'%20x='0'%20y='0'%20height='100%25'%20width='100%25'%20%0A%20%20%20%20%20%20%20%20%20%20xlink%3Ahref='data:image/jpg;base64,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'%3E%3C/image%3E%3C/svg%3E)
Samsung unveiled a newer GDDR6W memory standard that can deliver up to twice the performance, memory capacity, and memory bandwidth as compared to the standard GDDR6 and GDDR6X modules.
The best part of these memory modules is that they can be flawlessly used in consumer products like graphics cards and future consoles as they use the same ball grid array (BGA) packaging as the older GDDR6 ones. In contrast, HBM2E uses a much costlier micro pillar grid array (MPGA) packaging.
GDDR6W takes the memory standard closer to HBM2E, which is a much costlier alternative and is only used in industry-grade products like data server GPUs. With a much higher performance output, the altered variant of the VRAM modules will power next-gen computer hardware.
How do GDDR6W VRAM modules work?
The GDDR6W memory module utilizes two DRAM devices which effectively doubles the interface from 32 bits to 64 bits. It also means that for the same die, manufacturers can double the capacity. This development means two things.
First off, devices with more memory capacity can be manufactured without significantly increasing the PCB or die size. Thus, future graphics cards might come with more than 24 GB of VRAM, unlike the RTX 4090.
Secondly, manufacturers can cut costs on manufacturing devices with the same amount of VRAM when pitched against a GDDR6-powered product.
The newer memory modules are also $36 smaller in size thanks to recent improvements in packaging techniques that Samsung aptly calls Fan-Out Wafter-Level Packaging (FOWLP) technology.
This new technique is a breakthrough in the memory industry as it gets rid of the traditional PCBs that are often involved in connecting VRAM modules. It has been replaced with a redistribution layer (RDL), which is a two-layered thin connecting technology for the new memory standard. These layers are connected to fine copper posts that help in reducing the die height from 1.1mm to 0.7mm.
However, this sophisticated packaging technique is more expensive than placing the memory chips on a PCB. But the newer chips will be significantly cheaper than the much more expensive HBM2E memory standards that experts were considering as the next step up in memory module technology.
The newer memory modules also come with twice the memory bandwidth. While a 32 GB GDDR6 memory chip has a peak bandwidth of, say, 88 Gbps, the newer modules can deliver up to 176 Gbps.
Samsung is currently using the new memory standards in industrial applications. They are employed in artificial intelligence (AI) hardware and high-performance computing accelerators.
AMD and Nvidia have not announced the usage of the new memory modules in their graphics cards. However, Samsung has notified that they plan on working with GPU partners to integrate GDDR6W memory modules into consumer-grade graphics cards.
It can be safely assumed that the current-gen RTX 40 series and Radeon RX 7000 series GPUs will not get GDDR6W. In that case, it will be a full two years before these new memory modules start hitting the PC gaming hardware market.
