Blackwell B200 and GB200 NVL72 ramp across clouds
NVIDIA's Blackwell data-centre parts are now bookable by the hour, with 192 GB of HBM3e per GPU pushing single-node model sizes higher.
Source: nvidia.com
New silicon and accelerators — GPUs, memory and the systems they go into. Back to all the latest.
NVIDIA's Blackwell data-centre parts are now bookable by the hour, with 192 GB of HBM3e per GPU pushing single-node model sizes higher.
Source: nvidia.com
AMD's CDNA-3 refresh widens the memory lead for large-model inference, sharpening the open-ROCm alternative to NVIDIA in the data centre.
Source: amd.com
Intel's Battlemage flagship brings 16 GB and stronger ray tracing to the value bracket.
Source: intel.com
A higher-clocked, 24 GB RTX 5080 SUPER lands between the 5080 and 5090 for prosumer AI.
Source: nvidia.com
A 32 GB RDNA workstation card aimed at on-prem inference on the open ROCm stack.
Source: amd.com
141 GB Hopper parts get easier to rent as availability catches up with demand.
Source: nvidia.com
A compact Grace-Blackwell desktop brings 128 GB of unified memory to local development.
Source: nvidia.com
AMD's next data-centre accelerator lifts memory and low-precision throughput for inference.
Source: amd.com