Upgrade to Pro — share decks privately, control downloads, hide ads and more …

How we use GPUs in Cookpad

7dc8611c26c3ca62c551109c65d04270?s=47 Yuichiro Someya
November 06, 2017
Programming
0
84

How we use GPUs in Cookpad

@Tokyo Machine Learning Kitchen
https://tokyo-ml.github.io/

7dc8611c26c3ca62c551109c65d04270?s=128

Yuichiro Someya

November 06, 2017
Tweet

More Decks by Yuichiro Someya

See All by Yuichiro Someya
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
0
6.8k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
3
1.6k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
1
680
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
0
2.6k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
15
4.6k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
1
280
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
15
12k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
1
2.5k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
0
320

Other Decks in Programming

See All in Programming
746c5d84a717cf7f74822d248e06574e?s=48 nkmrh
3
710
583e920a7e9238a1c21e923025f8f641?s=48 elainenaomi
4
160
F301c0275838d562c72ea9177e7ddc68?s=48 fkino
1
140
9bda8755a689204677a198ee85c0eb07?s=48 banjun
PRO
0
160
Bf04d1ee6fa2216133b23af8f609c70a?s=48 gatchan0807
0
270
4fd283672253efa184ec8f63fb1a8d02?s=48 hexarf
2
220
2e72046813bf3dc01173fe54919a6d1d?s=48 cc4966
3
240
3931098ae486b6df619050c63e24f6cc?s=48 osyo
0
760
23a9ea59992e5349aa4bf659a6d583d3?s=48 nanashiki
1
510
Cc8ec9594b83b5dc3c7eef345c05dc8d?s=48 shiz
6
1k
3c031541ed3d92869414857dfef853de?s=48 koher
7
660
82d6167c4d14393c2e20b37a74b363c5?s=48 kasacchiful
0
230

Featured

See All Featured
761be20b5ebd271008bcee1244fc5b52?s=48 matthewcrist
73
7.7k
6804f1775cb4babfcc3851298566fbce?s=48 tanoku
260
24k
465724d73fe3a92c0879fdfb43a3a6f3?s=48 philhawksworth
195
9k
A60068bce2e73de3a37ca9d2dbe36092?s=48 bryan
105
11k
7fa6101cd43067653cf4ac6c7ca54305?s=48 akmur
255
20k
462dad0dd24c568a32dcdb0ae895ae1b?s=48 rasmusluckow
318
19k
Dad095ea7038f89f760419ce475d5d14?s=48 michaelherold
227
8.9k
864a009ac73600c7c02e1f26629dd989?s=48 paigeccino
0
95
3ab1249be442027903e1180025340b3f?s=48 reverentgeek
34
2.4k
812e1705ff0f8bc1bcf18587bde687d5?s=48 62gerente
593
210k
B3d6434763caa0ef5dc4b792662c49f7?s=48 smashingmag
287
47k
Ef32e0b1ac5082450dc8c1fca3a26b5c?s=48 cherdarchuk
77
270k

Transcript

  1. )PXXFVTF(16TJO$PPLQBE :VJDIJSP4PNFZB!$PPLQBE*OD3%

  2. ‣ Yuichiro Someya (ayemos) ‣ github.com/ayemos ‣ Machine Learning Enginner

    @ Cookpad Inc. # 2016(new grads) ~ Current
  3. None

  4. ‣ 0VS(16FOWJSPONFOU  )PXXFVUJMJ[F"84T(16JOTUBODFT  )PXXFLFFQPVSTDBMBCJMJUZPGUFBNTJO3% 

  5. 

  6.  /7*%*"7

  7.  All-in on AWS since 2011

  8.  All-in on AWS since 2011 Amazon RDS (Relational Data)

    Amazon Redshift (Data Warehouse)

  9.  All-in on AWS since 2011 Amazon S3 (Object Storage)

    Amazon RDS (Relational Data) Amazon Redshift (Data Warehouse)

  10.  All-in on AWS since 2011 Amazon S3 (Object Storage)

    Amazon RDS (Relational Data) Amazon Redshift (Data Warehouse) 7JSUVBM1SJWBUF$MPVE

  11. 7JSUVBM1SJWBUF$MPVE  All-in on AWS since 2011 Amazon S3 (Object

    Storage) Amazon RDS (Relational Data) Amazon Redshift (Data Warehouse) Amazon EC2 (Computation)

  12.  ‣ $6%" ‣ DV%//

  13.  ‣ $6%" ‣ DV%// (Snapshot)

  14.  ‣ $6%" ‣ DV%// (Snapshot) ‣ $6%" ‣ DV%//

  15.  ‣ $6%" ‣ DV%// (Snapshot) ‣ $6%" ‣ DV%//

    CUDA9 cuDNN7 CUDA8 cuDNN7 CUDA8 cuDNN6

  16.  ‣ $6%" ‣ DV%// (Snapshot) ‣ $6%" ‣ DV%//

    CUDA9 cuDNN7 CUDA8 cuDNN7 CUDA8 cuDNN6

  17. 5FNQMBUF  CUDA9 cuDNN7 CUDA8 cuDNN7 CUDA8 cuDNN6 5FNQMBUF KTPO

    QBDLFSCVJME

  18.  ‣ $6%" ‣ DV%// (Snapshot) ‣ $6%" ‣ DV%//

    CUDA9 cuDNN7 CUDA8 cuDNN7 CUDA8 cuDNN6

  19.  ‣ $6%" ‣ DV%// (Snapshot) ‣ $6%" ‣ DV%//

    IUUQTBXTBNB[PODPNBNB[POBJBNJT CUDA9 cuDNN7 CUDA8 cuDNN7 CUDA8 cuDNN6

  20.  CUDA9 cuDNN7 CUDA8 cuDNN7 CUDA8 cuDNN6 ... `ssh` ...

  21.  CUDA9 cuDNN7 CUDA8 cuDNN7 CUDA8 cuDNN6 ... `ssh` ...

  22.  CUDA9 cuDNN7 CUDA8 cuDNN7 CUDA8 cuDNN6 ... `ssh` ...

    AWS Lambda (Function) Stop! Idle? (Hourly)

  23.  CUDA9 cuDNN7 CUDA8 cuDNN7 CUDA8 cuDNN6 ... `ssh` ...

    AWS Lambda (Function) Stop! Idle? (Hourly)

  24. ‣ 0OEFNBOE(16XPSLCFODIFT  6UJMJ[F".*UPNVMUJQMFXPSLCFODIFOWJSPONFOUT  1BDLFSNBLFTJUFBTJFSUPVQEBUFBOENPSFTUBCMF  0QFSBUFWJB$IBUCPU  8SBQVQ