search

PointsCoder / ONCE_Benchmark

One Million Scenes for Autonomous Driving
177 stars 32 forks source link

About training hardware setting #7

Closed qiaozhijian closed 3 years ago

qiaozhijian commented 3 years ago

Hello. What's the type and number of GPUs when training? Thank you.

hengseuer commented 3 years ago

Hi, you can refer to the paper(https://arxiv.org/abs/2106.11037).

qiaozhijian commented 3 years ago

Hi, you can refer to the paper(https://arxiv.org/abs/2106.11037).

Could you please point it out in paper? Which section and paragraph?

qiaozhijian commented 3 years ago

Hi, you can refer to the paper(https://arxiv.org/abs/2106.11037).

I tried, but failed.

hengseuer commented 3 years ago

Hi, you can refer to the paper(https://arxiv.org/abs/2106.11037).

Could you please point it out in paper? Which section and paragraph?

I am sorry. I misread. According to the total batch size(32) and batch_size_per_gpu(4), I think the number of GPUs is 8.

qiaozhijian commented 3 years ago

Hi, you can refer to the paper(https://arxiv.org/abs/2106.11037).

Could you please point it out in paper? Which section and paragraph?

I am sorry. I misread. According to the total batch size(32) and batch_size_per_gpu(4), I think the number of GPUs is 8.

Nevermind. I'm wondering which type of GPU can load 5 batchs(1 labeled and 4 unlabeled).

PointsCoder commented 3 years ago

@qiaozhijian We use 8 Tesla V100 (32G). For ssl-training, batch_size 5 will take about 10G GPU memory.

qiaozhijian commented 3 years ago

Thank you for your reply.