Technology

Huawei unveils new Atlas 350 AI accelerator with 1.56 PFLOPS of FP4 compute and up to 112GB of HBM — claims 2.8x more performance than Nvidia's H20

Source

Tom's Hardware

Published

Mar 24, 2026

TL;DR

AI Generated

Huawei has introduced the Atlas 350 AI accelerator, featuring 1.56 PFLOPS of FP4 compute and up to 112GB of HBM, claiming 2.8 times more performance than Nvidia's H20. This new NPU is based on the Ascend 950PR chip and is optimized for FP4 precision, allowing for larger AI models on the same hardware with less memory. Despite challenges due to U.S. sanctions, Huawei's Atlas 350 showcases impressive specs and is priced competitively at around $16,000, aiming to reduce reliance on foreign hardware in China's AI landscape.

Read Full Article

Jensen Huang snubbed by White House for President Trump’s China state visit — Nvidia CEO not on roster, which includes Apple's Tim Cook and Elon Musk

Nvidia CEO Jensen Huang was notably absent from the roster of business leaders invited to join President Trump on his state visit to China, despite being a prominent figure in previous visits. The exclusion comes amidst tensions over Nvidia's chip exports to China and the U.S. blocking advanced semiconductors due to military concerns. This move has significantly impacted Nvidia's market share in China, allowing local chipmakers to gain ground. The absence of Huang from Trump's entourage is seen as a strong signal to China regarding the U.S.' stance on high-end AI chips and trade negotiations.

Tom's Hardware•

1 hour ago

Amazon employees admit to using AI unnecessarily to pump up internal usage scores — workers complain of intense pressure to use AI tools

Amazon employees have admitted to artificially inflating their usage of AI tools to meet internal targets, with some using an in-house platform called MeshClaw to boost their numbers. While Amazon stated that usage stats wouldn't affect performance evaluations, employees felt pressure to utilize these tools, leading to concerns about the accuracy of demand figures driving billions in AI infrastructure spending. This practice, known as "tokenmaxxing," has raised questions about the reliability of AI consumption data and the potential impact on the industry's growth. Experts suggest a shift towards measuring efficient token usage rather than sheer volume to ensure sustainable returns on massive investments in AI infrastructure.

Tom's Hardware•

2 hours ago

Compromised Mistral AI and TanStack packages may have exposed GitHub, cloud and CI/CD credentials in 'mini Shai Hulud' malware infection — supply-chain campaign spreads across npm and AI developer ecosystems like wildfire

Microsoft is investigating a compromise of the mistralai PyPI package, where attackers injected malicious code that automatically executed on import, downloaded a secondary payload, and launched malware on Linux systems. The compromised package contained code that silently downloaded a file and executed it in the background, potentially stealing credentials and executing destructive commands. This incident is part of the broader "Mini Shai-Hulud" software supply-chain campaign affecting npm and AI developer ecosystems. Additionally, security firm Aikido warned of compromised TanStack JavaScript packages and Mistral npm SDK packages in separate attacks, urging developers to rotate credentials if affected. The incidents highlight the increasing targeting of developer infrastructure, emphasizing the importance of securing high-value credentials in modern development environments.

Tom's Hardware•

3 hours ago

Dell SupportAssist update is crashing PCs with constant blue screens and reboot loops — the boot service built for system recovery is the culprit of unending instability

A recent update to Dell's SupportAssist Remediation software is causing constant blue screen errors and reboot loops on various Dell laptop models. The issue stems from the update version 5.5.16.0 released on April 30th, affecting systems like the XPS 15 9530 and Dell Pro Plus 14. Users have identified the culprit as the DellSupportAssistRemedationService.exe process and have found workarounds to stop the crashes, such as disabling the service or uninstalling the software entirely. This isn't the first time SupportAssist Remediation has caused such issues, with a similar pattern reported last year. Dell has yet to acknowledge the problem or provide a fix.

Tom's Hardware•

4 hours ago

Jensen Huang snubbed by White House for President Trump’s China state visit — Nvidia CEO not on roster, which includes Apple's Tim Cook and Elon Musk

Tom's Hardware•

1 hour ago

Amazon employees admit to using AI unnecessarily to pump up internal usage scores — workers complain of intense pressure to use AI tools

Tom's Hardware•

2 hours ago

Compromised Mistral AI and TanStack packages may have exposed GitHub, cloud and CI/CD credentials in 'mini Shai Hulud' malware infection — supply-chain campaign spreads across npm and AI developer ecosystems like wildfire

Tom's Hardware•

3 hours ago

Dell SupportAssist update is crashing PCs with constant blue screens and reboot loops — the boot service built for system recovery is the culprit of unending instability

Tom's Hardware•

4 hours ago

Huawei unveils new Atlas 350 AI accelerator with 1.56 PFLOPS of FP4 compute and up to 112GB of HBM — claims 2.8x more performance than Nvidia's H20

TL;DR

Similar Articles

Jensen Huang snubbed by White House for President Trump’s China state visit — Nvidia CEO not on roster, which includes Apple's Tim Cook and Elon Musk

Amazon employees admit to using AI unnecessarily to pump up internal usage scores — workers complain of intense pressure to use AI tools

Compromised Mistral AI and TanStack packages may have exposed GitHub, cloud and CI/CD credentials in 'mini Shai Hulud' malware infection — supply-chain campaign spreads across npm and AI developer ecosystems like wildfire

Dell SupportAssist update is crashing PCs with constant blue screens and reboot loops — the boot service built for system recovery is the culprit of unending instability

We use cookies

Huawei unveils new Atlas 350 AI accelerator with 1.56 PFLOPS of FP4 compute and up to 112GB of HBM — claims 2.8x more performance than Nvidia's H20

TL;DR

Similar Articles

Jensen Huang snubbed by White House for President Trump’s China state visit — Nvidia CEO not on roster, which includes Apple's Tim Cook and Elon Musk

Amazon employees admit to using AI unnecessarily to pump up internal usage scores — workers complain of intense pressure to use AI tools

Compromised Mistral AI and TanStack packages may have exposed GitHub, cloud and CI/CD credentials in 'mini Shai Hulud' malware infection — supply-chain campaign spreads across npm and AI developer ecosystems like wildfire

Dell SupportAssist update is crashing PCs with constant blue screens and reboot loops — the boot service built for system recovery is the culprit of unending instability