Samsung could double HBM2 memory production and still not meet demand
Samsung could double HBM2 memory production and still not meet demand
Another factor that needs to be considered is demand, as while HBM2 memory is rarely seen in the consumer market, HBM2 is widely used to make professional-grade GPUs, FPGAs and other devices, with Intel, AMD and Nvidia all acting as large buyers of the new memory type.Â
At ISC 2018, Samsung reportedly stated that they could double their production of HBM2 memory and still not satisfy market demand, another factor that allows HBM2’s pricing to remain high. This factor also makes HBM2 an extremely high margin product for HBM2 memory makers like Samsung and SK Hynix.Â
Â
(Image from Glenn K. Lockwood)
HBM2 memory has its benefits, Samsung’s latest Aquabolt HBM2 memory modules can offer 8GB of capacity and speeds of 307GBps with a single memory chip, which is more memory bandwidth than a Nvidia GTX 1070 graphics card can provide. Not bad for a single chip that only requires 1.2V for power. Â
Another benefit of HBM2 memory is that it allows users to fit faster products into smaller spaces, as an equivalent GDDR5/GDDR6 setup would take up significantly more space and power, a factor which is useful for small form factor systems or products where power savings can have a huge impact on power cost and/or battery life.Â
As it stands, it looks like HBM2 will not receive much more adoption within the consumer market, as the cost of the memory is likely to be a lot higher than GDDR6, which is now being mass produced by Samsung, SK Hynix and Micron. HBM2 needs to become a lot cheaper before it can become anything more than a niche product.Â
You can join the discussion on Samsung’s HBM2 supply failing to meet demand on the OC3D Forums.Â