ray-tracing and DLSS for less than 300 euros, is it possible?

It’s 2022, and NVIDIA now covers all segments of the desktop graphics card market with its product portfolio ranging from the GeForce RTX 3090 (Ti) to the GeForce RTX 3060. All? No ! The entry level is still populated by irreducible GeForce GTX 10/16 Series which resist again and again, and life is not easy every day for these garrisons of players forced to drastically reduce the graphics settings of their favorite game. to hope to reach the bar of 60 constant FPS in Full HD. This is where the GeForce RTX 3050 comes into its own: this card replaces the aging GTX 1060 and 1660, which are no longer able to run recent AAA games decently at 1080p, with the added bonus of support for ray-tracing and DLSS.

In addition to completing its catalog from below, this new graphics card allows NVIDIA to compete (without great difficulty) with the Radeon RX 6500 XT which has just appeared at AMD. Its technical characteristics have been carefully defined so that it is powerful enough to decent play in Full HD with any game, while maintaining an enticing MSRP in these times of GPU shortages (among others).

In practice, the manufacturer has therefore chosen to use a GA106 GPU, the same as the RTX 3060, but with a large part of the transistors disabled. The RTX 3050 also embeds 8 GB of GDDR6 memory : NVIDIA is not making the same mistake here as AMD with its RX 6500 XT with only 4 GB of VRAM. We can even hope to taste the game in QHD (1440p) with this card, but we will necessarily have to make big concessions on the graphics settings side.

NVIDIA has chosen, as is often the case with its entry-level products, not to offer Founders Edition versions of its RTX 3050, limiting itself to providing its various partners with a reference design. We were supposed to receive a model from Gigabyte to test this GeForce RTX, but a setback delayed its arrival in our lab. We therefore had to turn to our colleague Igor Wallossek from Igor’s Labs to carry out the tests that will follow, based on a model from Palit, a RTX 3050 Dual OC 8GB. Factory overclocked model, this RTX 3050 is displayed at a recommended retail price higher than the 279 euros announced by NVIDIAbut it is useless to remind you that these prices are completely theoretical and that it is in any case currently impossible to find any graphics card displayed at its recommended retail price…

NVIDIA GeForce RTX 3050 Spec Sheet

The manufacturer uses a Ampere GA106 chipset identical to that of the RTX 3060, still engraved in 8 nm by Samsung, but with some parts disabled. The number of Streaming Multiprocessors thus goes from 28 to 20, a reduction from 3584 to 2560 cores. The number of Tensor Cores and RT Cores also decreases, from 112 and 28 units respectively on the RTX 3060 to “only” 80 and 20 units. In summary, it is a Graphics Processing Clusters complete that NVIDIA has deactivated on the GPU of its GeForce RTX 3050. In return, the base frequency increases a little, reaching 1552 MHz.

The RTX 3050 has four VRAM controllers (compared to six on a full GA106) and embeds 8 GB of GDDR6 memory (14 Gbps), which gives a total bandwidth of up to 224 GB/s with a 128-bit bus. Here we are also below an RTX 3060 and its 360 GB/s, but well above the RX 6500XT whose 4 GB of GDDR6 only offers a bandwidth of 144 GB/s; the two graphics cards obviously do not play in the same category…

The TDP of this graphics card is logically lower than that of the RTX 3060: with 130W announced, the single PCI-Express 8-pin power connector will be more than enough to operate the RTX 3050. This of course benefits from all the technologies specific to GPUs in the Ampere range, including Resizeable BAR. Second-generation RT Cores and third-generation Tensor Cores deliver the ray-tracing and DLSS support, and NVIDIA Reflex, NVIDIA Image Scaling and NVIDIA Broadcast features are included. Compatible DirectX 12 Ultimatethe RTX 3050 also supports the Variable Rate Shading. On the other hand, it will be limited in terms of cryptocurrency mining, Lite Hash Rate model obliges, even if some programs now allow this limitation to be partially overridden. Palit’s card has an HDMI 2.1 output and three DisplayPort outputs, which should be the case with almost all other RTX 3050 models. Finally, the GPU is capable of hardware decode an AV1 stream.

In dual-slot format, the Palit card is rather compact with dimensions of 24.4 x 11.5 cm, for a very reasonable weight of 555 grams on the scale. The PCB is really small since it barely protrudes from the end of the PCI-Express 16x port! It should be noted in passing that not all the slots for the surface-mounted components are used: Palit seems to have reused the PCB of its RTX 3060 Dual OC by slightly lightening the power stage. This factory overclocked model benefits from a slightly higher Boost frequency than the reference model (1822 MHz), the rest of the characteristics, on the other hand, respecting NVIDIA’s recommendations.

Performance in Full HD

The performance measurements were carried out in the laboratory on a standardized platform, bringing together high-end components: a AMD Ryzen 9 5900X processor overclocked (PBO + 500 MHz), an MSI MEG X570 Godlike motherboard, two 16 GB modules of DDR4-4000 Corsair Vengeance RGB Pro memory (CL18-22-22-42), a 2TB Gigabyte Aorus NVMe Gen4 SSD, a second 2TB Corsair MP400 SSD, a power supply Be Quiet Dark Power Pro 12 1200W as well as a Raijintek Paean case and a BenQ PD3220U display. Cooling is entrusted to a watercoling system consisting of an Alphacool Eiswolf (modified) and an Eisblock XPX Pro, with Thermal Grizzly Kryonaut thermal paste.

Palit’s RTX 3050 was pitted against a few current and previous models from AMD and NVIDIA. The test panel consists of ten recent games, seven using the DirectX 12 API (Borderlands 3, Horizon Zero Dawn, Marvel’s Avenger, Metro Exodus Enhanced Edition, Necromunda, Shadow of the Tomb Raider, Watch Dogs Legion) and three the Vulkan API (Ghost Recon Breakpoint, Wolfenstein Youngblood, World War Z). The tests were carried out in Full HD, with the graphics settings pushed to the maximum.

The difference in average performance between the RTX 3060 and the RTX 3050 reaches 27%, which is almost proportional to the difference in terms of calculation units between these two graphics cards. Minimum FPS also drops by around 25% between the RTX 3060 and RTX 3050, but the card is powerful enough to stay above the 60 FPS mark (on the average of the ten games in our panel).

In other words, the actual performance is in line with the claims of the graphics card and the announcements of NVIDIA, namely the possibility of play recent games in Full HD at maximum graphics settingsall with a frame rate correct. On the competition side, the RX 6600 is unsurprisingly faster (but also significantly more expensive), and the RX 6500XT is totally outdated. And there is even a small place between the RTX 3050 and RTX 3060 for a hypothetical GeForce RTX 3050 Ti…


The idle consumption of the GeForce RTX 3050 is reasonable, especially since a part is to be credited to the RGB lighting of the Palit model. In game, we are around 130 watts whatever the definition, with however a slightly higher consumption in QHD than in FHD, the card then running at the maximum of its possibilities. We note all the same that the consumption exceeds the TBP (the total consumption of the card) announced by NVIDIA in certain cases such as the GPGPU calculations or the “torture” test. The explanation is to be found in the settings chosen by Palit: the manufacturer has indeed pushed the consumption limit to 140 watts on its model, the vBIOS also allowing manual adjustment of the TBP between 100 watts and 150 watts. In practice, however, the performance gain is negligible…

In detail, in-game Palit’s RTX 3050 draws up to 2.4 amps from the PCI-Express 16x port (well below the 5.5A limit) to power one of the power stage phases, the other three phases being powered by the 8-pin PCIe connector. The set therefore appears very well balanced and well below the limits imposed by the PCI-Express standard. The Boost frequency reaches 1980 MHz in play, but quickly stabilizes around 1850 MHz. Note in passing that just like the RTX 3060, the GeForce RTX 3050 does not really seem made for overclocking: it will be almost impossible to increase the GPU frequency by more than 100 MHz, which translates into an increase of 60 MHz after the heating period. Of course, these values ​​are generic and will depend heavily on chance, depending on the quality of the GPU installed.

Cooling system

NVIDIA does not offer a Founders Edition model of its GeForce RTX 3050, the cooling system and therefore the noise pollution vary depending on the model tested. In the case of the RTX 3050 Dual OC from Palit, we find a cooler consisting of two 90 mm fans with 9 blades each, six nickel-plated copper heat pipes and several radiators. The GPU and GDDR6 memory chips are topped with a heatsink to which the heat pipes are attached, while the power stage components directly contact three of the heat pipes.

At the back, the backplate ABS plastic does not participate in cooling and only serves to reinforce the assembly of the cooling system with the PCB, in addition to contributing to the general aesthetics of the board.

the cooling system is quite efficient, but its job is made easier by the board’s reasonable TBP. Noise pollution reaches 34.1 dB(A) in game, which is relatively quiet. In-game temperature readings show a maximum of 49.6°C at the back of the PCB at the GPU level, but the VRMs still climb up to 81.7°C. Even if it seems high in absolute terms, we remain significantly below what these components can support: here too, the power stage seems well sized with regard to GPU and GDDR6 memory consumption.

Leave a Comment