The Nvidia RTX 5090 is one of the most hotly anticipated graphics cards, and the excitement only grows as we move closer to its release. Recently, a boatload of details were leaked about the GPU, shedding light on its rumored capabilities and discarding many earlier speculations circulating online. The leaks present new information about the GPU and the architecture that powers it.
In this article, we'll analyze the leaks, explain what they mean and what kind of performance you can expect.
Note: Aspects of this article are based on leaks and are subject to change.
What is the Blackwell technology powering the RTX 5090?
The upcoming RTX 5090 will be based on the Blackwell architecture, named after David Blackwell, a Howard University mathematician. He published over 90 papers on game theory, statistics, and more.
The Blackwell architecture brings in 5th-generation Tensor cores, a new Decompression Engine, an RAS engine, and more. The new Decompression Engine could come in real handy in future games. It uses the GPU to decompress and load assets from the SSD, which should reduce issues such as texture pop-in and traversal stutters.
Such issues linger in many Unreal Engine games, even today. However, a title must be designed to use that feature, or else you may not see its full benefits.
Previously, I wrote about a leak that suggested that this new Decompression Engine will be in the Nintendo Switch 2 as well, which launches later this year.
What are the specs of the RTX 5090?
The Blackwell GB100 series is for the server class GPUs, while the GB200 series die will power the consumer class. The RTX 5090 will utilize the GB202 die. A new leak from @kopite7kimi on X/Twitter suggests that GB202 will be a monolithic die. @kopite7kimi is a renowned PC hardware leaker with 66% accuracy in their leaks, according to a hardware-claims-tracking database.
But when a user asked if the GB202 die is essentially two GB203 dies fused together, @kopite7kimi answered: "Logically." Therefore, it may not be a traditional monolithic design, but rather two GPU dies fused with an interconnect in the middle. So, it's "logically" acting as a monolithic die instead of being a true monolithic die.
Also, @kopite7kimi suggested that there would be three PCBs (Printed Circuit boards) in the Founders Edition of the RTX 5090 GPU. The three PCBs would be the Main Board, an IO Rigid Board, and a PCB board for the PCIe slot. The main board would house the actual GPU die, memory chips, voltage-regulating modules, and more.
The IO Rigid Board should have HDMI, DP, and possibly a USB-C connector. The third PCB should have the PCIe connector that goes into the motherboard. What prompted the sudden change in the PCB design? As far as I can understand, this solves two problems: the size and GPU sagging. Modern graphics cards' PCB sizes have become quite large, which has resulted in such issues over time.
Having separate PCBs should allow Nvidia to save space and design shorter and smaller graphics cards for its future Founders Edition cards. It would solve another issue called GPU sagging. It happens when the PCB is longer, and the weight of the heatsink and cooler makes the PCB sag from one side. It would later snap from the continuous weight, thereby killing itself.
The idea of a shorter and slimmer RTX 5090 FE could become real. When replying to a user named @OC3D, @kopite7kimi mentioned that the flagship GPU is a 2-slot cooler.
Previously, we wrote about the leaked specs of the RTX 5090, which helped us present the following specs for the GPU.
What is the expected performance of the RTX 5090?
According to the leaked specs of the RTX 5090, it has a 50% increase in shader cores, a 77.7% increase in L2 cache size, and a 75% increase in VRAM bandwidth compared to the RTX 4090. These numbers seem to indicate that the theoretical performance of the new RTX GPU should be around 50 - 60% more than the previous flagship Nvidia graphics.
However, specs don't always translate to real-world performance. On top of that, these numbers are based on leaked specs, therefore actual gaming performance may end up more or less than that.
What is the expected price of the RTX 5090?
The Nvidia RTX 4090 was launched for $1,599, but is currently selling for $1,750 to $2,000; this is significantly higher than its launch price. With the crazy high demand in the AI market, it's unlikely that Nvidia will sell the upcoming RTX 5090 any less than the launch price of RTX 4090.
In fact, the new GPU may end up launching somewhere between $1,750 and $2,000. Nonetheless, it's an exciting release. The new technologies Nvidia is bringing with the RTX 5090 may rejuvenate interest in the gaming industry once again.