4 GBWith 8K context: 22.1 GBWith 32K context: 27.7 GB8-bit, 8K context: 36.7 GB
Parameters
34.4B
Longest context
195K
Released
Nov 2023
License
Apache 2.0
Strongest at
Not rated

How much VRAM Yi 34B 200K needs

Memory Yi 34B 200K needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit64.1 GB66.7 GB72.3 GB94.8 GB
8-bit34.0 GB36.7 GB42.3 GB64.8 GB
6-bit26.3 GB28.9 GB34.6 GB57.1 GB
5-bit22.8 GB25.4 GB31.1 GB53.6 GB
4-bit19.4 GB22.1 GB27.7 GB50.2 GB
3-bit15.7 GB18.3 GB23.9 GB46.4 GB
2-bit12.0 GB14.7 GB20.3 GB42.8 GB
1.6-bit6.6 GB9.3 GB14.9 GB37.4 GB

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

GPUs and Macs that run Yi 34B 200K

The best way Yi 34B 200K runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Yi 34B 200K on RTX 4060 report
RTX 5060Too bigPDF: Yi 34B 200K on RTX 5060 report
RTX 3070Too bigPDF: Yi 34B 200K on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Yi 34B 200K on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Yi 34B 200K on RTX 3060 12 GB report
Arc B580Too bigPDF: Yi 34B 200K on Arc B580 report
RTX 4070Too bigPDF: Yi 34B 200K on RTX 4070 report
RTX 4070 SuperToo bigPDF: Yi 34B 200K on RTX 4070 Super report
RTX 5070Too bigPDF: Yi 34B 200K on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Yi 34B 200K on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Yi 34B 200K on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Yi 34B 200K on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Yi 34B 200K on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Yi 34B 200K on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Yi 34B 200K on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Yi 34B 200K on RTX 5070 Ti report
RTX 5080Too bigPDF: Yi 34B 200K on RTX 5080 report
Radeon RX 7900 XTX3-bit24K31 tok/s3.3 sPDF: Yi 34B 200K on Radeon RX 7900 XTX report
RTX 30903-bit24K37 tok/s2.1 sPDF: Yi 34B 200K on RTX 3090 report
RTX 40903-bit24K40 tok/s0.9 sPDF: Yi 34B 200K on RTX 4090 report
Radeon AI PRO R97005-bit28K15 tok/s2.1 sPDF: Yi 34B 200K on Radeon AI PRO R9700 report
RTX 50905-bit28K48 tok/s0.7 sPDF: Yi 34B 200K on RTX 5090 report
2× RTX 30908-bit40K18 tok/s1.2 sPDF: Yi 34B 200K on 2× RTX 3090 report
2× RTX 40908-bit40K19 tok/s0.5 sPDF: Yi 34B 200K on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit44K25 tok/s1.1 sPDF: Yi 34B 200K on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit44K18 tok/s0.8 sPDF: Yi 34B 200K on RTX 6000 Ada report
L40S8-bit44K16 tok/s1.3 sPDF: Yi 34B 200K on L40S report
2× RTX 50908-bit104K33 tok/s0.4 sPDF: Yi 34B 200K on 2× RTX 5090 report
4× RTX 30908-bit195K18 tok/s0.6 sPDF: Yi 34B 200K on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit195K33 tok/s0.6 sPDF: Yi 34B 200K on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Yi 34B 200K on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Yi 34B 200K on Mac mini M4 24 GB report
Mac mini M4 32 GB3-bit16K5.8 tok/s47 sPDF: Yi 34B 200K on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Yi 34B 200K on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB6-bit28K7.7 tok/s24 sPDF: Yi 34B 200K on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit44K6.0 tok/s24 sPDF: Yi 34B 200K on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit195K11 tok/s12 sPDF: Yi 34B 200K on M4 Max 128 GB report
M5 Max 128 GB8-bit195K12 tok/s3.4 sPDF: Yi 34B 200K on M5 Max 128 GB report
M3 Ultra 96 GB8-bit136K14 tok/s6.6 sPDF: Yi 34B 200K on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit195K14 tok/s6.6 sPDF: Yi 34B 200K on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit195K14 tok/s6.6 sPDF: Yi 34B 200K on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit195K20 tok/s2.1 sPDF: Yi 34B 200K on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit44K4.2 tok/s6.8 sPDF: Yi 34B 200K on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit195K4.2 tok/s6.8 sPDF: Yi 34B 200K on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit195K5.8 tok/s2.4 sPDF: Yi 34B 200K on RTX Spark report
DGX Spark8-bit195K5.3 tok/s2.4 sPDF: Yi 34B 200K on DGX Spark report
Dual DGX Spark8-bit195K5.3 tok/s1.4 sPDF: Yi 34B 200K on Dual DGX Spark report
DGX Station GB3008-bit195K110 tok/s0.2 sPDF: Yi 34B 200K on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit195K33 tok/s0.3 sPDF: Yi 34B 200K on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit195K33 tok/s0.2 sPDF: Yi 34B 200K on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit195K33 tok/s0.1 sPDF: Yi 34B 200K on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit168K37 tok/s1.5 sPDF: Yi 34B 200K on A100 80 GB SXM report
H100 SXM8-bit168K58 tok/s0.5 sPDF: Yi 34B 200K on H100 SXM report
4× H200 NVL8-bit195K80 tok/s0.2 sPDF: Yi 34B 200K on 4× H200 NVL report
8× H200 NVL8-bit195K80 tok/s0.1 sPDF: Yi 34B 200K 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 TheBloke/Yi-34B-200K-GGUF
vLLM
vllm serve 01-ai/Yi-34B-200K

Questions

How much VRAM does Yi 34B 200K need?

Yi 34B 200K needs 19.4 GB for its weights at 4-bit, 22.1 GB with an 8K context and 27.7 GB with 32K. At 8-bit, 36.7 GB.

What is the minimum VRAM for Yi 34B 200K?

About 18.3 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 Radeon RX 7900 XTX.

How big is Yi 34B 200K?

34.4B parameters: 64.1 GB of weights at 16-bit, 19.4 GB at 4-bit.

Can Yi 34B 200K run on an RTX 4090?

Yes. At 3-bit it holds 24K of context and writes about 40 tokens a second.

More models