$199,000 Nvidia monster works with AMD Epyc CPUs
Rivals can also be friends; Lisa Su congratulated and thanked the partnership
Published by Char Aznable - 15/05/2020
Rivals can also be friends; Lisa Su congratulated and thanked the partnership
Published by Char Aznable - 15/05/2020
[SHOWTOGROUPS=4,20]
Nvidia DGX A100 (Image credit: Nvidia)
The DGX A100 is an ultra-powerful system that has many features of Nvidia on the outside, but there are also some of AMD on the inside. A heavy pair of AMD Epyc 7742 processors (code-named Rome) is at the heart of Nvidia's new $ 199,000 creation.
The DGX A100 employs up to eight Ampere A100 data center GPUs, offering up to 320 GB of total GPU memory and providing approximately 5 petaflops of AI performance. The A100 may be doing most of the heavy lifting, but the team still needs a leader. However, Intel does not fit the account.
The A100 uses PCIe 4.0, but Intel does not currently have processors compatible with the interface. AMD, on the other hand, has openly adopted the PCIe 4.0 standard on most of its modern CPUs. Nvidia has finally found comfort in AMD's arms, more specifically Red Team's second generation Epyc offerings.
The DGX A100 features two 7nm Epyc 7742 processors. Each Zen2 processor comes with 64 cores and 128 threads that work with a base clock of 2.25 GHz and a boost clock of 3.4 GHz. The double Epyc 7742 is responsible for 128 cores and 256 threads in the DGX A100.
The Epyc 7742 is not just generous with cores; it is also quite cumbersome in the cache, providing up to 256 MB of L3 cache. Most importantly, the 64-core part makes 128 high-speed PCIe 4.0 lanes available to Nvidia.
A DGX A100 also comes with 1 TB of memory (upgradeable to 2 TB), two 1.92 TB NVMe M.2 SSDs in RAID 1 for the operating system (Ubuntu Linux) and up to four PCIe 4.0 NVMe U.2 out of 3 , 84 TB drives in RAID 0 for secondary storage. Nvidia also offers the option of adding four additional SSDs to increase the RAID 0 volume from 15 TB to 30 TB.
The DGX A100 was designed with the latest generation network. Nvidia recently acquired Mellanox Technologies in a whopping $ 6.9 billion deal, and it's already paying off. The DGX A100 includes eight single-port Mellanox ConnectX-6 VPI HDR InfiniBand adapters for clustering and a ConnectX-6 VPI dual-port Ethernet adapter for network storage and use. The adapters mentioned above have the capacity to provide up to 200 Gbps throughput.
Lisa Su also congratulated Nvidia and thanked the collaboration between the companies.
или Зарегистрируйся
Proud to have @AMDServer #EPYC CPUs '' Rome '' on the new DGX servers. Congratulations @nvidia on the new A100 release and thanks for choosing @AMD as your CPU partner.
SOURCE: Для просмотра ссылки Войдиили Зарегистрируйся
[/SHOWTOGROUPS]
Nvidia DGX A100 (Image credit: Nvidia)
The DGX A100 is an ultra-powerful system that has many features of Nvidia on the outside, but there are also some of AMD on the inside. A heavy pair of AMD Epyc 7742 processors (code-named Rome) is at the heart of Nvidia's new $ 199,000 creation.
The DGX A100 employs up to eight Ampere A100 data center GPUs, offering up to 320 GB of total GPU memory and providing approximately 5 petaflops of AI performance. The A100 may be doing most of the heavy lifting, but the team still needs a leader. However, Intel does not fit the account.
The A100 uses PCIe 4.0, but Intel does not currently have processors compatible with the interface. AMD, on the other hand, has openly adopted the PCIe 4.0 standard on most of its modern CPUs. Nvidia has finally found comfort in AMD's arms, more specifically Red Team's second generation Epyc offerings.
The DGX A100 features two 7nm Epyc 7742 processors. Each Zen2 processor comes with 64 cores and 128 threads that work with a base clock of 2.25 GHz and a boost clock of 3.4 GHz. The double Epyc 7742 is responsible for 128 cores and 256 threads in the DGX A100.
The Epyc 7742 is not just generous with cores; it is also quite cumbersome in the cache, providing up to 256 MB of L3 cache. Most importantly, the 64-core part makes 128 high-speed PCIe 4.0 lanes available to Nvidia.
A DGX A100 also comes with 1 TB of memory (upgradeable to 2 TB), two 1.92 TB NVMe M.2 SSDs in RAID 1 for the operating system (Ubuntu Linux) and up to four PCIe 4.0 NVMe U.2 out of 3 , 84 TB drives in RAID 0 for secondary storage. Nvidia also offers the option of adding four additional SSDs to increase the RAID 0 volume from 15 TB to 30 TB.
The DGX A100 was designed with the latest generation network. Nvidia recently acquired Mellanox Technologies in a whopping $ 6.9 billion deal, and it's already paying off. The DGX A100 includes eight single-port Mellanox ConnectX-6 VPI HDR InfiniBand adapters for clustering and a ConnectX-6 VPI dual-port Ethernet adapter for network storage and use. The adapters mentioned above have the capacity to provide up to 200 Gbps throughput.
Lisa Su also congratulated Nvidia and thanked the collaboration between the companies.
Для просмотра ссылки Войди
Proud to have @AMDServer #EPYC CPUs '' Rome '' on the new DGX servers. Congratulations @nvidia on the new A100 release and thanks for choosing @AMD as your CPU partner.
SOURCE: Для просмотра ссылки Войди
[/SHOWTOGROUPS]
