Part family

AI Compute / accelerators

Sourced to order. Every part is classified and inspected before shipment. Every list gets a quote.

AI Compute / accelerators parts on the line card

  • 100-300000108H AMD · CDNA 3 datacenter accelerator in the OAM form factor with 256 GB of HBM3e memory, for eight accelerator universal baseboards.
  • 100-300000254H AMD · CDNA 4 datacenter accelerator in the OAM form factor with 288 GB of HBM3E memory, liquid cooled, for eight accelerator universal baseboards.
  • 900-21010-0020-000 NVIDIA · Hopper datacenter GPU as a dual slot PCIe Gen5 card with 94 GB of HBM3 memory and NVLink bridge support for card pairs, for standard servers.
  • 900-21010-0040-000 NVIDIA · Hopper datacenter GPU as a dual slot PCIe Gen5 card with 141 GB of HBM3e memory and NVLink bridge support, for standard servers.
  • HL-325L Intel · Intel Gaudi 3 AI accelerator in the OAM form factor with 128 GB of HBM2e memory and 24 integrated 200 Gb Ethernet ports.
  • HL-338 Intel · Intel Gaudi 3 AI accelerator as a PCIe Gen5 add-in card, 128 GB HBM2e, sold per card; added 2026-10-12 on Ilya's order
  • NVRTXPRO6000TCGPU-KIT NVIDIA · Blackwell datacenter GPU as a passively cooled dual slot PCIe Gen5 card with 96 GB of GDDR7 ECC memory, for standard servers. Supplied as the PNY kit with an auxiliary power cable.
  • VCNRTXPRO6000B-PB NVIDIA · Blackwell workstation GPU as an actively cooled dual slot PCIe Gen5 card with 96 GB of GDDR7 ECC memory and four DisplayPort 2.1 outputs, for desktop AI, rendering and simulation workstations. Supplied as the PNY retail card.
  • VCNRTXPRO6000BQ-PB NVIDIA · The 300 W Max-Q edition of the RTX PRO 6000 Blackwell, an actively cooled dual slot PCIe Gen5 card with 96 GB of GDDR7 ECC memory and four DisplayPort 2.1 outputs, for multi GPU workstations and dense desktop builds. Supplied as the PNY retail card.

Parts not shown are sourced to order. Every part is classified and inspected before shipment. Add them by part number on the request page. We reply with a quote.

Guides