Nvidia GeForce RTX 5090 and 5080 specs just leaked, only one of them looks good

The Nvidia GeForce RTX 5090 specs have just leaked, according to the latest word in the rumor mill, along with the RTX 5080 specs too. If these leaks are genuine, the new flagship Nvidia gaming GPU looks like a seriously powerful monster, but the RTX 5080 looks comparatively weak.

Not that this should worry Nvidia. With AMD expected to only be focusing on the mid-range GPU market when it launches its RDNA 4 architecture, Nvidia should end up having the flagship best graphics card market all to itself. Nvidia also hopefully learned a lesson with the launch of the GeForce RTX 4080 back in 2022, which is that you can’t actually charge whatever you want for a graphics card if the value is terrible – people will just choose not to buy it. Nvidia later addressed this mistake with the cheaper and more powerful RTX 4080 Super earlier this year.

Let’s start with the RTX 5090 specs, which have seemingly just been dropped by regular tech leaker kopite7kimi. In a post on X (formerly Twitter), kopite7kimi revealed a string of numbers, which appear to add up to an incredibly powerful new graphics card, using the GB202-300-A1 Blackwell chip. According to the leak, the RTX 5090 will feature 21,760 FP32 (32-bit floating point) units, meaning it has 21,760 CUDA cores, as Nvidia’s core design can process both FP32 and INT32 calculations.

With 128 CUDA cores in each streaming multiprocessor (SM) unit, this spec would give the RTX 5090 170 SMs, with each SM expected to have one RT core for ray tracing. In terms of numbers, this is significantly bigger than the RTX 4090, which has 16,384 CUDA cores spread over 128 SMs, although this isn’t a like-for-like comparison when it comes to performance, as the RTX 5090 will use the Blackwell architecture rather than the Ada Lovelace architecture used by the RTX 4090.

The leak also shows a fat memory system, in terms of both capacity and bandwidth. If this leak is genuine, then the RTX 5090 has a massive 32GB of GDDR7 VRAM attached to a 512-bit interface. As a point of comparison, the RTX 4090 has 24GB of GDDR6X VRAM and a 384-bit interface.

There’s no mention of the frequency of the VRAM in the leak, but if it’s 28Gbps, which is certainly feasible, then the total memory bandwidth would be a huge 1.79TB/s, compared to just over 1TB/s on the RTX 4090. The leak also claims the RTX 5090 will draw 600W of power, which corresponds with previous leaks about the RTX 5090 power draw.

This had previously led some people to speculate that the RTX 5090 has two 12Vx6 power connectors, but tech site Benchlife has seemingly put that speculation to bed in a recent report, claiming that “there are no two sets of 12v on the GeForce RTX 5090, as reported by some media.” Benchlife also claims that the RTX 5090 will have a 14-layer PCB and use the PCIe 5.0 interface.

Nvidia GeForce RTX 5090 specs estimate

This is our best estimate for the Nvidia GeForce RTX 5090 specifications, based on rumor and speculation, as Nvidia hasn’t made an official announcement yet.

Next up is the RTX 5080 which, much like the RTX 4080 when compared to the 4090, looks like it has much less power than the 5090, although it should still be capable. According to kopite7kimi, this new GPU is based on the GB203-400-A1 chip and has 10,752 CUDA cores, less than half the number of the 5090, equating to 84 SMs (and 84 RT cores).

Meanwhile, the memory setup looks very similar to that of the RTX 4080, with 16GB of VRAM attached to a 256-bit interface, but with GDDR7 memory potentially increasing the bandwidth. On the plus side, it reportedly draws less power, with a rating of 400W.

Nvidia GeForce RTX 5080 specs estimate

This is our best estimate for the Nvidia GeForce RTX 5080 specifications, based on rumor and speculation, as Nvidia hasn’t made an official announcement yet.

Of course, none of this has been officially confirmed by Nvidia, so take all this with a pinch of salt right now. However, if these are genuinely the specs of the RTX 5080, then let’s at least hope that Nvidia doesn’t give it a ridiculous price, as it did with the RTX 4080 when it first launched. It currently looks as though there’s an enormous gulf between the 5080 and 5090 when it comes to both GPU power and the memory setup, and it would be bad news for gamers if there isn’t much of a difference in price between the two chips.

It would also be disappointing if Nvidia’s second-in-command did indeed only have a 256-bit memory interface and 16GB of VRAM. As a point of comparison, even AMD’s current-gen Radeon RX 7900 XT has 20GB of VRAM and a 320-bit interface. If this GPU ends up costing $999 or more, then I’d expect it at least to have a 20GB/320-bit setup to help see it through the next few years of game releases at high settings.

For more information about what’s coming with Blackwell for gamers, check out our Nvidia GeForce RTX 5090 guide, where we share everything we currently know about the new chip. You can also read our Nvidia GeForce RTX 5080 guide for more information about the rumored second-in-command in Nvidia’s Blackwell lineup.