Artwork

Konten disediakan oleh PyTorch, Edward Yang, and Team PyTorch. Semua konten podcast termasuk episode, grafik, dan deskripsi podcast diunggah dan disediakan langsung oleh PyTorch, Edward Yang, and Team PyTorch atau mitra platform podcast mereka. Jika Anda yakin seseorang menggunakan karya berhak cipta Anda tanpa izin, Anda dapat mengikuti proses yang diuraikan di sini https://id.player.fm/legal.
Player FM - Aplikasi Podcast
Offline dengan aplikasi Player FM !

Asynchronous versus synchronous execution

15:03
 
Bagikan
 

Manage episode 298430836 series 2921809
Konten disediakan oleh PyTorch, Edward Yang, and Team PyTorch. Semua konten podcast termasuk episode, grafik, dan deskripsi podcast diunggah dan disediakan langsung oleh PyTorch, Edward Yang, and Team PyTorch atau mitra platform podcast mereka. Jika Anda yakin seseorang menggunakan karya berhak cipta Anda tanpa izin, Anda dapat mengikuti proses yang diuraikan di sini https://id.player.fm/legal.

CUDA is asynchronous, CPU is synchronous. Making them play well together can be one of the more thorny and easy to get wrong aspects of the PyTorch API. I talk about why non_blocking is difficult to use correctly, a hypothetical "asynchronous CPU" device which would help smooth over some of the API problems and also why it used to be difficult to implement async CPU (but it's not hard anymore!) At the end, I also briefly talk about how async/sync impedance can also show up in unusual places, namely the CUDA caching allocator.

Further reading.

  continue reading

83 episode

Artwork
iconBagikan
 
Manage episode 298430836 series 2921809
Konten disediakan oleh PyTorch, Edward Yang, and Team PyTorch. Semua konten podcast termasuk episode, grafik, dan deskripsi podcast diunggah dan disediakan langsung oleh PyTorch, Edward Yang, and Team PyTorch atau mitra platform podcast mereka. Jika Anda yakin seseorang menggunakan karya berhak cipta Anda tanpa izin, Anda dapat mengikuti proses yang diuraikan di sini https://id.player.fm/legal.

CUDA is asynchronous, CPU is synchronous. Making them play well together can be one of the more thorny and easy to get wrong aspects of the PyTorch API. I talk about why non_blocking is difficult to use correctly, a hypothetical "asynchronous CPU" device which would help smooth over some of the API problems and also why it used to be difficult to implement async CPU (but it's not hard anymore!) At the end, I also briefly talk about how async/sync impedance can also show up in unusual places, namely the CUDA caching allocator.

Further reading.

  continue reading

83 episode

Tất cả các tập

×
 
Loading …

Selamat datang di Player FM!

Player FM memindai web untuk mencari podcast berkualitas tinggi untuk Anda nikmati saat ini. Ini adalah aplikasi podcast terbaik dan bekerja untuk Android, iPhone, dan web. Daftar untuk menyinkronkan langganan di seluruh perangkat.

 

Panduan Referensi Cepat

Dengarkan acara ini sambil menjelajah
Putar