Video Memory GDDR6

Samsung – GDDR6 memory is already very close


The new module has already received the award. Samsung boasted that its 16-gigabit system memory GDDR6 won the Best Innovation Award CES-2018, in its category.

This is a system that has the designation K4ZAF325BM-HC14 – it’s 16 Gigabit bones GDDR6, working with a voltage of 1.35 V. This is the first chip of this kind from Samsung, which, however, is still under development, but the company has prepared sample specially for presentation.

Samsung can boast that its 16 Gbps GDDR6 system works at speeds up to 16 Gbps, reaching a bandwidth of 64 GB / s. If you apply this kind of memory on a video card with a GPU that has a 384-bit memory interface, then the entire subsystem would have a throughput of about 768 Gbit / s. This is twice as much as in 8 Gb GDDR5 chips, which reach a maximum of 384 GB / s.

Graphics GDDR6

It is not yet known when Samsung will be ready to launch mass production of new GDDR6 systems. Nvidia has already announced that on January 7, as part of the CES 2018 exhibition, Jen-Hsun Huang (company manager) will make a big presentation. And in March, the next GTC conference (GPU Technology Conference) will be held. Early next year, Nvidia will unveil the next generation of GeForce series 2000 video cards with processors based on the Volta architecture. It is possible that this series will already appear with GDDR6 memory systems.

Geforce GDDR6

A small comparison of GDDR5 and GDDR6:


  • Density: 4Gb ~ 8Gb vs 8Gb ~ 16Gb;
  • Package: 170B (12×14) 0.8mm pitch vs 180B (12×14) 0.75mm pitch;
  • Burst Length: 8bit (DDR) vs 16bit (DDR/QDR optional);
  • External IO: X32 (x16) vs 2CH x32 (2CH x16 & PC Mode x32);
  • WCK Granularity: byte vs Byte / word (optional);
  • Pre-fetch per CH: 256bit (32GB access) vs 256bit (32B);
  • WCK Rate: 2f (DDR) vs 4f (DDR) / 2f (QDR);
  • Pin Data Rate: 8Gbps (Target 10Gbps) vs Up to 16Gbps;
  • IO: POD vs POD;
  • Voltage: 1.5V (1.35V) vs 1.35V;
  • VPP: – vs 1.8V;
  • Rx: CTLE per word vs 1-tap DFE per DQ;
  • CA Training @SRF: NO vs YES;
  • EDC Rate: Full (0x83 72bit) vs Full / Half (0x83 144bit / 0x83 XOR).

Leave a Reply