One of the most important launches that happened around this year's CES was Nvidia's new GPU lineup. The AI behemoth announced RTX 50-series GPUs, bringing massive performance gains. One of its key components, the GDDR7 memory, is made by Samsung.
Nvidia RTX 50-series GPUs use Samsung's GDDR7 memory chips
Yesterday, Nvidia CEO Jensen Huang confirmed that its new RTX 50-series GPUs use GDDR7 VRAM memory chips, and the initial batch of GPUs will use memory chips from Samsung Electronics. While talking to reporters, he said, “GeForce RTX 50 series will use GDDR7 products from different partners. The first one is Samsung Electronics.”
On January 6, on the day of the new GPU announcement, Jensen mistakenly said that the new GPUs use GDDR7 memory from Micron. When he talked to the reporters, he also said he didn't know Samsung and SK Hynix even made graphics memory chips. However, that isn't true. So, a couple of days later, he corrected his mistake.
Samsung was the first to announce GDDR7 memory. The company completed the development of GDDR7 memory chips in mid-2023. The chips probably went into production in the second half of last year. The final GDDR7 specification was published early last year.
The new chips offer bandwidth of up to 1.5TB/s, which is 40% higher than GDDR6 chips. Their per-pin speed is 32Gbps, which was achieved using the PAM3 (Pulse-Amplitude Modulation) signaling method.
Nvidia is currently the world leader in AI accelerator and GPU segments. Its latest AI chips use HBM3E memory chips from Micron and SK Hynix. While Samsung also makes HBM3E chips, its chips haven't received Nvidia's certification, which led to heavy losses in Samsung's semiconductor division.
