X
Business

Samsung denies it is failing Nvidia's tests for HBM chips

Samsung says its plan to supply HBM, arguably the most important memory chip for AI processing, is on track as pressure mounts to deliver.
Written by Cho Mu-Hyun, Contributing Writer
hbm3e-12h-1.png

Image: Samsung

Samsung on Friday denied a report that claimed its high-bandwidth memory (HBM) chips were failing to pass Nvidia's tests due to heating and power consumption issues.

The South Korean tech giant was responding to a Reuters report earlier in the day, citing unnamed sources, that alleged Samsung failed to get approval in its latest test results in April for its HBM3E chips by the US GPU giant to use with their AI processors. The South Korean tech giant has been testing its HBM3 and HBM3E with Nvidia since last year, the report said.

Also: Samsung replaces chip boss amid concern over leadership position

Samsung, without naming Nvidia, in a statement said that its tests with various global partners to supply HBM were "proceeding smoothly." The company was "running various tests to strictly verify the quality and performance of the HBM chips," Samsung added.

HBM3E is the latest generation of HBM, a chip that stacks multiple DRAM dies vertically to reduce space and power consumption while offering high bandwidth that allows them to process large amounts of data. Because of this, they are considered a must with AI processors such as H100, B100, and H200 launched by Nvidia. They are what double data rate (DDR) DRAMs were to CPUs in PCs and low-power double data rate (LPDDR) DRAMs to application processors in smartphones for AI processing.

Also: Samsung's chip division wants the top spot back: What's happening?

Reuter's report comes at a sensitive time for Samsung, the world's largest memory maker, as it has been falling behind its domestic rival SK Hynix in supplying HBM to the market currently dominated by Nvidia.

Any quality tests by clients taking months is not uncommon in the chip industry, where a new chip will start development usually two years in advance before it rolls out commercially. Back and forth with a client over varying test results is also not uncommon.

However, just a few days earlier this week, the South Korean tech giant announced that it replaced its chip division head to overcome what it called a "semiconductor crisis". The highly unusual move was considered a response to its failure to take the lead in the HBM space that is rapidly growing from the AI boom. In other words, the claims of test failures come at a critical juncture when Samsung looks considerably vulnerable.

Samsung cannot miss out on the US GPU maker as a customer in the AI space. On Wednesday, Nvidia said it saw revenue jump 260% from a year earlier in its latest fiscal quarter from February to April, thanks to high demand for its AI processors from cloud service providers. According to analyst firm TrendForce, Nvidia has an 80% market share in the AI processor market.

Also: Samsung expects AI to drive sales of chips and smartphones in 2024

SK Hynix, meanwhile, is effectively the exclusive supplier of HBMs to Nvidia at the current time and has a huge lead. The South Korean memory maker, which already has its HBM3E approved, said earlier this month that its capacity for HBM chips for this year has already been sold out.  

While Samsung is a leader in other products such as smartphones and TVs, memory chips are considered the backbone that made the company the global brand it is today and its pride. Whether it can maintain its leadership in the AI era will be revealed in the coming months.

Editorial standards