Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Sign up for free
Menu
Search
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Pricing
Search
Sign in
Sign up for free
GCPと機械学習
Search
kmt-t
February 24, 2019
Programming
50
1
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
GCPと機械学習
クラウドでの深層学習環境についての発表です。
kmt-t
February 24, 2019
More Decks by kmt-t
See All by kmt-t
組込エンジニアのための深層学習
kmt_t
1
24
Deep Learningと超解像
kmt_t
0
31
Design by Contractとホーア論理
kmt_t
0
52
進化するART
kmt_t
0
35
Other Decks in Programming
See All in Programming
エージェントを使い続けるためのメタワーク / Meta-work for keeping using the agent
rkaga
1
160
AWSに止められる覚悟してますか?
morizo_1984
2
540
ActionController meets Protocol Buffers
sorah
0
560
自サークルのIdPからログアウトできるようになるまで
taiseiue
1
140
ソニーのクラウド共通基盤の変遷とAI時代の開発スタイルに合わせた進化 / The Journey of Sony’s Common Cloud Platform and Its Evolution for AI-Native Development
kenjiyoneyama
0
280
Agents on Rails - Rails at Scale 2026
irinanazarova
0
570
SalesForceを内製化!? ~ HR事業を支える顧客管理基盤のインフラを大公開 ~
oku053
0
180
npm との違いを手がかりにGo の依存管理を理解したい
optim
0
780
Herb in Rails 8.2: Your ERB views, now HTML-aware @ Rails World 2026, Austin, Texas
marcoroth
0
220
當我們開始不讀 Code - AI Agent 時代,Code Review 與 Clean Code 還剩下什麼?
jame2408
1
260
AIで開発が速くなった先で、開発チームは何を目指すか
shihochan
1
1.6k
FactoryBotをやめればRSpecは速くなるのか?
yuta_onishi_97
0
700
Featured
See All Featured
How People are Using Generative and Agentic AI to Supercharge Their Products, Projects, Services and Value Streams Today
helenjbeal
2
360
Connecting the Dots Between Site Speed, User Experience & Your Business [WebExpo 2025]
tammyeverts
11
1.1k
The Organizational Zoo: Understanding Human Behavior Agility Through Metaphoric Constructive Conversations (based on the works of Arthur Shelley, Ph.D)
kimpetersen
PRO
0
500
Marketing to machines
jonoalderson
1
5.9k
Leading Effective Engineering Teams in the AI Era
addyosmani
9
2.7k
Embracing the Ebb and Flow
colly
88
5.2k
How to Build an AI Search Optimization Roadmap - Criteria and Steps to Take #SEOIRL
aleyda
1
2.2k
Documentation Writing (for coders)
carmenintech
77
5.6k
Kristin Tynski - Automating Marketing Tasks With AI
techseoconnect
PRO
0
550
Jess Joyce - The Pitfalls of Following Frameworks
techseoconnect
PRO
1
450
Sharpening the Axe: The Primacy of Toolmaking
bcantrill
46
3k
What Being in a Rock Band Can Teach Us About Real World SEO
427marketing
0
1.1k
Transcript
GCPͱػցֶश ൃදऀ kmt-t 1/24
ࣗݾհ • ໊લ : kmt-t • ৬ۀ : ΈࠐΈϓϩάϥϚ •
ஶॻ : AndroidͷԾϚγϯ • ϚΠίϯ͓͡͞ΜͰ͢ 2/24
ͱͬͱΔͼʔհ • ௗऔࢢͷRubyษڧձ • 10͙Β͍ܧଓத • 3ϲ݄ʹҰճ։࠵ • Ruby͡Όͳ͍ൃද͋Γ 3/24
ௗऔͷΫϥυࣄ Ͳ͏ʁ 4/24
ൃදςʔϚ • ػցֶशͬͯԿ͢Δͷʁ • ΫϥυͰػցֶशΛ͢Δҙຯ • ΫϥυαʔϏεൺֱ • Ϋϥυͱਂֶश 5/24
ػցֶशͬͯԿ͢Δͷʁ 6/24
ػցֶशͰͰ͖Δ͜ͱ • ॳ৺ऀ͚ʹҎԼͷ;͕ͨͭͰ͖ΕΦοέʔ • ճؼ • ྨ 7/24
ճؼ • Y = f(X) • Կ͔ͷྔ͔Β͔ΒผͷྔΛਪ͢Δ • ྫ͑ؾԹ͔ΒϏʔϧͷൢചΛ༧ଌ •
XYϕΫτϧ (ͷ) ͩͬͨΓ 8/24
ྨ • σʔλ͕Կͳͷ͔ྨ͢Δ • ը૾ྨ͕༗໊ • ݘʁೣʁυϥ͑Μʁ 9/24
ػցֶशͷྲྀΕ • σʔλऩू : ֶशσʔλͷ࣭͕ॏཁ • ετϨʔδ : Ͳ͜ʹอଘ͢Δͷ͔ʁ •
લॲཧ : ܽଛิؒɺσʔλ૿ڧɺϕΫτϧԽ • ˒ֶश : ͕͔͔࣌ؒΔɺΧοτΞϯυτϥΠ • ˒ਪ : ࣮ࡍͷαʔϏεͰਪͷΈ͢Δ 10/24
ΫϥυͰػցֶशΛ͢Δҙຯ 11/24
ΫϥυͰΔϝϦοτ • ϚωʔδυαʔϏε • άϥϑΟΧϧͳઃܭ • εέʔϧΞτ༰қ • λΠϜϘοΫε͕͋ΔϓϩδΣΫτͩ ͱ͍҆
• ࣭ͷྑֶ͍शࡁΈϞσϧ 12/24
ΫϥυͰΔσϝϦοτ • GPUΠϯελϯε͕ߴ͍ • GPU͕ݹ͍ • ֶशσʔλͷσϓϩΠ͕ͭΒ͍ • ετϨʔδ͕͍ •
ѱ͍ҙຯͰϩʔΧϧͰ։ൃ͢ΔͷͱมΘΒͳ͍ 13/24
Ϋϥυͱਂֶश 14/24
େࡶͳ • ૬͕͋ΔͷͰϕʔεͷΠϯελϯεྉۚɺੑೳͷࠩখ͍͞ • GCP͕શʹଞͱҧ͏ͷ͕ΞΫηϥϨʔλ • GPU (AmazonɺϚΠΫϩιϑτ) • TPU
(Google) • GCPͰTPU͕͑ΔΑ͏ʹͳͬͨͷ࠷ۙ • ॲཧ͕͍࣌ؒͷͰܭࢉೳྗউෛ 15/24
GPUͷܭࢉೳྗ • ਂֶशͷܭࢉͷຆͲੵܭࢉ • ࢦඪTFLOPS/TOPS͕Ұൠత HW fp32 fp16 GeForceRTX2080Ti 13.4TFLOPS
107TFLOPS※ Tesla K80 5.6TFLOPS N/A Tesla V100 15.7TFLOPS 125TFLOPS※ (※TensorCoreͷੑೳ) 16/24
TPUͷܭࢉೳྗ • 4νοϓߏ • INT8ͰֶशͰ͖ͳ͍ HW bfloat16 INT8 TPUv1 N/A
92TOPS TPUv2 180TFLOPS ʁʁʁ TPUv3 v2ͷ2ഒ ʁʁʁ 17/24
ྉۚൺֱ • Ұϲ݄720࣌ؒɺ1υϧ110ԁͰܭࢉ Ϋϥυ αʔϏε HW ྉۚ TFLOPS୯Ձ GCP Cloud
TPU TPUv2 356ઍԁ/݄ 1980ԁ݄/TFLOPS GCP ϓϦΤϯϓςΟϒ TPU TPUv2 107ઍԁ/݄ 545ԁ݄/TFLOPS GCP ϓϦΤϯϓςΟϒ TPU TPUv3 190ઍԁ/݄ 528ԁ݄/TFLOPS AWS p3.2xlarge NVIDIA V100 242ઍԁ݄ 1938ԁ݄/TFLOPS AWS p2.xlarge NVIDIA K80 71ઍԁ݄ 12728ԁ݄/TFLOPS Azure NC6 NVIDIA K80 79ઍԁ݄ 14245ԁ݄/TFLOPS 18/24
TPUΞʔΩςΫνϟ 19/24
TPUͱGPUͷҧ͍ • ૬ରతʹTPUਪɺGPUֶश͕ಘҙ • GPUϕΫτϧɺTPUϚτϦοΫε • TPUͷํ͕ಋମͱͯ͠ͷޮ͕Α͍ • ͍҆GeForceσʔληϯλʔͰ͑ͳ͍ •
҆͘आΓΒΕΔGPU͕ݹ͍ • GPUआΓͳͯۙ͘ॴͷύιίϯͰ͍҆ 20/24
(ࢀߟ) ύιߪೖ • AMD RX470 • 5980ԁ • 5.186TFLOPS •
TensorFlow͑Δ • (Tesla K80/5.6TFLOPS/98ສ6ઍԁ) 21/24
Ͳ͏Ͱ͍͍ 22/24
ϚΠίϯ͓͡͞Μͷओு • ݸਓతʹΤοδAI͕Γ͍ͨ • ϚΠίϯ͓͡͞ΜతʹEdge TPU͔͠ڵຯͳ͠ • Edge TPUGoogleͷ֎ൢΈࠐΈTPU •
https://cloud.google.com/edge-tpu/?hl=ja 23/24
͝ਗ਼ௌ͋Γ͕ͱ͏͍͟͝·ͨ͠ 24/24