言語 予測 異常検知 文書 音声 AI Apps (SaaS) 分析・可視化サービス AI Services 生成AIサービス ML Services AI Apps: Finance, Human Resource, Sales, Service, Procurement 分析・可視化サービス: Oracle Analytics Service AI Services: 画像認識: OCI Vision 文書認識: OCI Document Understanding 言語: OCI Language 予測: OCI Forecasting 異常検知: OCI Anomaly Detection 音声: OCI Speech チャットボット: OCI Digital Assistant 生成AIサービス: Oracle Generative AI Service ML Services: モデル開発・学習: Data Science Service データベース ML開発: ML on Database 学習データ作成: Labelling service データ連携・管理: データベース: Oracle Database, MySQL データレイク: Object Storage, Big Data データ連携: Data Integrator, GoldenGate メタデータ管理: OCI Data Catalog Oracle AI さまざまな用途に対応可能なAIポートフォリオ。 データ連携・管理 テキスト生成 要約 検索 分類 モデル開発・学習 データベース上での ML開発・デプロイ 学習データ作成 チャットボット
A100 80GB Azure AWS Google Cloud OCI NDm A100 v4 P4de.24xlarge A2-ultragpu-8g BM.GPU.GM4.8 Region US East (N. VA) East US (N. VA) US-Central1 Any region Instance type Virtual machine Virtual machine Virtual machine Bare metal vCPU 96 96 96 256(128コア) CPU memory (GB) 1900 GiB 1152 GB 1360 GB 2048 GB GPU type NVIDIA A100 80GB NVIDIA A100 80GB NVIDIA A100 80GB NVIDIA A100 80GB GPU 8 8 8 8 Local storage(TB) 6.4 TB 8 TB 3 TB 27.2 TB 月額 (730時間) $23,922 $29,905 $29,602 $23,360 https://blogs.oracle.com/cloud-infrastructure/post/ai-infrastructure-cloud-cost-comparison-best-value Lowest published on-demand list prices for AI compute instances as of May 2, 2023
19 https://www.oracle.com/cloud/economics/ 6K IOPS 38K - 375K IOPS If the cloud provider offered multiple options, including the need to purchase “performance credits,” the cheapest option was selected. Additional options, such as replication, were declined. For AWS, the choice was gp3. For Azure, it was Premium SSD v2. For Google Cloud, it was balanced Persistent Disk. The cheapest option was selected for each cloud provider and any additional options were declined. For AWS, the choice was io2 as gp3 tops out at just 256K IOPS. For Azure, it was Premium SSD v2. For Google Cloud, it was extreme Persistent Disk.
フォーマンスが最大50%高速化し、コストが最大80% 削減。」 (With OCI’s high-performance AI infrastructure, MosaicML states that it has seen up to 50 percent faster performance and cost savings of up to 80 percent compared to other cloud providers.) お客様事例: MosaicML
https://arxiv.org/pdf/2210.15315.pdf "Noise in the Clouds: Influence of Network Performance Variability on Application Scalability" 28 • HPC向けインスタンスの、2ノード間の遅延と帯域を計測 • 結果: Oracleは、安定した低遅延と、スペック通りの帯域 AWS Oracleの遅延は、低遅延( 約数μs)、かつ安定した通信 Oracleでは、帯域を最大限使 用したパフォーマンスで通信 AWS 遅延 帯域 Google Google