Nvidia has approved the first-ever usage of Samsung Electronics' fourth-generation high bandwidth memory, or HBM3 chips, in its CPUs, according to three persons briefed on the situation.
However, the approval is a little subdued since, according to the individuals, Samsung's HBM3 chips will only be utilised in the H20, a less advanced Nvidia graphics processing unit (GPU) that was created for the Chinese market in accordance with US export regulations.
They also mentioned that it was unclear at the time if Nvidia will include Samsung's HBM3 chips into its other AI processors or if those chips would need to pass further testing.
These insiders, who wished to remain anonymous because they were not permitted to speak to the media, said that Samsung has also failed to match Nvidia's specifications for fifth-generation HBM3E chips, and that testing of those chips is still ongoing.
There were no comments from Samsung and Nvidia.
Chips are vertically stacked in HBM, a kind of dynamic random access memory (DRRAM) that was initially manufactured in 2013 in order to conserve space and lower power usage. It is a crucial part of GPUs for AI and aids in processing vast volumes of data generated by intricate applications.
The generative AI boom has produced a skyrocketing demand for advanced GPUs, which Nvidia and other producers of AI chipsets are finding difficult to satisfy. This is why Nvidia approved Samsung's HBM3 processors.
There are just three primary producers of HBM: SK Hynix, Micron, and Samsung. Since HBM3 is similarly in low supply, Nvidia is eager for Samsung to meet its requirements so that it may expand the variety of suppliers it works with.
According to two of the individuals, SK Hynix, the industry leader, intends to produce more HBM3E and less HBM3, which would boost Nvidia's requirement for more access to HBM3.
There were no comments available from SK Hynix.
According to Reuters in May, which cited sources, Samsung, the largest memory chip manufacturer in the world, has been trying since last year to pass Nvidia's testing for both HBM3 and HBM3E, but has had difficulty because of problems with heat and power consumption.
Following the May release of the Reuters piece, Samsung denied any allegations that it had failed Nvidia's testing because of issues with heat and power consumption.
Two of the sources claim that Samsung may start providing HBM3 for Nvidia's H20 CPU as early as August.
Nvidia designed three GPUs specifically for the Chinese market, the H20 being the most sophisticated. The U.S. increased export limits in 2023, hoping to obstruct advancements in AI and supercomputing that would be advantageous to China's military.
The H20's processing capability has been severely limited in compliance with US sanctions, in contrast to the H100 model that is marketed in non-China countries.
When deliveries of the H20 started this year, the American company priced it lower than a competing chip from the Chinese tech giant Huawei, which led to a rough beginning for the device, as Reuters noted in May.
But according to other reports, sales are currently rising quickly.
SK Hynix, as opposed to Samsung, is Nvidia's primary source of HBM chips and has been providing HBM3 since June 2022. Additionally, it started providing HBM3E to an unnamed client in late March. Sources claim that shipments ended up at Nvidia.
Micron has also stated that it will provide HBM3E to Nvidia.
(Source:www.scmp.com)
However, the approval is a little subdued since, according to the individuals, Samsung's HBM3 chips will only be utilised in the H20, a less advanced Nvidia graphics processing unit (GPU) that was created for the Chinese market in accordance with US export regulations.
They also mentioned that it was unclear at the time if Nvidia will include Samsung's HBM3 chips into its other AI processors or if those chips would need to pass further testing.
These insiders, who wished to remain anonymous because they were not permitted to speak to the media, said that Samsung has also failed to match Nvidia's specifications for fifth-generation HBM3E chips, and that testing of those chips is still ongoing.
There were no comments from Samsung and Nvidia.
Chips are vertically stacked in HBM, a kind of dynamic random access memory (DRRAM) that was initially manufactured in 2013 in order to conserve space and lower power usage. It is a crucial part of GPUs for AI and aids in processing vast volumes of data generated by intricate applications.
The generative AI boom has produced a skyrocketing demand for advanced GPUs, which Nvidia and other producers of AI chipsets are finding difficult to satisfy. This is why Nvidia approved Samsung's HBM3 processors.
There are just three primary producers of HBM: SK Hynix, Micron, and Samsung. Since HBM3 is similarly in low supply, Nvidia is eager for Samsung to meet its requirements so that it may expand the variety of suppliers it works with.
According to two of the individuals, SK Hynix, the industry leader, intends to produce more HBM3E and less HBM3, which would boost Nvidia's requirement for more access to HBM3.
There were no comments available from SK Hynix.
According to Reuters in May, which cited sources, Samsung, the largest memory chip manufacturer in the world, has been trying since last year to pass Nvidia's testing for both HBM3 and HBM3E, but has had difficulty because of problems with heat and power consumption.
Following the May release of the Reuters piece, Samsung denied any allegations that it had failed Nvidia's testing because of issues with heat and power consumption.
Two of the sources claim that Samsung may start providing HBM3 for Nvidia's H20 CPU as early as August.
Nvidia designed three GPUs specifically for the Chinese market, the H20 being the most sophisticated. The U.S. increased export limits in 2023, hoping to obstruct advancements in AI and supercomputing that would be advantageous to China's military.
The H20's processing capability has been severely limited in compliance with US sanctions, in contrast to the H100 model that is marketed in non-China countries.
When deliveries of the H20 started this year, the American company priced it lower than a competing chip from the Chinese tech giant Huawei, which led to a rough beginning for the device, as Reuters noted in May.
But according to other reports, sales are currently rising quickly.
SK Hynix, as opposed to Samsung, is Nvidia's primary source of HBM chips and has been providing HBM3 since June 2022. Additionally, it started providing HBM3E to an unnamed client in late March. Sources claim that shipments ended up at Nvidia.
Micron has also stated that it will provide HBM3E to Nvidia.
(Source:www.scmp.com)
