Skip to content

ALMA 13B R VRAM requirements

How much VRAM ALMA 13B R needs
4-bit weights: 7.3 GBWith 4K context: 11.1 GB8-bit, 4K context: 16.7 GB
Parameters
13B
Longest context
4K
Released
Jan 2024
License
MIT
Strongest at
Not rated

How much VRAM ALMA 13B R needs

Memory ALMA 13B R needs at each precision and context length
PrecisionWeightsWith 4K
16-bit24.2 GB28.0 GB
8-bit12.9 GB16.7 GB
6-bit9.9 GB13.7 GB
5-bit8.6 GB12.4 GB
4-bit7.3 GB11.1 GB
3-bit5.9 GB9.7 GB
2-bit4.5 GB8.3 GB
1.6-bit2.5 GB6.3 GB

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

GPUs and Macs that run ALMA 13B R

The best way ALMA 13B R runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: ALMA 13B R on RTX 4060 report
RTX 5060Too bigPDF: ALMA 13B R on RTX 5060 report
RTX 3070Too bigPDF: ALMA 13B R on RTX 3070 report
RTX 3080 10 GBToo bigPDF: ALMA 13B R on RTX 3080 10 GB report
RTX 3060 12 GB4-bit4K31 tok/s2.2 sPDF: ALMA 13B R on RTX 3060 12 GB report
Arc B5804-bit4K27 tok/s1.5 sPDF: ALMA 13B R on Arc B580 report
RTX 40704-bit4K43 tok/s1.0 sPDF: ALMA 13B R on RTX 4070 report
RTX 4070 Super4-bit4K43 tok/s0.8 sPDF: ALMA 13B R on RTX 4070 Super report
RTX 50704-bit4K56 tok/s0.9 sPDF: ALMA 13B R on RTX 5070 report
RTX 4060 Ti 16 GB6-bit4K19 tok/s1.3 sPDF: ALMA 13B R on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB6-bit4K29 tok/s1.2 sPDF: ALMA 13B R on RTX 5060 Ti 16 GB report
Arc A770 16 GB6-bit4K25 tok/s1.4 sPDF: ALMA 13B R on Arc A770 16 GB report
Radeon RX 9070 XT6-bit4K34 tok/s0.8 sPDF: ALMA 13B R on Radeon RX 9070 XT report
RTX 4070 Ti Super6-bit4K42 tok/s0.6 sPDF: ALMA 13B R on RTX 4070 Ti Super report
RTX 4080 Super6-bit4K46 tok/s0.5 sPDF: ALMA 13B R on RTX 4080 Super report
RTX 5070 Ti6-bit4K56 tok/s0.6 sPDF: ALMA 13B R on RTX 5070 Ti report
RTX 50806-bit4K59 tok/s0.5 sPDF: ALMA 13B R on RTX 5080 report
Radeon RX 7900 XTX8-bit4K39 tok/s1.2 sPDF: ALMA 13B R on Radeon RX 7900 XTX report
RTX 30908-bit4K45 tok/s0.8 sPDF: ALMA 13B R on RTX 3090 report
RTX 40908-bit4K49 tok/s0.3 sPDF: ALMA 13B R on RTX 4090 report
Radeon AI PRO R97008-bit4K27 tok/s0.8 sPDF: ALMA 13B R on Radeon AI PRO R9700 report
RTX 50908-bit4K83 tok/s0.3 sPDF: ALMA 13B R on RTX 5090 report
2× RTX 30908-bit4K45 tok/s0.5 sPDF: ALMA 13B R on 2× RTX 3090 report
2× RTX 40908-bit4K49 tok/s0.2 sPDF: ALMA 13B R on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit4K64 tok/s0.4 sPDF: ALMA 13B R on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit4K46 tok/s0.3 sPDF: ALMA 13B R on RTX 6000 Ada report
L40S8-bit4K42 tok/s0.5 sPDF: ALMA 13B R on L40S report
2× RTX 50908-bit4K83 tok/s0.2 sPDF: ALMA 13B R on 2× RTX 5090 report
4× RTX 30908-bit4K45 tok/s0.2 sPDF: ALMA 13B R on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit4K83 tok/s0.2 sPDF: ALMA 13B R on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB3-bit4K15 tok/s18 sPDF: ALMA 13B R on Mac mini M4 16 GB report
Mac mini M4 24 GB6-bit4K9.1 tok/s18 sPDF: ALMA 13B R on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit4K7.0 tok/s18 sPDF: ALMA 13B R on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB6-bit4K20 tok/s9.0 sPDF: ALMA 13B R on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit4K16 tok/s9.0 sPDF: ALMA 13B R on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit4K16 tok/s9.0 sPDF: ALMA 13B R on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit4K27 tok/s4.5 sPDF: ALMA 13B R on M4 Max 128 GB report
M5 Max 128 GB8-bit4K30 tok/s1.3 sPDF: ALMA 13B R on M5 Max 128 GB report
M3 Ultra 96 GB8-bit4K35 tok/s2.5 sPDF: ALMA 13B R on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit4K35 tok/s2.5 sPDF: ALMA 13B R on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit4K35 tok/s2.5 sPDF: ALMA 13B R on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit4K48 tok/s0.8 sPDF: ALMA 13B R on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit4K11 tok/s2.6 sPDF: ALMA 13B R on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit4K11 tok/s2.6 sPDF: ALMA 13B R on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit4K15 tok/s0.9 sPDF: ALMA 13B R on RTX Spark report
DGX Spark8-bit4K14 tok/s0.9 sPDF: ALMA 13B R on DGX Spark report
Dual DGX Spark8-bit4K14 tok/s0.5 sPDF: ALMA 13B R on Dual DGX Spark report
DGX Station GB3008-bit4K253 tok/s0.1 sPDF: ALMA 13B R on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit4K83 tok/s0.1 sPDF: ALMA 13B R on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit4K83 tok/s0.1 sPDF: ALMA 13B R on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit4K83 tok/s0.1 sPDF: ALMA 13B R on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit4K93 tok/s0.6 sPDF: ALMA 13B R on A100 80 GB SXM report
H100 SXM8-bit4K142 tok/s0.2 sPDF: ALMA 13B R on H100 SXM report
4× H200 NVL8-bit4K190 tok/s0.1 sPDF: ALMA 13B R on 4× H200 NVL report
8× H200 NVL8-bit4K190 tok/s0.1 sPDF: ALMA 13B R 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/ALMA-13B-R-i1-GGUF
vLLM
vllm serve haoranxu/ALMA-13B-R

Questions

How much VRAM does ALMA 13B R need?

ALMA 13B R needs 7.3 GB for its weights at 4-bit, 11.1 GB with an 4K context and 11.1 GB with 4K. At 8-bit, 16.7 GB.

What is the minimum VRAM for ALMA 13B R?

About 9.7 GB at 3-bit with an 4K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 3060 12 GB.

How big is ALMA 13B R?

13B parameters: 24.2 GB of weights at 16-bit, 7.3 GB at 4-bit.

Can ALMA 13B R run on an RTX 4090?

Yes. At 8-bit it holds 4K of context and writes about 49 tokens a second.

More models