CES 2025 is behind us, and we're eager to see if Nvidia's premiere RTX 5090 can live up to the hype. Come January 23 at 6:00 PM PST, the RTX 5090 Founders Edition review embargo will finally be lifted as we dissect Nvidia's performance claims. Reviews for custom AIB (Add-In Board) models will follow a day later at the same time (though not initially from us).
The RTX 5090 is based on Nvidia's Blackwell architecture, employing the GB202 die, and fabricated using TSMC's 4NP (5nm-class) process technology. With a die size of 750 mm2, this is Nvidia's largest consumer GPU since the Turing TU102 (RTX 20-series). All that glitters is not gold, however, as despite the near reticle-limit-sized chip, the transistor density sees no improvement over the last generation at ~123 million transistors per mm2 (92 billion transistors across a 744mm2 area). That's no surprise, as there's no change in process node — the RTX 50-series will use TSMC 4N, just like the RTX 40-series.
Powering the RTX 5090 are 21,760 CUDA cores (170 SMs), a 33% increase over the RTX 4090. Not only that, but Nvidia has also equipped the RTX 5090 with 32GB of GDDR7 memory across a wide 512-bit interface for 1.79 TB/s of bandwidth, a 78% improvement over the 4090. The TGP (Total Graphics Power) also sees a 27% uptick versus Lovelace, at 575W. Despite a $1,999 MSRP, RTX 5090 prices at launch are rumored to be inflated due to supply constraints.
RTX 5090 FE (Founders Edition) reviews will go live tomorrow at 6:00 PM PST (Pacific Standard Time). As of writing, only 19 hours remain before reviews drop (be sure to adjust for your timezone). We have already unboxed the RTX 5090 FE, where Nvidia boasts a 50% size reduction in packaging, now made of paper fibers (cardboard).
The standout feature is Nvidia's dual-slot, double flow through design for the RTX 5090 FE. It's pretty impressive how the company has managed to cram 575W of power dissipation into such a small form factor, especially when custom models are going up to four slots wide. This required creative thermal engineering as the RTX 5090 FE features a multi-PCB structure, enabling better airflow and cooling. There's also liquid metal for the thermal interface material.
No third-party benchmarks are allowed yet, but leaks suggest a 30-40% uplift in synthetics. We'll see how well that translates to gaming performance, with and without Multi Frame Generation. Regarding pricing, the RTX 5090 FE will set you back $1,999, assuming you can find it at MSRP.
| Graphics Card | RTX 5090 | RTX 5080 | RTX 5070 Ti | RTX 5070 |
|---|---|---|---|---|
| Architecture | GB202 | GB203 | GB203 | GB205 |
| Process Technology | TSMC 4N | TSMC 4N | TSMC 4N | TSMC 4N |
| Transistors (Billion) | 92.2 | 45.6 | 45.6 | 31 |
| Die size (mm^2) | 750 | 378 | 378 | 263 |
| SMs / CUs / Xe-Cores | 170 | 84 | 70 | 48 |
| GPU Shaders (ALUs) | 21760 | 10752 | 8960 | 6144 |
| Tensor / AI Cores | 680 | 336 | 280 | 192 |
| Ray Tracing Cores | 170 | 84 | 70 | 48 |
| Boost Clock (MHz) | 2407 | 2617 | 2452 | 2512 |
| VRAM Speed (Gbps) | 28 | 30 | 28 | 28 |
| VRAM (GB) | 32 | 16 | 16 | 12 |
| VRAM Bus Width | 512 | 256 | 256 | 192 |
| L2 / Infinity Cache | 96 | 64 | 48 | 48 |
| Render Output Units | 176 | 112 | 96 | 80 |
| Texture Mapping Units | 680 | 336 | 280 | 192 |
| TFLOPS FP32 (Boost) | 104.8 | 56.3 | 43.9 | 30.9 |
| TFLOPS FP16 (FP4/FP8 TFLOPS) | 838 (3352) | 450 (1801) | 352 (1406) | 247 (988) |
| Bandwidth (GB/s) | 1792 | 960 | 896 | 672 |
| TBP (watts) | 575 | 350 | 300 | 250 |
| Launch Date | Jan 30, 2025 | Jan 30, 2025 | Feb 20, 2025 (?) | Feb 27 2025 (?) |
| Launch Price | $1,999 | $999 | $749 | $549 |
In addition to aesthetics and design, the thermal performance and boost clocks set FE and custom variants apart. CES showcased numerous custom AIB designs with innovative technologies, such as Zotac's new safety light feature and yellow-tipped power connectors from MSI.
Asus is adding a new sub-brand to its ROG series, Astral, and will debut with four RTX 5090 variants at reference boost clocks. Gigabyte will launch the RTX 5090 across its Aorus Elite, Aorus Master, Aorus Master Ice, Xtreme Waterforce, Gaming and Windforce lineups. At MSI's end of the ring, we find its renowned Gaming and Ventus alongside high-end Vanguard, Suprim, and Special Edition offerings. Thus far, Zotac is the only AIB to offer a higher boost clock than reference, with the RTX 5090 AMP Extreme Infinity reaching 2,467 MHz.
Nvidia specifies that the RTX 5090 operates at a base and boost clocks of 2,017 MHz and 2,407 MHz, respectively. We expect the real-world boost clocks to be higher, as seen with the RTX 4090. You can see the 4090 results in our testing below, which clocked in at around 2,750 MHz on average in games despite a reference boost clock of only 2,520 MHz. Other 40-series GPUs show a similar delta between official boost clocks and real-world clocks.
So buckle up, as we have much to disclose and discuss tomorrow. Will the RTX 5090 match Nvidia's claims of a 2x jump in performance? The RTX 5090 will hit shelves on January 30 alongside its younger sibling, the RTX 5080, though supply-chain constraints may limit availability.
Rumors suggest the RTX 5070 Ti might be available later next month on February 20 but we'll have to wait until further confirmation from Nvidia. If the pattern continues, we can expect the RTX 5070 a week later — and Founders Edition reviews will be able to go live a day before the third party cards.
