At its annual Technology Day, Samsung detailed some of the technologies that will enable 36 Gbps for its upcoming GDDR7 graphics memory (opens in new tab). This is a very healthy boost over the current fastest GDDR6 and GDDR6X memory.

This 36 Gbps figure is also only a starting point. By comparison, GDDR6 started at 14 Gbps; the key to GDDR7's speed boost is the revised use of PAM (pulse amplitude modulation) signals for data transfer

.

Simply put, the zeros and ones of binary computer data are converted into electrical waveforms as they are transmitted from, say, a graphics memory chip to the GPU; the PAM-2 signal implemented in GDDR6 allows one bit per transmission cycle, while in GDDR6X the PAM -4 version is used, which doubles to two bits per cycle. [GDDR7 uses the PAM-3 signal instead, which supports 3 bits per cycle, resulting in an immediate 50% bandwidth speedup at any operating frequency. PAM-3 is also said to be 25% more energy efficient, which is useful given the rapid power consumption of modern graphics cards. [To put a number on how this affects available bandwidth, the new Nvidia GeForce RTX 4080 (opens in new tab) with 256-bit bus and 22.4Gbps GDDR6X has a raw bandwidth of 717GB/s. With GDDR7 running at 36 Gbps on the same 256-bit bus, the figure is 1,152 GB/sec.

And what about the RTX 4090 (opens in new tab) and the 21 Gbps GDDR6X running on a 384-bit bus; at GDDR7, the 4090 would rock 1,728 GB/sec. Fantastic.

As for when we might see GDDR7 on purchasable graphics cards, it is unclear if Nvidia's new RTX 40 series or AMD's RDNA 3 GPUs will support GDDR7 in the future. Earlier this year, there were rumors that the RTX 40 series would feature ultra-fast GDDR7 technology. However, the RTX 40 series and AMD's latest GPUs ultimately stayed at GDDR6X and GDDR6, respectively.

We also recently reported on Samsung's new GDDR6W (opens in new tab), which doubles the memory density and speed per memory package.