Preliminary specifications for Nvidia’s GeForce RTX 5080 and RTX 5090 graphics cards have been shared by @kopite7kimi, a well-known hardware leaker. This source often provides accurate information about Nvidia’s upcoming products. If these specs are correct, the RTX 5090 will be a powerhouse with a 600W total graphics power (TGP). Many of these details match earlier leaks but offer a bit more information in certain areas. For further details, you can check out our Nvidia Blackwell RTX 50-series GPUs guide.
Nvidia’s GeForce RTX 5090, the top-tier Blackwell GPU for desktops, is expected to use the GB202 graphics processor. It will feature 21,760 CUDA cores and 32GB of GDDR7 memory with a 512-bit interface. This large number of FP32 cores will require a lot of power, possibly up to 600W if the leak is accurate. Even if the final specs change slightly, the card will likely be among the best when released. Just don’t expect it to come cheap.
In comparison, the Nvidia GeForce RTX 5080 is said to run on the GB203 GPU with 10,752 CUDA cores, about half of what the RTX 5090 offers. The RTX 5080 is expected to have 16GB of GDDR7 memory with a 256-bit interface and a TGP of 400W. Thanks to the GDDR7 memory, it should offer better performance at higher resolutions than its predecessor, making it a strong graphics card.
Preliminary specifications of Nvidia’s GeForce RTX 5000-series graphics cards
Here’s the information in bullet points for easy readability:
GeForce RTX 5080:
- GPU: GB203-400-A1
- CUDA Cores: 10,752
- Memory: 16GB 256-bit GDDR7
- TDP: 400W
- PCB Design: PG144/147-SKU45
GeForce RTX 5090:
- GPU: GB202-300-A1
- CUDA Cores: 21,760
- Memory: 32GB 512-bit GDDR7
- TDP: 600W
- PCB Design: PG144/145-SKU30
The GeForce RTX 5090’s huge performance is certainly impressive, but what stands out is the large gap between it and the smaller RTX 5080. The RTX 5080 has almost half the stream processors and a narrower memory interface compared to the flagship RTX 5090. Its TGP is about two-thirds of the 5090, which means the clocks may need to be higher to close the performance gap. This difference between the two cards is even larger than what we saw with the 40-series GPUs
The potential performance gap between the RTX 5090 and RTX 5080 seems unusual and suggests that Nvidia may be trying to create a new performance tier. It could also be a way to limit the use of certain consumer cards for AI computing. On paper, the RTX 4090 had 68% more GPU cores, 50% more VRAM, 41% more memory bandwidth, and 13% more L2 cache than the RTX 4080. However, in real-world use, the 4090 was only about 35% faster at 4K ultra settings. In comparison, the RTX 3090 was only around 15% faster than the RTX 3080. Based on the new specs, the RTX 5090 could deliver up to twice the performance of the 4080.
We don’t know for sure why Nvidia designed its next-generation lineup this way. One possible reason is that the GB202 processor in the RTX 5090 might be made up of two GB203 dies. Rumors about Blackwell GPUs using a multi-chiplet design have been around for some time. Nvidia’s GB100/GB200 datacenter GPUs already use this architecture. However, using CoWoS-L packaging to achieve a high-speed (~10 TB/s) connection between dies in a consumer-grade product could be very costly.
One option for Nvidia is to create a monolithic graphics processor with over 21,760 CUDA cores using TSMC’s 4nm-class process. This design would result in a chip size of about 650 mm². However, large die sizes are difficult to produce, so companies usually include redundancies. For example, the RTX 4090’s AD102 chip has 144 Streaming Multiprocessors (SMs), but only 128 are active. This means Nvidia could choose this route if it wanted to, but a monolithic chip would also be very costly. It would be odd to have such a large gap between the RTX 5080 and RTX 5090. Other GPUs might fill this gap, and we could see lower-tier models with around 18,000 functional CUDA cores.
We do know that Nvidia uses the same chips across different products. Desktop, mobile, professional, and data center GPUs all share similar designs. With AI being very popular right now, Nvidia might focus on creating a large data center part first, then adapt it for consumer use. If this is the case, we could see much higher prices than the already expensive RTX 4090. It’s also possible that AI variants will be released before the consumer models.
Right now, all the information about Blackwell-based graphics cards for client PCs is unofficial. It’s important to be skeptical. Until Nvidia provides official details, things can change. There are also mixed rumors about the release date. Some say the RTX 50-series won’t launch until early 2025. If this is true, it gives Nvidia plenty of time to make adjustments before the release. Until there is an official announcement, we can expect a lot of rumors and speculation about the specifications and features.
Hi, I’m deoravijendra, a professional content writer and digital marketer with 5 years of SEO experience. I’m passionate about crafting compelling content and optimizing online presence for maximum impact.