One of the main appeals of GPU refreshes is the type of RAM configuration that will accompany next-generation products. To that end, Samsung is laying claiming to the "industry's highest capacity" GDDR7 with what it says are the first 24-gigabit (Gb) GDDR7 DRAM chips, which will find their way into numerous future hardware designs, including graphics cards and game consoles.
"With its high capacity and powerful performance, the 24Gb GDDR7 will be widely utilized in various fields that require high-performance memory solutions, such as data centers and AI workstations, extending beyond the traditional applications of graphics DRAM in graphics cards, gaming consoles, and autonomous driving," Samsung says.
A 24Gb memory chip is equivalent to 3 gigabytes (GB). This paves the way for big increases in VRAM, depending on how companies like NVIDIA and AMD want to implement the chips. For example, a lower end graphics card with four 24Gb GDDR7 modules would flex 12GB of fast VRAM, while a card with 12 memory chips, which is how many are found on the GeForce RTX 4090, would tout 36GB of VRAM. A card with 16 memory chips works out to 48GB.
"We will continue to lead the graphics DRAM market by bringing next-generation products that align with the growing needs of the AI market," said YongCheol Bae, Executive Vice President of Memory Product Planning at Samsung Electronics.
Samsung says its 24Gb GDDR7 chips are built on the company's 5th generation 10-nanometer class DRAM, paving the way for a 50% increase in density without increase the package size from the previous generation. They also feature tri-level Pulse-Amplitude Modulation (PAM3) signaling for blazing-fast speeds of up to 40 gigabits per second (Gbps), the same as what
SK hynix flexed earlier this year. That's a 25% jump over the previous generation. Additionally, Samsung says its sees a path to 42.5Gbps "depending on the usage environment."
While certainly exciting, we may not see these specific chips implemented on the upcoming GeForce RTX 50 series, or in particular, the
GeForce RTX 5090. NVIDIA's next flagship consumer GPU is rumored to feature 32GB of GDDR7 at either 28Gbps or 32Gbps, linked to a fat 512-bit bus. If so, we'd be looking at 1.792TB/s to a little over 2TB/s of memory bandwidth.
Employing Samsung's 24Gb chips would likely bump those specs up to 36GB (12 x 3GB) of GDDR7 VRAM, and at 40Gbps, memory bandwidth would jump to 2.56TB/s. At 42.5Gbps, memory bandwidth would be even higher, at 2.72TB/s.
That's just having some fun with math, though—don't expect the GeForce RTX 5090 to be equipped as such. Nevertheless, Samsung does specifically mention graphics cards and game consoles as landing destinations for its latest GDDR7 memory chips, so we'll see how it all plays out.
In the meantime, Samsung says "major GPU customers" will being the process of validating its
24Gb GDDR7 for use in next-gen AI computing systems this year, and deploy solutions in early 2025.