Samsung launched its new GDDR6 memory. This is new memory that we expect will be used in future generations of GPUs since it is a fairly large speed bump over previous generations.
Samsung GDDR6 24Gbps DRAM Launched
The packages themselves are 180FBGA packages that, if the platform supports it, could work in existing footprints. Given how GPUs are made, and that these are high-speed parts, we think this will be used on new cards.
Here is the specs for the K4ZAF325BC-SC24 which is the 24Gbps model. Samsung also has a 20Gbps model but this is the only part listed on Samsung’s website at 24Gbps GDDR6 speeds.
We can see that the product status is “Sample” but given the fact we have both a press release as well as the public listing of the new memory on the Samsung website, it seems like this is something designed for the next generation of GPUs. Usually, when technology is for 2-3 generations out, we get just a press release but not a model number, as we did with the Samsung CXL Memory Expander with DDR5 Launched back in 2021.
One of the biggest challenges of modern CPUs and GPUs is simply memory bandwidth. The new chips can help increase capacity (in areas using 8Gb chip designs) as well as boost memory bandwidth. Those two factors combined will hopefully make the next generation of GPUs more attractive. As GPU prices have fallen rapidly over the last quarter or so, that will put pressure on companies like AMD and NVIDIA to move inventory before the next generation comes out.
NVIDIA already is using GDDR6X on some GPUs like the NVIDIA GeForce RTX 3090 Review A Compute Powerhouse, but on some of the other GPUs in the market, they will see a benefit from the new Samsung chips as a way to increase memory bandwidth. Not only do these devices have NVIDIA and AMD as primary customers, but now Intel will be joining the fray with its Arc GPUs and this is a fairly “easy” way to increase performance, albeit likely with a higher price tag.
weird, why promoting a product for end-users when the end-users cannot buy or use it?
Those are all 32 bit wide, suggesting the higher end GPU will have at least 16 gigabytes RAM. The slower 14 to 18 Gbps chips are already available in quantity.
So what is the point of GDDR6X if GDDR6 can hit 24gbps? Would power still be lower for GDDR6X since it uses PAM4 to halve the signal rate?