JEDEC, a global organization that sets standards for the microelectronics industry, has released specifications for the GDDR7 memory standard.
The next generation of GDDR7 video memory is expected to be used in graphics cards based on the AMD RDNA 4 architecture and the Nvidia Blackwell architecture and is expected to be available next year.
The current generation of GDDR6 memory was first used in the NVIDIA RTX 20 series of graphics cards built on the Turing architecture, which was launched in September 2018.
The RTX 2080 and RTX 2080 Ti are among the first cards to use GDDR6 memory, with clock speeds of up to 14 GB/s and memory bandwidth of up to 56 GB/s.
Nvidia also uses GDDR6X memory in the RTX 3080, which runs at up to 19 GB/s, and in the RTX 4080 Super, up to 23 GB/s.
Micron's GDDR6X memory chips are among the fastest in the world, operating at up to 24GB/s and having a memory bandwidth of up to 96GB/s.
GDDR7 memory is expected to provide a huge increase in bandwidth. Benchmarks published by JEDEC confirm that the memory bandwidth is up to 192 GB/s, which is twice the speed of GDDR6X memory.
Interestingly, GDDR7 memory uses three signal levels (-1, 0, +1) to transmit three bits of data every two cycles. This is a big change from the NRZ signal used in GDDR6 memory, which moves two bits every two cycles.
This change alone means that data transfer efficiency is improved by 50%, as GDDR7 has twice as many independent channels (4 versus 2 for GDDR6), while still using PAM3 signals at a faster data transfer rate. higher. Performance improvement ring. the performance.
Finally, the following companies appear to be interested in GDDR7 memory: AMD, Micron, Nvidia, Samsung and SK Hynix. The first graphics cards based on GDDR7 memory are expected to be available in late 2024 or early 2025.