(+34) 91 361 10 02
comercial@sie.es
Dossier
Twitter
LinkedIn
YouTube
RSS
Sistema Informáticos EuropeosSistema Informáticos Europeos
  • Productos
    • Plataformas | Hardware
    • Software
  • Servicios
    • Redes
    • Almacenamiento
    • Seguridad
    • Virtualización
  • Proyectos
    • Desarrollos HPC
    • Proyectos I+D+i
  • Novedades
    • Noticias SIE
    • Casos de éxito
    • SIE Con la ciencia
    • Otras noticias
  • Conócenos
    • Nosotros
    • Contacto

NEW PCI-E 5.0 PLATFORMS

With the beginning of 2023, we already have the new PCI-E 5.0 PLATFORMS, both from AMD with Genoa and Intel with Sapphire Rapids.

 

The truth is that the first manufacturer to enter the new PCI-E 5.0 standard was NVIDIA with the H100 GPU and the new Mellanox HDR 200 Gb/s cards.

All this new technology involves doubling the bandwidth of the communication buses, as was previously the case with the move from PCI-E 3.0 to PCI-E 4.0.

Procesadores Sapphire Rapids | Intel | Nuevo estándar para arquitecturas de centros de datos

Procesadores Sapphire Rapids | Intel | Nuevo estándar para arquitecturas de centros de datos

Nuevos procesadores “Genoa” | AMD | Mejoran el rendimiento y la eficiencia

Nuevos procesadores “Genoa” | AMD | Mejoran el rendimiento y la eficiencia

Nueva GPU H100 | Nvidia | Un salto exponencial para la computación acelerada

Nueva GPU H100 | Nvidia | Un salto exponencial para la computación acelerada

Changes
Increased number of Cores
Increased memory
More speed
Better interconnectivity
New refrigeration technologies
Changes

There have also been other changes, which come hand in hand with the new buses and which we will be seeing in the coming months:

Increased number of Cores
AMD
0
Iniciales
0
Pronto
Intel
0
Iniciales
0
Pronto

Increase in the number of Cores in the processors: in the case of AMD, initially up to 96 Cores and in the immediate future up to 128 Cores, with the new Pergamo series. In Intel’s case, initially up to 48 Cores, but Sapphire will go up to 60.

Increased memory

DDR5 a 4800 MHz y 12 pipeline

Antes:  DDR4 a 3200 MHz y 8 pipeline

The move from DDR4 memory at 3200 MHz to DDR5 memory at 4800 MHz and from 8-pipeline to 12-pipeline memory, depending on the processors. This is logical, because the faster processors with more Cores, the greater the need for faster access to memory.

More speed
0
IOPS

Although they have already been announced, we will soon see NVME drives with PCI-E 5.0 technology, which in theory will allow up to 4 million IOPS.

These speeds are already achievable with the current and new LSI/Broadcom and Supreme Raid controllers (based on the NVIDIA T1000 chip).

Better interconnectivity

Of course, during the first half of the year, we will launch the Gigabyte HGX platforms based on NVIDIA H100 SXM5, with PCI-E 5.0 bus, but also with improved intercommunication through the internal bus of the NVLINK equipment and connectivity through NVSWITCH with the new 400 Gb/s Infiniband cards, which will also support direct connection to this bus.

All this, very focused on Artificial Intelligence solutions.

The H100 improves on the order of 2-3 times in HPC and 5-6 times in IA, the performance of its predecessor, the A100. In addition to the new bandwidth

New refrigeration technologies

During this quarter, we will have very dense platforms in these new Giga Computing technologies.

This will mean processors with power consumption up to 500 W and the new H100 SXM5 700 W.

That’s why we will also launch all new oil immersion technology.

Intel
AMD
Nuevos procesadores “Genoa” | AMD | Mejoran el rendimiento y la eficiencia
Nuevos procesadores “Genoa” | AMD | Mejoran el rendimiento y la eficiencia
NUEVAS PLATAFORMAS PCI-E 5.0
NUEVAS PLATAFORMAS PCI-E 5.0
  • Nota legal
  • Responsabilidad Social Corporativa
© 2020 Sistemas Informáticos Europeos, todos los derechos reservados.