Samsung Reveals 24Gb GDDR7: Faster, Bigger, and More Efficient

Samsung has introduced a new 24Gb GDDR7 DRAM, promising it will raise the bar for GPUs and AI applications. With a significant boost in memory capacity and transfer speeds that surpass previous versions, the company is gearing up for a major shift in both consumer and professional markets.

Faster Memory, Up to 42.5 Gbps

Unlike older models, the latest GDDR7 memory modules from Samsung bring a 50% increase in cell density. Thanks to its 5th-generation 10nm tech, this capacity upgrade doesn’t require any changes to the size of the memory package itself. What this means is better performance without the need for more physical space in devices.

While GDDR6 was limited to lower speeds, Samsung’s new GDDR7 ramps up the pace. This DRAM reaches speeds of 40 Gbps under typical use, and it can hit up to 42.5 Gbps when fully optimized. The improved data transfer rates are expected to support more demanding tasks, making it a strong choice for GPUs and AI-focused data centers.

Samsung is aiming for consumer GPUs to benefit from this, especially with upcoming NVIDIA Blackwell graphics cards potentially using the tech. In comparison, earlier models only managed 28Gbps to 32Gbps, making the jump to 40Gbps quite an upgrade.

Energy Efficiency Overhaul

One of the standout features here is improved power efficiency. GDDR7 memory will use up to 30% less energy compared to its predecessor. This is achieved through tech like clock control management and a dual VDD design, both of which are borrowed from mobile products.

VDD stands for Voltage Drain Drain, which refers to the power supply voltage in electronic circuits. The dual VDD design is a power management technique that allows the memory to operate at different voltage levels depending on its current workload.

So while performance jumps, power consumption remains low—a major plus for users needing long-term efficiency.

Potential GPU and AI Uses

It’s expected that this 24Gb memory will not only power high-end GPUs but also find its way into AI workstations. Fields that need substantial processing power—think autonomous driving, AI, and gaming consoles—could see major benefits. Early validation is expected to take place at the end of this year, with a commercial rollout scheduled for early 2025.

However, while NVIDIA seems like a clear candidate to integrate this faster memory, AMD appears to be sticking with GDDR6 for now, as rumors suggest. The Radeon RX 8000 series is likely to prioritize cost-efficiency, potentially ruling out the use of this new, faster DRAM.

The AMD Radeon RX 8000 series is the upcoming generation of graphics cards from AMD, expected to launch in early 2025.

Built on the RDNA 4 architecture, it’s expected to deliver substantial improvements in both performance and efficiency, potentially making it a serious contender against NVIDIA’s offerings.

Architecture and Performance Expectations

The new RX 8000 series is built on AMD’s RDNA 4 architecture, bringing an estimated 50% performance increase over the previous RDNA 3 generation. Early reports suggest that the top-tier RX 8800 XT will feature 76 Compute Units (CUs) and around 4800 Stream Processors (SPs), alongside clock speeds nearing 2.5 GHz. It’s expected to be equipped with 24GB of GDDR6X memory, offering a memory bandwidth of up to 24 Gbps.

• Top model RX 8800 XT: 76 CUs, 4800 SPs, 24GB GDDR6X
• Mid-range RX 8700 XT: 60 CUs, 3840 SPs, 16GB GDDR6X
• Entry-level RX 8600 XT: 48 CUs, 3072 SPs, 12GB GDDR6X

Despite skipping GDDR7, these memory configurations suggest a strong focus on balancing performance and cost, especially when targeting consumers in the $400 to $500 price bracket.

One of the standout features of the RX 8000 series is the expected enhancement in ray tracing capabilities, potentially bringing AMD’s offerings closer to NVIDIA’s RTX 40 series. Additionally, a new Infinity Cache design with up to 96 MB capacity is rumored, which should improve overall performance in games and heavy workloads. The GPUs are also set to be more power-efficient, with a top-end TDP of around 200W, showcasing AMD’s continued focus on performance-per-watt improvements.

While AMD has opted to stick with GDDR6 instead of GDDR7, likely to keep costs down, the memory options ranging from 12GB to 24GB will still cater to a wide range of users. These cards should offer a compelling mix of high-end performance without the extreme power demands of their NVIDIA counterparts.

AMD is positioning this lineup to compete directly with NVIDIA’s upcoming RTX 50 series, focusing on a balance of performance and affordability to challenge the high-end GPU market.

NVIDIA RTX 50 Series: What We Know So Far

The upcoming NVIDIA RTX 50 series, based on the highly anticipated Blackwell architecture, is expected to push the limits of GPU performance once again. Built on TSMC’s advanced 4NP process, this next-gen lineup could also see some models transition to 3nm, signaling a major leap forward for gaming and professional graphics applications.

Rumored to deliver 2-3 times the performance of the current RTX 4090, the flagship RTX 5090 will lead the charge with as many as 24,576 CUDA cores and memory configurations offering either 24GB or 32GB of GDDR7. Lower-tier cards, such as the RTX 5080 and RTX 5070, will also see significant performance upgrades, bringing cutting-edge capabilities to a broader range of users. The RTX 5090 is also speculated to feature a 512-bit memory bus, which could substantially boost memory bandwidth.

Key Specifications Across the Lineup:

• RTX 5090: Up to 24,576 CUDA cores, 24GB or 32GB GDDR7, 512-bit memory bus
• RTX 5080: 10,752 CUDA cores, 16GB GDDR7
• RTX 5070: 6,400 CUDA cores, 12GB GDDR7
• RTX 5060: 4,608 CUDA cores, 8GB GDDR7

The inclusion of GDDR7 memory across most models hints at significant improvements in data transfer speeds, with the RTX 5090 potentially achieving 1.5 to 1.8 TB/s bandwidth—more than enough to handle the most demanding workloads in gaming, AI, and professional applications.

With Blackwell, NVIDIA aims to further enhance its ray tracing performance while building upon its existing AI-driven technologies, such as DLSS (Deep Learning Super Sampling). These improvements will likely make the RTX 50 series an even more compelling choice for gamers and professionals who require high fidelity in real-time rendering. Other advanced features expected in the series include support for DisplayPort 2.1, HDMI 2.1, PCIe 5.0, and enhanced AI-driven optimizations.

Power consumption for the high-end models is expected to be significant, with rumors indicating a total graphics power (TGP) between 400W and 600W for the RTX 5090. However, thanks to architectural advancements, NVIDIA is expected to deliver improved performance-per-watt, ensuring that the new cards will be more efficient compared to their predecessors.

The release of the RTX 50 series is expected to be staggered, with the RTX 5090 and RTX 5080 potentially arriving by the holiday season of 2024. Meanwhile, the lower-tier models like the RTX 5070 and 5060 are expected to follow in early to mid-2025. A full reveal at CES 2025 seems likely, where NVIDIA may unveil the official specs and performance metrics.

Last Updated on November 7, 2024 2:30 pm CET