Samsung’s 24 Gb GDDR7 Memory Can Pave The Way For Higher-Capacity Consumer GPUs, Up To 42.5 Gbps Speeds

Samsung has developed its next-gen 24 Gb GDDR7 memory solution, offering higher VRAM capacities & speeds of up to 42.5 Gbps.

Samsung Plans To Bring Higher Capacity "24 Gb" GDDR7 Memory To Consumer GPUs Early Next Year, Up To 42.5 Gbps & 32 Gb Modules Also Hinted

Samsung was among the first memory makers to announce its next-gen GDDR7 memory last year. Initially, Samsung listed speeds of up to 32 Gbps and 16 Gb flavors (2 GB). As we have recently reported, the first round of GDDR7 memory will likely be rated at speeds ranging from 28 to 32 Gbps, but it looks like Samsung has faster and higher-capacity memory in the works already. The Samsung solutions were also slated to achieve up to 37 Gbps speeds but it looks like the company is setting new performance standards for the industry.

In today's announcement, Samsung confirms that it has developed the industry's first 24 Gb GDDR7 DRAM which would utilize its 5th Generation 10nm process technology, offering up to 50% increase in cell density while maintaining the same package size as the predecessor aka GDDR6. Samsung's GDDR7 memory would also incorporate clock control management & a dual VDD design to improve power efficiency by over 30%.

After developing the industry’s first 16Gb GDDR7 last year, Samsung has reinforced its technological leadership in the graphics DRAM market with this latest achievement,” said YongCheol Bae, Executive Vice President of Memory Product Planning at Samsung Electronics.

We will continue to lead the graphics DRAM market by bringing next-generation products that align with the growing needs of the AI market.

The 24Gb GDDR7 utilizes 5th-generation 10-nanometer (nm)-class DRAM, which enables cell density to increase by 50% while maintaining the same package size as the predecessor.

via Samsung

With 24 Gb DRAM, GPUs will have access to 3 GB of memory capacities compared to 2 GB provided by 16 Gb DRAM. This will lead to a nice uplift in memory capacities. A 128-bit GPU can scale up to 12 GB of VRAM, a 256-bit GPU can scale up to 24 GB of VRAM, a 384-bit GPU can scale up to 36 GB of VRAM while a 512-bit GPU can scale up to 48 GB of VRAM.

At the same time, Samsung also promises even higher GDDR7 memory speeds by leveraging PAM3 signaling. A 25% improvement over the current GDDR6 memory is promised with 40 Gbps speeds but in certain products, the memory would be able to scale up to 42.5 Gbps speeds, a 2.36x uplift over 18 Gbps GDDR6 modules.

One should also mention that Samsung announced a faster variant of its GDDR6 memory called GDDR6W which was going to double the channel to 64-bit and offer increased densities back in 2022 however we didn't see any consumer-level implementation for that. The company also had 24 Gbps memory ready within its GDDR6 portfolio but the fastest GDDR6 solution that we got was the RTX 4080 SUPER with 23.0 Gbps speeds based on Micron dies.

Following are the speeds that you can expect from 42.5 Gbps memory modules:

• 512-bit - 2720 GB/s (2.7 TB/s)
• 384-bit - 2040 GB/s (2.0 TB/s)
• 320-bit - 1700 GB/s (1.7 TB/s)
• 256-bit - 1360 GB/s (1.4 TB/s)
• 192-bit - 1020 GB/s (1.0 GB/s)
• 128-bit - 680 GB/s

Recent reports that the NVIDIA GeForce RTX 5080 GPU might come in a 24 GB capacity option make a lot more sense given that Samsung has already planned commercialization of 24 Gb modules by early 2024. As such, we can see other memory makers such as Micron & SK Hynix follow suit, allowing for higher capacities and speeds on both mainstream and high-end GPUs.

GDDR Graphics Memory Evolution: