GDDR SDRAM is one of the key core components of a graphics card, and its strength and capacity will directly affect the final performance of the card.

Ⅰ What is GDDR?

GDDR stands for Graphics Double Data Rate, a high-speed SDRAM used in graphics cards. GDDR SDRAM is a high-performance DDR memory specification designed specifically for high-end graphics cards, with an exclusive operating frequency, clock frequency, and voltage, so it is different from standard DDR memory on the market. Generally, it has a higher clock frequency and generates less heat than the normal DDR memory used in the main memory, so it is more suitable for use with high-end display chips.

Samsung GDDR6

High-performance DRAM designed for graphics cards has two main features, one is high density addressing capability, which means the capacity of the particles should be high to meet the memory capacity requirements of the graphics card and the requirements of the graphics card design. High-density graphics memory addressing is only half of the problem the design has to face, the other half is performance. The graphics memory must be able to provide high-speed transfer, now GDDR can provide more than 1GT/s data transfer rate.

Ⅱ GDDR vs. DDR vs. LPDDR

DDR is actually DDR SDRAM, called Double Data Rate Synchronous Dynamic Random Access Memory.

GDDR is the abbreviation for Graphics Double Data Rate, which is also the double data rate memory for display, used in graphics cards. It is used as the memory for the graphics card. It runs faster than DDR to display video and pictures more smoothly.

DDR is a kind of DRAM, the standard for memory particles, while GDDR is the standard for graphics memory particles.

In fact, DDR and GDDR, DDR2, and GDDR2 technology are similar, and they can be replaced with each other. However, as the rate is increased, DDR is developed towards high-rate high rates. GDDR is developed toward small capacity and higher rate. GDDR is mainly used to display memory storage.

When the memory matches the CPU, the CPU needs to make a lower delay in the task that does not need type. Storage particles should transmit a small amount of data extremely quickly. The graphics card needs to be high throughput, and the storage particles should transmit very large amounts of data.

As a result, DDR and GDDR develop in different directions in the later standards. GDDR memory is optimized for data stored on graphics, which can withstand greater throughput. DDR is optimized for the CPU, which can store the applications and data currently processing, with lower delays.

GDDR and DDR BW Roadmap

LPDDR is based on DDR with the extra LP (Low Power) prefix, also known as mDDR (Mobile DDR SDRAM). Time back to the era of Android 1.6 and IOS 3.0, the DDR SDRAM used in early cell phones could not meet the needs of smartphones no matter the speed or power consumption. So, LPDDR, the memory that has the lower power consumption and smaller volume emerged.

Since the usage scenario of LPDDR is closer to DDR, the development process of LPDDR has a greater intersection with DDR. By the LPDDR3 era, LPDDR3 began to be adopted by notebook platforms, including Mac Book air, for its good performance and even better power consumption ratio.

LPDDR4 began to establish its own specification system in accordance with the needs of its own application scenarios, and LPDDR5 completed mass production before the DDR5 standard was officially finalized, surpassing DDR4 in theoretical data transfer performance and taking a different path from DDR memory.

The following image has made a DDR family development route according to the product delivery time of DDR memory.

DDR family development route

Ⅲ GDDR Generations

So far, the GDDR family includes GDDR, GDDR2, GDDR3, GDDR4, GDDR5, and GDDR6. GDDR5 and GDDR6 are the most common types of memory on the market today. Other types of GDDR have been slowly eliminated from the market.

GDDR

Like DDR, GDDR uses 2-bit prefetching and can transfer data once on the rising and falling edges of a clock cycle, so as to achieve twice the transfer rate of a single time cycle. However, the GDDR frequency is not much higher than DDR, but GDDR has relaxed the latency requirement in order to pursue higher frequency.

GDDR particles are in the pursuit of large bit width, so the capacity will be particularly small. At that time, specifications are all 8 × 16Bit. That is, the size of a GDDR memory is only 16MB. The same period of DDR memory particles can reach 32 or even 64MB.

Later, due to the upgrade of the lithography process and the pursuit of more packaging area, storage manufacturers implemented BGA packaging technology on GDDR, which means that the pins are not exposed and are all hidden at the bottom. At the same time, the 32bit memory bit width was achieved and has been continued to this day.

GDDR2

GDDR2 is packaged in BGA (Ball Grid ArrayPackage). The speed of the memory varies from 3.7ns to 2ns, and the maximum default frequency is from 500MHz to 1000MHz, but it is obviously inferior to GDDR3 memory. Its single-particle bit width is 16bit, and it takes 8 pieces to form a 128bit specification.

GDDR3

GDDR3 is a type of memory developed specifically for graphics processing, also using BGA packaging technology. Its single-particle bit width is 32bit, 8 particles can form 256bit/512MB of memory bit width and capacity. The memory speed is between 2.5ns (800MHz) and 0.8ns (2500MHz). Compared to GDDR2, GDDR3 has three advantages, which are low power consumption, high frequency, and large single capacity.

GDDR4

GDDR4 and GDDR3 are basically the same technologies. GDDR4 single memory particles can achieve 64bit bit width 64MB capacity, that is to say, only 4 memory chips can achieve 256bit bit width and 256MB capacity. 8 memory chips can easily achieve 512bit bit width 512MB capacity. GDDR4 memory particles are currently concentrated at a speed of 0.7ns to 0.9ns. However, GDDR4 has an overall sequence, and the GDDR3 of the same frequency is existently leading to the product of GDDR4 memory. And the GDDR4 memory does not solve high power, high fever problems because the voltage is lower. The result has led to the lack of advantages over GDDR4 with GDDR3.

AMD Radeon HD 2600 XT GDDR4

GDDR5

Compared to GDDR3 and GDDR4, GDDR5 SDRAM has many technical advantages including higher bandwidth, lower power consumption, and higher performance. If paired with the same number of memory particles with the same memory bit width, the total bandwidth provided by GDDR5 memory particles is more than 3 times that of GDDR3. Since GDDR5 memory can achieve higher bit widths, it means that graphics cards with GDDR5 memory will have more flexibility and a significant increase in performance. Therefore, all mainstream graphics cards without exception are using GDDR5.

GDDR6

GDDR6 and GDDR5X use the same 16bit prefetch, which has been proven to be the most effective way to improve data transfer speed. GDDR5X/5/4/3/2/1 memory versions are single-channel read/write design, while GDDR6 memory uses dual-channel read/write design. The operating voltage has been reduced compared to the previous generation of GDDR5 memory, which means that GDDR6 is not only more powerful but also consumes less power. Memory bandwidth can reach 16Gbps × 384bit ÷ 8 = 768GB / s. Take GTX1080 for example, its memory bandwidth is 11 * 256/8 = 352GB, almost double the bandwidth.

GDDR6X

GDDR6X has a higher memory bandwidth

The main difference between GDDR6X and GDDR6 is in their equivalent frequencies, GDDR6 has an equivalent frequency of 14~16Gbps, while GDDR6X has an equivalent frequency of 19~21Gbps, which means that with the same memory bit width, the memory bandwidth of GDDR6X increases by 35% compared to GDDR6, reaching 912GB/s~1008GB/s.

GDDR6X has a higher power efficiency ratio

Because GDDR6X adopts a completely different PAM4 signal modulation method with MTA coding technology, GDDR6X not only has higher memory bandwidth but also has a better energy efficiency ratio. According to Micron's data, GDDR6X consumes 15% less energy per bit than GDDR6, while GDDR6X also provides 35% more memory bandwidth.

Since GDDR6X memory was developed by NVIDIA and Micron, it has not yet been standardized by JEDEC, so there is no second manufacturer that can produce GDDR6X memory. Now only the GA102 core RTX3080 and RTX3090 cards use GDDR6X memory, RTX3070 models below are still along with the GDDR6 memory.