Samsung Unveils Shinebolt HBM3E Memory At Nearly 10Gbps And Blistering 32Gbps GDDR7
![Samsung HBM3E news](https://images.hothardware.com/contentimages/newsitem/62866/content/Samsung-HBM3E-news.png)
Due to its incredibly high memory bandwidth per chip and its stackable nature, HBM has been popular for high-end GPUs and even CPUs since JEDEC finalized its specification in 2013. NVIDIA's Grace Hopper GH200 is getting an HBM3E variant, though the original version already used HBM3. Intel also used HBM2E for its Xeon Max Sapphire Rapids CPU.
SK Hynix releaesd its HBM3E to partners in May, while Micron did the same in July. Samsung differentiates its Shinebolt chips by boosting performance considerably; while HBM3E from SK Hynix and Micron can hit 8Gbps and 9.2Gbps per pin, respectively, Samsung's HBM3E can do 9.8Gbps per pin. That's a ton of bandwidth in general, and a substantial lead over the competition.
![Samsung GDDR7 news](https://images.hothardware.com/contentimages/newsitem/62866/content/Samsung-GDDR7-news.png)
While the focus of Samsung's announcement was clearly HBM3E, the company also announced its upcoming GDDR7 memory chips, which will hit 32Gbps per chip. For reference, GDDR6X (made by Micron for NVIDIA) maxes out roughly at 22Gbps, and regular GDDR6 at 16Gbps. GDDR6 has been around for a while, first debuting on NVIDIA's GeForce RTX 20 series in 2018, while GDDR6X came out with the RTX 30 series. Still, GDDR6X is three years old, so the upgrade GDDR7 will provide for next-generation GPUs will be significant.
Previously, Samsung detailed other technical specifications of its GDDR7 memory: it will be 20% more efficient than GDDR6 and will allow for memory bandwidth of around 1.5 terabytes per second in some configurations. Those kinds of speeds on consumer GPUs were previously only possible with HBM2, which was able to deliver just over 1TB/s worth of bandwidth on the Radeon VII all the way back in 2019. Even the GeForce RTX 4090 has just a tad less memory bandwidth than the Radeon VII, but with GDDR7 it looks like GPUs will be able to enjoy the speeds HBM2 previously delivered.