As the demand for graphics cards has increased, so too have the requirements for improved memory. Graphics card manufacturers are looking to HBM2 as a way to meet those demands. But is it worth it? Or will GDDR6 be what we need to stay ahead of the game?
This question can depend on your specific needs, but some general trends will help you make an informed decision. This blog post will look at these two types of memory and break down the differences between them.
HBM stands for high-bandwidth memory, and GDDR is short for graphics double data rate. HBM offers a substantial bandwidth increase over traditional buffer systems that store data in the GPU’s onboard RAM or system memory, as low as 14 GB/s to 50GB/s with DDR (double data rate) DRAM. This means that two pixels are processed at once by an individual cell of the die.
Thus, theoretically eliminating bottlenecks while more quickly processing information without consuming power. Whereas GDDR, just like its name implies, operates at twice the speed compared to standard DDR SDRAM, meaning it has a throughput of 16Gbps per pin versus 11 Gbps respectively – a 33% increase.
Despite the difference in these two types of memory, HBM is still costly to implement. GDDR does not offer a significant bandwidth advantage over DRAM when compared at similar speeds.
Overview of HBM2:
Standard specifications of the memory:
HBM is used on the high-end graphics card and other similar, advanced graphics units. Typical use cases include highly banded memory interfaces with low latency.
In terms of construction, HBM memory offers a non-planar structure with a 3D design. The design follows a cube or cuboid structure– stacking up memory chips one over the other to create a cubed look. The improved design allows for less space while packing in more performance.
Place it close to the CPU for enhancing the processing speed. The HBM signaling specification is less power-hungry than other standards.
With the HBM2 as a standard, it is clear that this technology is the second generation of high bandwidth memory. In contrast to the HBM standard, the HBM2 offers options for higher memory capacity and better bandwidth.
The HBM2 has 8 DRAM dies per stack, offers up to 2 Gbps speeds, and is first seen in Nvidia’s Tesla P100 graphics cards.
The HBM2 standard has become relevant due to its support of virtual reality, augmented reality, and other large-memory intensive applications such as machine learning or neural processing. The following graphics cards support the HBM2 standard:
Titan V, Vega Frontier Edition, Vega RX 64, and Quadro GP100.
Characteristics of HBM2:
HBM2 reduces power consumption by providing a faster data transmission speed
The memory usually takes up a smaller form factor, saving space.
HBM2 features a new, more efficient design without the need for a bulky cooler
HBM2 provides a higher bandwidth option to complement GDDR6
The high-bandwidth memory interface has greater bandwidth
The price of the memory could be pretty high
Overview of GDDR6:
GDDR6 is the newest graphics standard, succeeding GDDR5 and GDDR5X. GDDR6, a recently developed memory standard, provides significantly faster speeds than HBM2. The GDDR6 reaches up to 16 Gbps and offers speeds ranging from 1.35 V to 2.0V.
Dimensioning for the bandwidth offered by GDDR6 is usually a factor of 72 GB per second; however, this can be adjusted according to the user’s needs.
The standard is designed on the 10-nanometer manufacturing process and offers a higher density of 32 GB per die. It is expected to be used for graphics cards from Nvidia, beginning with Volta and Turing ranges. DDR4 and GDDR6 are designed for gaming, VR, and cryptocurrency mining.
Samsung, Micron, and Hynix manufacture GDDR6 memory. DDR4 cards made by Samsung and Micron max out at 16 Gbps; this will satisfy the needs of advanced users. To cater to mainstream application areas and offer speeds that fall between 10 and 14 Gbps, Hynix will be manufacturing memory standards.
GDDR6 is an alternate DDR3 update designed and implemented by Samsung and Micron. The style is made available on the Turing architecture first, offering 16 and 32 GB dies instead of the 8 GB found in the GDDR5 family.
Some graphics cards with the GDDR6 memory standards include – Quadro RTX 8000, Quadro RTX 6000, Quadro RTX 5000, GeForce RTX 2080 Ti, RTX 2080, Ren Huang RTX (Ren Huang 2060), and Nvidia TITAN RTX.
GDDR6 offers functionality and performance that’s better than HBM2. GDDR6 is the latest iteration of memory standard, succeeding GDDR5 and HBM2. It provides rates that are faster than the latter.
Detailed Comparison between HBM2 and GDDR6:
Ultimately – for those looking for an immediate improvement without having to transition their systems or who have been waiting on updates from AMD’s Polaris architecture – this might be your best option.
But suppose you’re willing to wait out that particular storm or are happy with what you currently have. In that case, it may be worth sticking with traditional GPU RAM like DDR SDRAM (double data rate synchronous dynamic random access memory).