NVIDIA’s latest workstation is a beast equipped with four A100 GPUs of 80GB each



[ad_1]

NVIDIA has revealed its latest workstation for professionals. An impressive golden obelisk that holds enough energy within it to be used in serious machine learning and artificial intelligence work in general. The DGX Station A100, among other things, carries a total of 320 GB of GPU memory.


The battle between design, glitz and the future: all the details of battle stations from an expert point of view

Waste of energy encapsulated in a gold box

The new DGX is a workstation that part of the A100 GPUs which NVIDIA introduced in May of this year. As we saw at the time, they were not meant for the average consumer, but for a professional environment. The new DGX A100s are just that, a professional rig equipped with up to four A100 GPUs.

Inside the box are a number of other components that are worth highlighting along with the GPUs. For example, we find a file 64-core AMD CPU with 512GB of memory. As for storage, it comes with 7.68TB NVME SSD. We also find PCIe Gen-4 support and a DGX display adapter with four DisplayPort outputs and support for 4K resolutions.

Nvidia A100 Sxm4 80GB

Most interesting of all is that it does not require special power or cooling equipment. According to NVIDIa, just plug it into your home outlet like a normal computer.

Going back a little to the GPU of this DGX A100, it uses third generation voltage cores, along with third generation NVLink and NVS technology. This doubles the GPU-to-GPU bandwidth, for transfers of up to 600GB per second. NVIDIA indicates that it is the only workgroup that supports NVIDIA Multi-Instance GPU (MIG) technology. With it the DGX A100 can offer up to 28 separate GPU instances work in parallel with different users or different activities.

NVIDIA

NVIDIA declined to say how much one of these workstations cost. However, we can be clear that the figure is certainly high. Just to contextualize things each of the A100 GPUs are priced at $ 200,000.

Street | Engadget
More information | NVIDIA

[ad_2]
Source link