Introduction

A few weeks ago, NVIDIA released the new GeForce RTX 3050 6 GB as its new entry level discrete GPU—here's our review. This new GPU slipped into NVIDIA's product stack with much less fanfare than the RTX 40 SUPER series, and replaces the likes of the GTX 1650 and GTX 1660 as sub-$200 options. In case you haven't noticed, the latest crop of discrete graphics cards are expensive—three-figure prices are normal, gone are the days of the $70 entry-level, because processor integrated graphics solutions have grown powerful over the years. Besides an entry level price, the RTX 3050 6 GB is particularly energy efficient, and is designed to operate without additional power connectors, needing just the PCIe slot power to run.

The GeForce RTX 3050 6 GB that we review today is based on the older Ampere graphics architecture, which meets the full DirectX 12 Ultimate feature set, including real time ray tracing; as well as support for DLSS. Is the card fast enough for ray tracing? We'll answer this question later in the review. If you recall, NVIDIA's argument against ray tracing for entry-level products was that the hardware would be too slow for gameplay at acceptable frame-rates, which is why when it debuted RTX with Turing, it had carved out an entire lower end product stack—the GTX 16-series, which lacked ray tracing capabilities. So, what's changed now? We believe it's the growth in popularity of DLSS, and Intel's debut into the GPU market.



Intel in 2022 released its Arc Alchemist discrete graphics, and even its entry-level Arc A380 features the full DirectX 12 Ultimate feature set along with XeSS, a modern upscaling technology. The A380 is a really slow GPU, but later in 2023, the company released the Arc A580 for less than $200, which is based on a larger silicon, and capable of 1080p gaming, or gaming with ray tracing when you engage XeSS. Combine this with Intel's channel marketing muscle, and its ability to get DIY retailers to push combos of the A580 with entry-level Core i3 processors and H610 chipset motherboards, and the company was in a position to grab a slice of the entry-level e-sports gaming pie. We believe that NVIDIA is responding to exactly this trend—to give the entry-level a convincing new GeForce RTX option that features DirectX 12 Ultimate, including ray tracing; and NVIDIA's popular DLSS super resolution technology. NVIDIA's play with the RTX 3050 6 GB is to address the crowd that games on a 1080p display, but with DLSS enabled, which means the game is actually rendering at a lower resolution, such as 720p, but being upscaled and made to look close to 1080p native.

The GeForce RTX 3050 6 GB is a significantly different product from the original RTX 3050 that NVIDIA launched two years ago—January 2022 to be exact. The original RTX 3050, with 8 GB of memory, had released at a time when the gaming GPU market was reeling with acute shortages in supplies due to the double whammy of the pandemic and the GPU crypto-currency mining craze. It was the only GPU that was available at a relatively acceptable price, and was very much capable of 1080p native resolution gameplay. The RTX 3050 8 GB was initially based on a cut-down GA106 silicon—the same chip which NVIDIA used to build its popular RTX 3060—but later switched to the smaller GA107 silicon without any change in specs, as the RTX 3050 8 GB specs sheet exactly aligned with that of the GA107. In fact, these newer revisions of GA107-based RTX 3050 8 GB were more energy efficient, with a 10-20 W reduction in gaming power draw, compared to the GA106-based ones.

For the new RTX 3050 6 GB, NVIDIA stuck with the same 8 nm GA107 silicon, but lowered various specs—not only memory size. To begin with, the most obvious change with the RTX 3050 6 GB is the VRAM size, which sees a 25% reduction to 6 GB from 8 GB, and a proportionate reduction in memory bus width, to 96-bit from 128-bit. The memory bus ticks at 14 Gbps, which yields 168 GB/s in memory bandwidth, compared to the 224 GB/s on the original RTX 3050. That's not all, the core count is slightly reduced, too. While the original RTX 3050 maxes out all 20 streaming multiprocessors (SM) present on the GA107, the new RTX 3050 6 GB enables 18 out of those 20. This works out to 2,304 CUDA cores, 72 Tensor cores, 18 RT cores, 72 TMUs, and the chip's full 32 ROPs.

Perhaps the biggest impact of this reduction in SM has been on power—it comes with a TGP of just 70 W. This means that the card is optimized to work without any power connectors (the original RTX 3050 came with a 6-pin connector to meet the 115 W TGP). The spartan 70 W power reflects in thermals, and partners are allowed to use simple extruded aluminium fan-heatsinks that cost just a few dollars to make. It also vastly simplifies the power design of the GPU.

NVIDIA has no reference design for the RTX 3050 6 GB, and so we have with us the Gainward RTX 3050 6 GB Pegasus. We chose Gainward because its PCBs tend to be designed close to NVIDIA's internal reference design boards and their cards are sold at highly competitive pricing. This card offers all there is to have with this GPU, and even makes things easy for the entry-level crowd, with legacy display connectivity, such as a DVI connector. It features a simple aluminium monoblock fan-heatsink. Gainward is pricing the card at $180, which is the baseline price set by NVIDIA for the RTX 3050 6 GB.