The South Korean Giant has started the development of DDR6 memory, a next-generation standard that, as you may have imagined, will succeed the newly released DDR5. As is customary with each new generation, this standard will mark an important evolution in terms of performance, and also efficiency, at least in theory, as it is still taking “its first steps”.
DDR6 memory will be planted in impressive 12,800 MHz frequency, a figure that we must consider as the base level of this new generation. This means that this will not be the ceiling of DDR6, but the starting point that will appear in the JEDEC specifications. According to the source of the news we may meet maximum speeds up to 17,000 MHz, a truly impressive figure. To put a bit of context, just remember that, right now, with 3200 MHz DDR4 memories and CL14 latencies we already enjoy excellent performance.
Compared with DDR5 memory, the jump that DDR6 memory would make would also be quite large, since that would peak, in theory, at 8,500 MHz. However, we must bear in mind that it is possible that variations will end up in the final frequencies, so we are going to take the figures that we have given with caution, since some sources say that the real ceiling of the DDR6 will be in the 20,000 MHz (20 GHz).
DDR6 memory will arrive between 2025 and 2026, and before that we will see GDDR6 +
There are still several years to go before the debut of DDR6 memory occurs, but it is interesting to begin to know its main keys, and to get an idea of everything that this new standard can offer. Personally I think that, thanks to this enormous increase in working frequencies, we will see a major performance improvement for integrated GPUs that they turn to RAM to use it as graphics memory, since DDR6 will offer higher bandwidth.
And speaking of GPUs, Samsung has also revealed that it is preparing new GDDR6 + memory chips, a solution that will coexist with GDDR6 and that will be able to achieve frequencies over 21 GHz. So far, Micron has been the only one to hit the 21 GHz barrier with GDDR6X memory, a standard that NVIDIA has used in its more powerful GeForce RTX 30-series graphics cards.
We do not have confirmation of when the first Samsung GDDR6 + memory chips will be available, so we do not know if they will arrive in time to integrate into the next generation of graphics from NVIDIA and AMD, provisionally known as GeForce RXT 40 and Radeon RX 7000. All in all Yes, we are clear that with the memory operating at that frequency and a 384-bit bus it would be possible to achieve an approximate bandwidth of 1 TB / s, while with a 256-bit bus we would be in the 768 GB / s.
On the other hand, memory GDDR7 It is a project that is also taking its first steps, and that promises an important generational leap, since in theory it could reach the 32 GHz frequency. With that speed, a memory system based on a 256-bit bus could reach 1 TB / s of bandwidth, while in systems with 384-bit and 512-bit buses we would have a total bandwidth of 1.5 TB / s and 2 TB / s.
I do not want to end without reminding you that, in addition, this new generation of graphics memory should also allow an increase in density per chip, which it would go from the current 16 Gb (gigabits) to 32 Gb. This means, in short, that it will allow doubling the maximum amount of VRAM integrated in a graphics card without doubling the space occupied on the PCB. It is a very important detail, since this type of memory is distributed horizontally (2D).