selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

tencent logo

Can the NVIDIA A100 80GB run Hy-SimulMT-30B-A3B?

Hy-SimulMT-30B-A3B·30.4B (3B active)·NVIDIA A100 80GB·80 GB VRAM

Saved in this browser, used on every model page.
Yes

Yes. The NVIDIA A100 80GB runs Hy-SimulMT-30B-A3B fully in VRAM at FP16 / BF16 (about 68 GB of 80 GB).

best quant
FP16 / BF16
needs
68 GB
usable
80 GB

Estimates from the model's size.Try other contexts on the Hy-SimulMT-30B-A3B page →

Every quantisation of Hy-SimulMT-30B-A3B on the NVIDIA A100 80GB

QuantMemoryFit
Q4_K_M 21 GBfits
Q8_0 37 GBfits
FP16 / BF16 pick68 GBfits

Speed is a bandwidth-based estimate for generating text (2039 GB/s); prompt processing and long contexts are slower.

Run Hy-SimulMT-30B-A3B on the NVIDIA A100 80GB

Transformers · Run with the translation pipelinedocs
from transformers import pipeline

pipe = pipeline("translation", model="tencent/Hy-SimulMT-30B-A3B")
result = pipe("Self-hosting puts you in control of your data.")
print(result)

Translation models that fit the NVIDIA A100 80GB comfortably

All that fit →
ModelParamsBest quantMemoryLicence
30.1BA3BFP16 / BF16≈68 GBFits in memoryCommercial OK
31.8BFP16 / BF16≈71 GBFits in memoryConditions
facebook logonllb-moe-54bfacebook
54BQ8_0≈65 GBFits in memoryNon-commercial
9.7BFP16 / BF16≈22 GBFits in memoryCommercial OK
tencent logoHy-MT2-7Btencent
8BFP16 / BF16≈18 GBFits in memoryCommercial OK
8BFP16 / BF16≈18 GBFits in memoryCheck licence

Other GPUs that run Hy-SimulMT-30B-A3B

Frequently asked questions

Can the NVIDIA A100 80GB run Hy-SimulMT-30B-A3B?
Yes. The NVIDIA A100 80GB runs Hy-SimulMT-30B-A3B fully in VRAM at FP16 / BF16 (about 68 GB of 80 GB).
Which quantisation of Hy-SimulMT-30B-A3B should I use on the NVIDIA A100 80GB?
FP16 / BF16: the largest one that fits. Expected quality: lossless compared with the full-precision weights.

Memory and speed are estimates from the model's size and the card's memory bandwidth; real numbers depend on the runtime, drivers and context length. Community figures are medians of user reports.