RTX 5090 might be way more powerful than 4090, per recent specs leak

If you thought the RTX 4090 was a powerful graphics card, think again. A recent specs leak suggests the RTX 5090 will bring vastly improved capabilities to the flagship gaming GPU. Besides a newer architecture, the card is reportedly getting a significant specs boost, including more CUDA cores, Streaming Multiprocessors (SMs), faster operating clock speeds, and cache.

Reliable industry insider and leaker Kopite7Kimi previously took to X to share information about the 5090. Per his expectations, the flagship card will use the GB202 graphics processor, following the footsteps of the 4090's AD102. This new consumer-focused version of the GB102 is expected to retain the AI capabilities of the data center Blackwell GPUs.

We have also had multiple other leaks, most notably from Panzerlied on Chip Hell, claiming that the upcoming graphics card will be way more stacked than the 4090.

---

RTX 5090 specs leaks make it seem like a monster

The RTX 4090 is the most powerful graphics card today (Image via Sportskeeda)

Leaked info suggests that the RTX 5090 will pack an eye-watering 24,576 CUDA cores, up from 16,384 on the 4090. Given the card will also use a new architecture, this is a stunning uplift and should directly correlate to improvements in rasterization performance.

Besides, the card is also rumored to pack more Tensor cores (768 vs. 512), making it better for machine learning workloads and DLSS. The RT core and Streaming Multiprocessor count is also expected to go up, which should translate to gains at higher resolutions like 4K.

There also have been rumors claiming that the upcoming RTX 5090 would pack a 512-bit memory bus, up from the 384-bit bus the current gen uses. This is expected to translate to a 50% memory bandwidth gain. Some leaks also suggest a 32 GB VRAM size, but we don't think that's going to be the case.

Here is a recap of the specs list, according to current leaks and rumors:

	RTX 4090	RTX 5090
CUDA cores	16,384	24,576
Ray tracing cores	128	192
Tensor cores	512	768
Streaming Multiprocessors	128	192
Boost clock	2.52 GHz	2.9 GHz
L2 cache	72 MB	128 MB
Memory bandwidth	1.01 TB/s	1.53 TB/s

The community has had mixed reactions to these leaks. Some believe it's impossible to scale the GPU die to make a near 50% jump in core count. Others think Blackwell won't be much more powerful than Ada Lovelace, a trend we saw with the jump from Ampere to the RTX 40 series.

However, with the 4090, Nvidia sought to build a GPU that's like no other. It costs a fortune but delivers class-leading performance that no other graphics card can match. It would make perfect sense for the RTX 5090 to continue this trend.