NVIDIA's upcoming GPUs are shaping up to be powerhouses, if recently leaked specifications hold true. Trusted hardware leaker @kopite7kimi has provided some early details on the company's next top-tier models, the RTX 5080 and RTX 5090.
Upcoming cards are part of NVIDIA's Blackwell series, and the leaks reveal not only impressive performance numbers but also extremely high power consumption, with the RTX 5090 reportedly drawing 600 watts of power and the 5080 needing 400 watts.
These specs indicate that the next generation of NVIDIA GPUs will be tailored for users who need immense computational power, likely targeting high-end gaming and resource-heavy applications like AI development.
Unpacking the Specs: Performance and Power
The RTX 5090 stands out for its sheer processing power, boasting 21,760 CUDA cores and 32GB of GDDR7 memory, tied together by a 512-bit bus. In contrast, the RTX 5080 offers half as many cores—10,752—but still provides 16GB of GDDR7 memory with a 256-bit interface. All the data a clear performance gap between the two models, even wider than what was seen with NVIDIA's previous 40-series cards.
GeForce RTX 5090
PG144/145-SKU30
GB202-300-A1
21760FP32
512-bit GDDR7 32G
600W— kopite7kimi (@kopite7kimi) September 26, 2024
Both cards bring substantial increases in bandwidth, with the 5090 reaching an eye-watering 1568 GB/s compared to the 5080's 784 GB/s. That difference, combined with the core count disparity, signals that the 5090 is meant for users who demand top-tier performance across the board, from high-end gaming at 4K to GPU-intensive workloads like rendering and AI.
Power Demands and Potential Trade-Offs
What sets the new Blackwell series apart isn't just performance; it's also the massive power draw. The RTX 5090, for instance, requires a 600W power supply, which will likely necessitate advanced cooling solutions. Those looking to install this card will need to ensure their systems are equipped to handle both the power requirements and the heat output that comes with it.
On the other hand, the RTX 5080 requires a slightly more manageable 400W, though it still represents a notable jump compared to its predecessor, the RTX 4080, which draws 320W. Gamers and professionals alike will need to consider upgrading their power supplies and cooling systems to support these cards, especially if planning to overclock or run intensive applications for extended periods.
Speculation on Design: Monolithic or Multi-Chiplet?
One of the most intriguing aspects of the leak is the potential design of the RTX 5090. Rumors have suggested that NVIDIA may opt for a multi-chiplet architecture for its top-tier card, allowing for better performance scaling while managing heat more effectively. This would mark a departure from the traditional monolithic designs used in consumer-grade GPUs and could signal a shift toward technologies typically reserved for data center products.
If true, this would explain the stark contrast in power and performance between the RTX 5090 and 5080. However, building such a product comes with challenges, not least of which is cost. A chip of this complexity could be expensive to manufacture, and NVIDIA might be forced to push prices even higher than the $1,600 launch price of the RTX 4090.
Challenging Path Towards Launch
Initially presented at Nvidia's GTC conference, the Blackwell architecture is set to significantly outperform its H100-class predecessor, offering 2.5 to 5 times the performance and doubling both memory capacity and bandwidth. While a mid-2024 release was initially planned, the schedule has now shifted to Q4.
Manufacturing issues that necessitated a mask change led to production delays, as disclosed during Nvidia's Q2 earnings call. CEO Jensen Huang acknowledged these setbacks and their effects on stakeholders, asserting that Nvidia has ramped up production to satisfy strong demand.
