If you're thinking about building a powerful PC or upgrading your existing one, you've probably felt a bit lost with the various NVIDIA 30 series cards. Things get complicated because, in addition to the officially released models, some unusual cards have appeared on AliExpress , such as the 20GB RTX 3080, which has more than a few enthusiasts wondering if it's really worth the expense.
The reality is that these days, raw processing power alone isn't enough; the amount of VRAM has become the central issue. If the design program you use or the latest game you've installed requires more memory than your graphics card has, you'll experience annoying lag or, in the worst-case scenario, the application simply won't launch at ultra settings.
First, let's clarify something: the 20GB RTX 3080 is real, even though it wasn't distributed through NVIDIA's usual channels. Technically, it's not a new card; it uses the same chip as the 10GB version, but with double the graphics memory. If you play light games, you won't even notice the difference, but as soon as you start playing titles like Cyberpunk 2077 or The Last of Us Part I, the difference becomes apparent.
For example, in Cyberpunk 2077 at 1440p and high settings, the 20GB version achieves around 101 FPS compared to 93 FPS for the standard version. But the real drama unfolds at 4K resolution with ray tracing : while the 10GB version struggles with minimums of just 4 FPS, the 20GB version performs much better with a minimum of 16 FPS. In this scenario, having those extra gigabytes is what separates a playable experience from a mere slideshow.
An even more extreme case is DOOM The Dark Ages on ultra-nightmare settings. This game can devour more than 13 GB of VRAM at 4K. Here, the 10 GB RTX 3080 throws in the towel and drops to a miserable 5 FPS, while the 20 GB version maintains a respectable average of 37 FPS without needing to use upscaling.
Technical analysis: Ampere's leap versus Turing
To understand the impact of this card, it's helpful to look back and compare it to the RTX 2080 Ti. The Ampere architecture was a massive leap forward, doubling the number of CUDA cores from 4352 to 8704. This isn't just a number to boast about; it translates into a brutal FP32 computing capacity , approaching 30 TFLOPS.
- RT Cores: They have been optimized to make ray tracing up to 70% more efficient.
- Tensor Cores: Although there are fewer cores than in the 2080 Ti, they are third generation and much more powerful in AI tasks.
- GDDR6X Memory: They are basically memories on steroids that allow you to move extremely heavy textures with incredible bandwidth.
In synthetic benchmarks like Time Spy, the 3080 outperforms the 2080 Ti by a margin of between 25% and 33%. And if we look at real-world 4K gaming, the improvement can reach up to 44% on average, clearly demonstrating that the generational leap was worthwhile, even though power consumption increased by about 50W.
RTX 3080 or RTX 3080 Ti?
If you have deep pockets, you could look at the Ti version, which ups the ante with 12GB of VRAM and a 384-bit bus. In terms of pure performance, the Ti is roughly 7% more powerful, but the price is usually much higher. For most, the standard RTX 3080 is the most balanced option, proving to be up to 45% more cost-effective than the Ti version.
In terms of design, NVIDIA introduced an innovative cooling system in the Founders Edition, featuring fans at both ends. One draws cool air into the chip, while the other exhausts heat upwards. They also implemented the 12-pin connector , which, although controversial due to the need for adapters, allows for significantly more power to be handled in a smaller space.
It's not all about pressing play. For those who do productive creative work, VRAM is sacred. If you render 4K videos or work with generative AI models locally, 10 GB can quickly become insufficient. In these cases, many people look to the Radeon RX 6800, which offers 16 GB of VRAM as standard and provides greater peace of mind in the long run to avoid bottlenecks.
To power these beasts, don't skimp on the power supply. A minimum of 750W is recommended, although ideally you'd go for an 850W one to have some headroom, since the power consumption peaks of the 30 series can be quite treacherous and give you a nasty surprise if the power supply is barely adequate.
This entire range is complemented by the NVIDIA Encoder (NVENC), which allows for streaming with stunning image quality and no lag, optimizing the experience for those who want to transmit their content with maximum professional fluidity.
The choice between these cards depends entirely on whether you prioritize raw power or memory capacity. While the Ampere architecture blew away the previous generation thanks to its CUDA cores and GDDR6X memory, users should keep an eye on the VRAM if they plan to work in creative environments or play games in 4K ultra. The 20GB version is the solution for extreme resolutions, while the 10GB version is the logical choice for tighter budgets.