[Hardware] NVIDIA GEFORCE RTX 3050

[Mr.Bada]

Graphics cards are a real pain point for PC gaming. Hell of an opening sentence to a new GPU review, I know, but that is the world we inhabit in 2022. Supply is constrained and demand is as sky-high as the prices of said GPUs. So, it was with a faint taste of something akin to hope on my lips I approached the launch of new mainstream graphics cards from both AMD and Nvidia. If you've peeped the lowest score I think I've given anything in my time since coming back to PC Gamer then you'll know how the AMD Radeon RX 6500 XT turned out. Now it's Nvidia's turn to either save the day or disappoint us all with the GeForce RTX 3050. I mean, there can't be any middle ground, can there? Somewhat inevitably, that's going to depend very much on your point of view.

Still, I'm feeling a little better about the budget-conscious end of the PC gaming market since enjoying the fruits of Intel's Core i5 12400 labours, and its seemingly accidental overclocking bounty. A $200 chip that can comfortably nail 5.2GHz, with an affordable motherboard platform to boot? Well, that has the potential to be the perfect base for your next gaming rig. All we need is an affordable GPU to go with it. This is where Nvidia steps in with the nominally $250 RTX 3050, promising to blow the RX 6500 XT out of the water for just $50 more. As it turns out MSRPs evaporate as soon as the first wave of GPUs sell out—generally in the first ten minutes—so we have the $280 Gigabyte RTX 3050 Eagle here, which is actually marginally cheaper than the $300 Gigabyte RX 6500 XT we reviewed as our first taste of AMD's budget GPU.And it does indeed blow the RX 6500 XT out of the water. But we are still essentially talking about a new Nvidia graphics card with GTX 1660 Ti gaming performance, though one with the added veneer of ray tracing hardware and DLSS support. What are the RTX 3050 specs?

The comparisons both AMD and Nvidia have made with their new cards generally hides where they actually sit in terms of relative performance. Both have spent an irrelevant amount of time putting their new cards up against the GTX 1650, with AMD also pitting the RX 6500 XT against the RX 570, and Nvidia the RTX 3050 against the GTX 1050. None of these comparisons make sense. The AMD card's poor performance is more ably highlighted by a comparison with the RX 5500 XT and RX 480—similarly specced and priced cards from generations back—and it is more meaningful to compare the RTX 3050 with the $279 GTX 1660 Ti than the $149 GTX 1650 or GTX 1050.The GA106 GPU Nvidia reportedly has at the heart of the RTX 3050 is a smaller version than that used in the RTX 3060 which makes it roughly the same size as the chip used in the old GTX 1660 Ti. It does though have almost double the number of transistors at its heart and, thanks to the excellent Ampere architecture it comes with over one thousand more shaders inside it. 

It is also notable that AMD's latest GPU, the Navi 24 silicon inside the RX 6500 XT is a 107mm2 chip, while the GA106 is something in the region of 276mm2. That's something which will likely come into play when it comes to manufacturing and the numbers of GPUs that each company can produce to fulfil the demand for their cards.

Those 2,560 CUDA cores in the RTX 3050 are split between 20 SMs which also then delivers 20 RT cores and 80 Tensor cores. It's those last two specs that separate the RTX 3050 from its GTX 1660 Ti brethren and ought to make it a far more tempting choice for the mainstream gamer. The RT cores are what allow for ray tracing and the Tensor Cores then allow you to enable DLSS to make ray tracing actually playable at a decent frame rate, all thanks to the magic of AI-powered super sampling.The RTX 3050 has a similar total graphics power (TGP) compared with the GTX 1660 Ti, too, with 130W vs. 120W for the older card. Along with that extra 10W, you also get another 2GB of GDDR6 memory for a total of 8GB. Though that runs on a thinner 128-bit aggregated memory bus compared with the GTX 1660 Ti's 192-bit bus, which is why you end up with less memory bandwidth as a result.