Slide 11
Slide 11 text
[Compute] GPU利用可能シェイプの拡充
11 Copyright © 2025, Oracle and/or its affiliates
詳細は担当営業までお問い合わせください
2025/11
32,768 GPUs
3,200 Gb/sec
RDMA
3,200 Gb/sec
RDMA
16,384 GPUs
3,200 Gb/sec
RDMA
65,536 GPUs
3,200 Gb/sec
RDMA
8,064 GPUs
NA
131,072 GPUs
3,200 Gb/sec
RDMA
131,072 GPUs
28,800 Gb/sec
RDMA
GPU.A10.1
GPU.A10.2
GPU.A10.4
VM, bare metal
NVIDIA
1 x A10 24 GB
2 x A10 24 GB
4 x A10 24 GB
GPU.GB200.4
NVL72
(18 nodes/rack)
Bare metal
36 Grace
CPUs
72
Blackwell
GPUs
13.6 TB
GPU memory
30.6 TB
GPU+CPU Memory
553 TB
Local NVMe
7,200 Gb/sec
Front-end network
GPU.H200.8
Bare metal
8 x H200
141 GB
1,128 GB
GPU memory
30.7 TB
Local NVMe
200 Gb/sec
Front-end network
GPU.H100.8
Bare metal
8 x H100
80 GB
640 GB
GPU memory
61.4 TB
Local NVMe
100 Gb/sec
Front-end network
GPU.A100-v2.8
Bare metal
8 x A100
80 GB
640 GB
GPU memory
27.2 TB
Local NVMe
100 Gb/sec
Front-end network
GPU.L40S.4
Bare metal
4 x L40S
48 GB
192 GB
GPU memory
7.7 TB
Local NVMe
200 Gb/sec
Front-end network
大規模,スケールアウト,AIトレーニング,推論,ファインチューニング
中規模
小規模
Object storage and/or FSS FSS HPMT—20/40/80 Gb/sec per MT Managed Lustre—8 Gb/sec per TB
GPU.GB300.4
NVL72
(18 nodes/rack)
Bare metal
36 Grace
CPUs
72
Blackwell
GPUs
20.7 TB
GPU memory
38.7 TB
GPU+CPU Memory
553 TB
Local NVMe
7,200 Gb/sec
Front-end network
3,840 GPUs
800 Gb/sec
RDMA
Bare metal
8 x
MI300X
192GB
GPU.MI300X.8
1,536 GB
GPU Memory
30.7 TB
Local NVMe
100 Gb/sec
Front-end
network
131,072 GPUs
6,400 Gb/sec
RDMA
GPU.RTXPro.8
Bare metal
8x RTX
Pro
6000
96GB
15.35 TB
Local NVMe
400 Gb/sec
Front-end network
16,384 GPUs 131,072 GPUs
6,400 Gb/sec
RDMA
OCI
Superclusters
GPU.MI355X.8
Bare metal
8 x
MI355X
288 GB
2 x 64C
AMD
CPU, 3TB
DDR5
2.3 TB
GPU memory
61.4TB
Local NVMe
400 Gb/sec
Front-end network
提供開始
131,072 GPUs
3,200 Gb/sec
RDMA
131,072 GPUs
57,600 Gb/sec
RDMA
GPU.B200.8
Bare metal
8 x B200
180 GB
1,440 GB
GPU memory
30.7 TB
Local NVMe
400 Gb/sec
Front-end network
注文受付中
GPU.B300
Bare metal
8 x B300
288GB
2,304 GB
GPU memory
30.7 TB
Local NVMe
400 Gb/sec
Front-end network
提供開始
提供開始
GPU.MI455X.4
Bare metal
4 x
MI455X
432
GB/GPU
2 x AMD
CPU
Local NVMe
400 Gb/sec
Front-end network
注文受付中
OCI offers NVIDIA L40S GPUs for Compute Cloud@Customer and NVIDIA L4 GPUs for Roving Edge Infrastructure
OCI offers NVIDIA H100 and H200 Tensor Core GPUs for OCI Dedicated Region and Oracle Alloy. NVIDIA B200 Tensor Core GPUs are orderable now.
関連URL: https://blogs.oracle.com/cloud-infrastructure/post/announcing-general-availability-of-oci-amd-mi355x
https://www.youtube.com/watch?v=ST443ix6-Bs&list=PLcFwxJMrxygDvjxvkOjqfHWMxjvvIglVc&index=11