RTX 5090 Vs RTX 4090: Our Thoughts On The Specs

Expert Verified By

Nvidia claims the new RTX 5090 packs a punch, but do the specs support? Keep reading to find out!

Story Highlights

  • The RTX 5090 has about 33% more GPU cores than the RTX 4090, which leads to a similar increase in the Tensor and RT cores as well.
  • The VRAM of the newer card is even higher than the RTX 4090, at about 32 gigabytes instead of the 24 gigabytes of the 4090.
  • There are many more bells and whistles added to the RTX 5090, some of which will come to the other RTX cards, while the multi-frame generation is exclusive to the RTX 50 series.

Ignoring all the issues with the pricing, Nvidia knocked it out of the park with the RTX 4090 in terms of engineering an excellent product. Now, we have the official announcement of its successor: the Nvidia GeForce RTX 5090. Come along as we explore the difference between these two cards!

Comparison Table

Technical SpecsNVIDIA RTX 4090NVIDIA RTX 5090
GPU DieGB202AD102
Process NodeTSMC 4nmTSMC 4nm
ArchitectureAda LovelaceBlackwell
SMs170128
GPU Shaders2176016384
Tensor Cores680512
RT Cores170128
Boost Clock (MHz)24072520
VRAM Speed (Gbps)2821
VRAM (GB)3224
VRAM Bus Width512384
Texture Mapping Units680512
TFLOPS FP32 (Boost)104.882.6
TFLOPS FP16 (INT8 TOPS)1676? (3352)661 (1321)
Bandwidth (GB/s)17921008
TBP (watts)575450
Launch DateJan 2025Oct 2022
MSRP$1599$1999

Technical Differences

GeForce RTX 5090 Reference Design
GeForce RTX 5090 Reference Design (Source: Nvidia)

This section will deal with the most important differences in the specifications of these two cards. However, the numbers might not reflect the real-world performance, a snippet of which we have already seen; more on that later.

GPU Cores

Nvidias fundamental processing units for parallel computations, which they have self-dubbed their CUDA Cores, saw a 33% improvement on the new card. The RTX 4090 has an impressive 128 CUDA Cores, but they have been raised to 170 on the Backwell GPU.

This links to an increase in the rest of the fundamental units of their GPUs, too, which translates into a similar increase in the AI-focused Tensor Cores and the real-time Ray Tracing-focused RT cores.

Clock Speeds

The RTX 5090 actually has had a downgrade with its boost clock speeds, with the RTX 4090 boosting about 4.7% higher than the newer card.

However, this might not be an accurate representation of the performance as doing some back-of-the-napkin math, we see that the RTX 5090 still maths out to about 27% better performance than the 4090, taking into account its higher core count.

This 27% estimation will change with the other specs taken into account, and also, the architectural differences will take a bigger role in showing off the performance of this new card. 

VRAM

The RTX 5090 leaps to GDDR7 memory, up from the GDDR6 VRAM that the RTX 4090 utilizes. It also helps that the new card jumps to 32 gigabytes of video memory, compared to the still very impressive 24 gigabytes of VRAM that the RTX 4090 comes with.

The bus width also goes from 384 bits to 512 bits with the Blackwell card, which will also unlock the full potential of the GPU die and restrict any memory limitations for many years to come.

Bells And Whistles

DLSS 4.0
DLSS 4.0 Features (Source: Nvidia)

Nvidia also debuted its new DLSS 4.0 technology with the Blackwell lineup of cards. Some features of this technology will come to all the RTX cards, but the multi-frame generation will sadly be exclusive to the new Blackwell cards. The features for all RTX cards include better Ray Reconstruction, enhanced Super Resolution, and Deep Learning Anti Aliasing.

The new tech allows for multi-frame generation, which means their new technology can insert more generated frames between the physically rendered ones. Nvidia claims that this makes 4K 240FPS a reality, even with all the other intensive features like Ray-Tracking, heck, and maybe even path tracing turned on.

Additionally, Nvidia’s Reflex, which is their latency-reducing tech, also receives an uplift with the debut of Reflex 2. This improvement is said to reduce latency from about 27ms on the old tech down to only 14ms with the new cards. Reflex 2, unlike DLSS 4, is said to come out to the rest of the RTX cards in Nvidia’s lineup.

A Note On The Real-World Performance

RTX 5090 Vs RTX 4090 official benchmarks
Looking at NVIDIA’s official benchmarks from the CES announcement 2025. (Source: NVIDIA)

Linus Tech Tips’ exclusive first look at the new card showed that it performed almost twice as well as the RTX 4090 in Cyberpunk. Though, as Linus mentioned, this showcase was done with curated settings, the results were still very impressive.

Was our article helpful? 👨‍💻

Thank you! Please share your positive feedback. 🔋

How could we improve this post? Please Help us. 😔

Gear Up For Latest News

Get exclusive gaming & tech news before it drops. Sign up today!

Join Our Community

Still having issues? Join the Tech4Gamers Forum for expert help and community support!

Latest News

Join Our Community

104,000FansLike
32,122FollowersFollow

Trending

RTX 5090 Vs RTX 4080 Super: What We Think

Let's see how the new top-of-the-line Nvidia card does against something half its price in the RTX 5090 Vs RTX 4080 Super comparison!

RTX 5070 Can’t Beat RTX 4090 Without Using Multi Frame Gen, Admits Nvidia Dev

Nvidia CEO claimed that the RTX 5070 will have similar performance to the RTX 4090, but it won't be possible without the multi-frame gen.

G.SKILL Previews ASUS ROG Crosshair X870E Crosshair Hitting DDR5-10600 With 48GB Memory

G.Skill has achieved various new records for DDR5 overclocking using an ASUS ROG X870E motherboard, including one for DDR5-10600 speed.

TeamGroup Introduces T-Force GE Pro PCIe 5.0 SSD In Huge 8TB Capacity

TeamGroup's blazing fast T-Force GE Pro PCIe 5.0 SSD is getting an 8TB facelift that keeps the same speed intact.

Nvidia CEO Teases Desktop CPU Plans, Corroborating Past Rumors

Nvidia CEO Jensen Huang has recently hinted to investors at his plans to enter the PC CPU market in the future with the rumored 20-core CPU.