Tech News
Back to home
Technology

HighPoint's adapter enables GPUDirect storage — up to 64 GB/s from drive to GPU, bypassing the CPU

Source

Tom's Hardware

Published

TL;DR

AI Generated

HighPoint has unveiled the Rocket 7638D PCIe 5.0 switch card, facilitating GPUDirect connection between AI GPUs and NVMe storage devices, bypassing the CPU for speeds of up to 64 GB/s. This technology enhances AI training and inference workloads with compatible software. The adapter supports up to 16 NVMe drives, offering configurations with up to 2PB of high-performance storage. The Broadcom PEX 89048 switch chip in the adapter is independent and compatible with both Arm and x86 platforms, simplifying integration with major operating systems. HighPoint's Rocket 7638D also includes features for health monitoring and troubleshooting, catering to multi-node or hyperscale installations.

Read Full Article

Similar Articles

GameStop NVMe SSD 2TB SSD Review: A surprisingly capable budget drive for the PS5

GameStop NVMe SSD 2TB SSD Review: A surprisingly capable budget drive for the PS5

GameStop surprised with a capable PS5-friendly NVMe SSD that offers good performance and pricing across all capacities. The drive features a reliable Silicon Motion controller paired with Samsung flash, making it a solid budget pick. While lacking detailed specifications, the drive's performance met expectations in testing. With competitive pricing and a PS5-compliant heatsink, the GameStop NVMe SSD is a good deal for those looking for an affordable SSD option.

Tom's Hardware•
Samsung preps PCIe 5.0 QLC SSD with a controller based on open-source RISC-V architecture — BM9K1 delivers speeds up to 11.4 GB/s for 'personal AI workloads'

Samsung preps PCIe 5.0 QLC SSD with a controller based on open-source RISC-V architecture — BM9K1 delivers speeds up to 11.4 GB/s for 'personal AI workloads'

Samsung has introduced the BM9K1 SSD, a PCIe 5.0 drive with QLC NAND that boasts sequential read speeds of up to 11.4 GB/s, targeting 'personal AI computing' workloads. The drive features a controller based on the open-source RISC-V architecture, offering improved energy efficiency and firmware optimization for QLC NAND management. Samsung plans to launch the BM9K1 in 2027 in capacities of 512GB, 1TB, and 2TB. This move puts Samsung in competition with Micron's 3610 PCIe 5.0 QLC SSD, with Samsung's drive offering higher sequential read speeds but lower capacities.

Tom's Hardware•
Microsoft blocks registry trick that unlocked performance-boosting native NVMe driver on Windows 11 — workarounds still exist to enable support, however

Microsoft blocks registry trick that unlocked performance-boosting native NVMe driver on Windows 11 — workarounds still exist to enable support, however

Microsoft has blocked a registry trick that allowed Windows 11 users to unlock native NVMe support, offering significant performance boosts. The trick, which had delivered up to 85% higher random write performance, is no longer functional due to changes in recent Windows 11 Insider builds. Despite this, workarounds like using ViVeTool still exist to enable native NVMe support, although they come with compatibility risks. Microsoft's native NVMe driver promises substantial performance improvements, but its official rollout to Windows 11 users is still pending.

Tom's Hardware•
Kioxia announces new Super High IOPS SSD that helps accelerate AI workloads on Nvidia GPUs — 25.6TB drive provides more GPU-accessible memory for faster data access

Kioxia announces new Super High IOPS SSD that helps accelerate AI workloads on Nvidia GPUs — 25.6TB drive provides more GPU-accessible memory for faster data access

Kioxia has introduced the E3.S CM9 series SSD in collaboration with Nvidia to enhance AI workloads on Nvidia GPUs. This "Super High IOPS" SSD offers 25.6TB capacity and can sustain high write speeds throughout its warranty period. Equipped with XL-Flash technology, the drive achieves over 10 million IOPS, significantly surpassing traditional datacenter SSDs. By directly linking these drives to GPUs, Nvidia aims to optimize data access and overcome limitations in existing memory solutions for AI models with massive parameters.

Tom's Hardware•

We use cookies

We use cookies to ensure you get the best experience on our website. For more information on how we use cookies, please see our cookie policy.