The second graphics card of NVIDIA's Blackwell generation to be released, the GeForce RTX 5080 targets the 4K Ultra gaming market HD by offering a more affordable alternative to the recently launched RTX 5090. Based on a brand new GB203 graphics chipset accompanied by 16 GB of GDDR7 memory, it presents itself as an interesting replacement for the RTX 4080 and 4080 Super, allowing you to take advantage of the latest graphics technologies.
Its price positioning aims to appeal to a gaming audience with a substantial budget, but still lower than that required to acquire an RTX 5090. The reference Founders Edition model is thus displayed at a recommended retail price of 1179 euros, and it is also at this price that PNY officially offers the model that interests us here. Be careful though, it is quite possible that the real prices will be a little higher for a while, or even that the availability of the different cards will be... quite random.
Attractive on paper in terms of performance and features, and accompanied by software innovations that a priori seem rather interesting, should you really fall for this GeForce RTX 5080?
GeForce RTX 5080: the Blackwell architecture
At the heart of this GeForce RTX 5080, we find a GB203 chipset adopting NVIDIA's new Blackwell architecture. Physically smaller than the GB202 of the GeForce RTX 5090, it still includes all the new features and improvements, both hardware and software. Engraved by TSMC using its 4N process, the same one used to manufacture the graphics chipsets of the previous generation, this GB203 has a physical size of 378mm² for 45.6 billion transistors, values similar to the AD103 which runs the RTX 4080.
The RTX 5080 GPU packs 84 streaming multiprocessors (SM), organized into 7 GPCs (Graphics Processing Clusters), all of which are functional in this case. This translates to a total of 10,752 CUDA cores, compared to 10,240 for the RTX 4080 Super. There are also 336 5th-gen Tensor cores, 84 4th-gen RT cores, 336 texture units (TMU), and 112 ROPs (Render Output Unit). The size of the shared L2 cache remains unchanged from the previous generation, at 64 MB. The RTX 5080, however, benefits from a significant upgrade in memory bandwidth. Although the bus width remains at 256 bits, it uses faster GDDR7 memory (30 Gbps), thus achieving a total bandwidth of 960 GB/s.
In addition to the units dedicated to 3D rendering and AI, the RTX 5080 includes two ninth-generation NVENC encoding engines generation and two sixth-generation NVDEC decoding engines. Independent, these units support 4:2:2, AV1 UHQ and MV-HEVC video formats (for 3D video streams and VR). Equipped with a PCI-Express 5.0 x16 interface, the card offers DisplayPort 2.1b video outputs with UHBR20 and HDMI 2.1 to achieve very high definitions and refresh rates: 480 Hz in 4K, and up to 120 Hz in 8K (with DSC compression). The whole thing finally benefits from a power limit of 360 watts, higher than the 320 watts achievable by the RTX 4080 Super.
DLSS 4 and multiple frame generation: new software features
NVIDIA is introducing several new software features with its Blackwell generation, including DLSS 4 and Multi Frame Generation. Generally speaking, these technologies rely on artificial intelligence to improve image quality and gaming performance.
More specifically, DLSS 4 uses an AI model based on transformers by default instead of the convolutional neural networks (CNN) used until now; enough to obtain better image quality at all levels of performance. NVIDIA adds in passing that this new AI model is compatible with the previous RTX 40, 30 and 20 series, which suggests possible quality gains for older generations of graphics cards.
The Multi Frame Generation is however – for the moment – an exclusive to the RTX 50 series. The AI here allows the generation of up to three consecutive frames (compared to just one with the classic Frame Generation) after a traditionally rendered image, which allows to considerably increase the number of frames per second, without significant impact on latency thanks to the joint use of NVIDIA Reflex (and soon Reflex 2), another technology from the manufacturer.
The Blackwell architecture is also designed to support neural rendering, where generative AI plays a direct role in the rendering process, and not just in specific features such as DLSS Super Resolution: by standardizing this approach in the DirectX 12 API, NVIDIA then allows 3D applications to directly access the Tensor cores. Finally, the new hardware scheduler (or AI Management Processor) allows generative AI and 3D graphics rendering workloads to be executed in parallel, optimizing the use of available hardware resources.
PNY's RTX 5080 OC in detail
The factory overclocked version of PNY deviates quite significantly from the design of the Founders Edition model, particularly on the cooling system side. The manufacturer has indeed chosen to take up a design proven on previous generations, with a large heatsink equipped with three fans, the middle one operating in reverse. We will also appreciate its semi-fanless operation: the fans will only start up from 50% load.
As the “OC” version requires, the graphics chipset of this card benefits from a few additional Megahertz in Boost: we thus officially go from 2617 MHz to 2780 MHz. The manufacturer has, however, taken the 360W consumption limit of the reference model to the letter, limiting in practice the interest of this factory overclocking.
With its dimensions of 32.9 x 13.8 x 7 cm, the PNY RTX 5080 OC is a little more imposing than the Founders Edition model and will take up three full slots. So be sure to check that you have the necessary space in your case.
Of course, we find the 12VHPWR 12+4-pin power connector already present on previous generation RTXs, and the manufacturer supplies with its card a triple adapter allowing you to continue using your “old” power supply if it does not have a suitable socket.
PNY GeForce RTX 5080 OC: what performance in game?
To test this GeForce RTX 5080 OC from PNY, we used a configuration adapted to the gaming market, powerful enough to let the graphics card express its full potential. It includes an AMD Ryzen 9 7950X processor, an Asus ROG Strix X670E-F Gaming WiFi motherboard, 32 GB of DDR5-5200 memory from Kingston in CL16, a BeQuiet! Pure Power 11 FM 1000W power supply, a BeQuiet! cooling system Pure Loop 2 FX 240 and a 4TB Samsung 990 Pro SSD, all installed in a BeQuiet! Pure Base 500 case. The measurements were carried out with the case closed in order to obtain values similar to those that any gamer would obtain at home.
The range of games used to measure the performance of this RTX 5080 OC includes more or less recent and demanding titles, based on various 3D engines: Cyberpunk 2077 (REDengine 4), Black Myth: Wukong (UE5), Hogwarts Legacy (UE4), Forza Motorsport (ForzaTech), Marvel Rivals (UE5), Shadow of the Tomb Raider (Foundation) and The Riftbreaker (Schmetterling). The measurements were made under Windows 11 24H2 in Full HD, QHD and 4K UHD with different graphics settings, but given the power of the GPU of the day, only the measurements in 4K UHD (with and without raytracing, as the case may be) are really relevant.
Performances en jeu
In classic 3D rendering, the performance of the GeForce RTX 5080 is generally above the RTX 4080 Super, but quite logically below the RTX 5090. The performance gain compared to the previous generation, of the order of 5% to 20% depending on the game, is still less marked than that of the RTX 5090 compared to the RTX 4090. On the other hand, the difference is much more noticeable between the RTX 5080 and an older card like the RTX 3080: on average, we can expect a performance gain of around 70% (without using frame generation).
Compared to a Radeon RX 7900 XTX, which is the best that AMD currently offers on the market (while waiting for the upcoming arrival of the Radeon 9070 and 9070 XT), the observation is clear: the overclocked GeForce RTX 5080 from PNY always comes out on top in 4K, sometimes in a significant way. In any case, this new RTX 5080 demonstrates its ability to handle complex scenes with a stable number of frames per second.
Our opinion on DLSS and Multi Frame Generation: very useful in raytracing
Activating ray-tracing in games obviously impacts the average number of frames per second, but the RTX 5080 still generally manages to maintain a correct framerate, even in high definitions; However, it is in this context that technologies such as DLSS 4 and MFG take on their full meaning, allowing drastic improvements in performance while maintaining good image quality.
Cyberpunk 2077
Hogwarts Legacy
Marvel Rivals
Black Myth: Wukong
The RTX 5080 takes particular advantage of the Blackwell architecture and its 84 RT cores to manage complex raytracing calculations such as advanced lighting and reflection effects in games. In this area, the RTX 5080 does better than all competing Radeons currently on the market, including the Radeon RX 7900 XTX, and in practice places itself once again between the RTX 4080 Super and the RTX 5090. However, the latter offers considerably better results, in return for a much higher drain on the wallet...
Finally, the Frame Generation and even more so the brand new Multi Frame Generation offer an increase in framerate that is simply impressive, and this is perhaps one of the main assets of this GeForce RTX 5080.
Reasonable noise at full power
What could be better, finally, than a little shot of Furmark to stress this GeForce RTX 5080 OC a little and see how it behaves under full load? After about thirty minutes of operation, the GPU temperature is stabilized at 70°C – a reasonable temperature, knowing that the measurements are carried out with the case closed – with fans rotating at a little over 1600 revolutions per minute.
The frequency of the graphics chipset is then around 2300 MHz, which corresponds to the base frequency announced by PNY. Remember that this is an ultimate load test designed to push the graphics card to its limits: it is clear that it does very well since in addition to being perfectly stable despite the 360 watts consumed, it is also very quiet. The winning trifecta, so to speak.
0 Comments