At re:Invent 2023, AWS CEO Adam Selipsky said the second generation of the company’s Trainium chip can train models four times faster than its predecessor. Trainium 2 has three times the memory as the first chip.
AWS will be the first cloud provider to deploy Nvidia’s GH200 chips, originally announced for release in 2024. Nvidia said GH200 has triple the memory of the popular H100 chips.
Nvidia CEO Jensen Huang, a surprise guest during AWS CEO Adam Selipsky’s keynote at the AWS re:Invent conference, also announced Nvidia’s AI “factory” DGX Cloud is coming to AWS.
Peter Desantis, general manager of AWS EC2, announced during a re:Invent keynote that the new chip helps figure out how to correct errors — like flipping a binary code from a 1 to a 0 — that occasionally pop up when working with quantum computing queries. If quantum computing errors are mitigated, it could pave the way for more usable quantum computing, Desantis said.
“We are still in the very early stages, but this chip represents an important step in error correction for quantum computing.”
Instead of relying on hard drives, thin clients like the cube offload most computing power to the cloud and can connect virtual desktops remotely. AWS’ Melissa Stein said the company wants to provide a low-cost option for clients in high employee turnover spaces, like call centers, to provide access to enterprise applications.
“We looked for options and found that the hardware we used for the Amazon Fire TV Cube provided all the resources customers needed to access their cloud-based virtual desktops. So, we built an entirely new software stack for that device, and since we didn’t have to design and build new hardware, we’re passing those savings along to customers.”
The AWS thin clients will cost $195, or $280 bundled with a hub to connect a second monitor, with a monthly $6 management and maintenance fee.
OpenAI said it had to fight off the attack which caused ChatGPT and its APIs to go down for 90 minutes on Wednesday.
“We are dealing with periodic outages due to an abnormal traffic pattern reflective of a DDoS attack. We are continuing work to mitigate this,” the company said.
CEO Sam Altman originally pointed to an increase in users as the cause of the problem. He said people may continue seeing “service instability”
[status.openai.com]
