The fact is that with these GDDR6X VRAMs, Micron gets a lot of marketing in terms of the first company that is able to offer graphic discrete memory that exceeds the speed of TB / s, although it is not really visible yet.
And that this technology has been around for 14 years, at which point new ways to increase speed in VRAM were being learned.
Micron VRAM GDDR6X, extra speed and high production volume
The biggest secret to Micron GDDR6X is IPAM 4 o 4-level Pulse Width Modulation. The technology is constantly being updated and their climax is up to 21 Gbps for this type of dedicated memory.
When Micron engineers are able to work with PAM 4 they have reached a new level of complexity, because technology is useless if you cannot reduce costs and do it with volume. Producing the GDDR6X VRAM with PAM 4 is not a Micron problem, but it was a problem to do it with the sufficient volume required by NVIDIA.
But what makes the GDDR6X so new without the speed achieved? and above all, how did you do it? There are three key issues to consider:
- Double bandwidth with low cost per transaction to achieve 1 TB per second of memory width.
- The already developed PAM 4 uses four voltage levels to encode and transmit up to two bits of data per clock.
- It can be built and installed in GDDR6 systems with multiple products.
To this should be added the fact that DRAM times have been reduced by doubling the data. This has made the demands of the same memory members easier, but the features inside the chips have increased exponentially, when for example at 16 GB / s the recording and transfer time has fallen. 62.5 picosegundos.
Therefore, the acceleration has also resulted in waves bringing difficulties to the accuracy and precision of circuits in which they exchange data. Also that transferring 2 pieces of data at the same time leads to the same speed compared to GDDR6 circuits can work in half and with greater clarity.
PAM 4, or how voice fluctuations affect speed
As we have said, managing 2 pieces per cycle requires a series of small, but complex structural changes. PAM 4 solves most of that complexity by adding 4 different signal levels known as symbols and the unit of measurement for symbols per second.
Therefore, where GDDR6 requires 16 lengths to transfer data via NRZ, the PAM 4 with GDDR6X will have half a chain logically using a few cycles. Also, the GDDR6X requires two different operating modes to operate at high-end clocks.
With PAM 4 the memory clock is similar to a command clock, which simplifies the way to send data without the need for NRZ. The data driver contains 2 broad UI, of which all 8 symbols are contained 2 pieces of data in 2 command cycles per hour, offering 8 Tail Brust Length.
This has a direct effect on a frequency called WCK or CK. From the GDDR5 to the GDDR6X, the type of memory retains a very clear standard for the four existing VRAM versions:
- Literacy access has two clocks per cycle, or in other words, they have a tCCD of 2 TCK. Therefore, we will only have access to the use of 100% VRAM buses when reading and writing are done in cycles per second.
- Although the data in CMD, ADDR and DATA are similar to GDDR6 and GDDR6X, the latter requires half CK frequency with the same functionality.
- It also shares with GDDR6 the fact of receiving commands and addresses on Double Data Rate (DDR) instead of SDR.
If this is not enough, we must also address the perceptions of a performance perspective.
Better energy efficiency at higher speeds
Micron doesn’t say much in this section, but it should be mentioned as it stands by looking at GDDR6X temperatures on Ampere cards.
According to the company, GDDR6X at 21 Gbps requires ia 15% less power Each transmission is compared to the GDDR6 operating at 14 Gbps, which makes a difference in the width range 50%.
We have to assume that the fact that we need a low CK frequency, and two bits of data per UI is very much related to this, because in Vpp view we are talking about the same output of 1.8V and Vdd and Vddq are the same compared to GDDR6: 1.35 volts and 1.25 volts respectively.
So we can’t shed much light on this, and we’ll have to stick to the statistics provided by Micron and NVIDIA, where it still seems like the GDDR6X is hot as a rule.
So, and in summary, the development of this new VRAM is key, where it is included in certain categories that do not reduce its compatibility with its previous version and instead, provides more rapid development and efficiency of NVIDIA graphics cards. Will AMD be able to integrate it into its new RDNA 2 GPU?