Tuesday, January 4th 2022
NVIDIA Launches GeForce RTX 3050 and RTX 3090 Ti
NVIDIA today launched the GeForce RTX 3050 desktop graphics card, positioned a notch below the RTX 3060. It is based on the same 8 nm "GA106" silicon as the RTX 3060, but with a third of its streaming multiprocessors (SMs) disabled. While the "GA106" physically has 3,840 CUDA cores, only 2,560 of them are enabled on the RTX 3050. The RTX 3060 doesn't max it out, either, featuring 3,584. Another key differentiator between the RTX 3050 and the RTX 3060 is memory. The new kid on the block features 8 GB of GDDR6 memory across a 128-bit wide memory bus, while the RTX 3060 has 12 GB of it across 192-bit.
The reference-spec sees the RTX 3050 GPU clock boost up to 1780 MHz. With a typical board power if 130 W, the card can make do with even a 6-pin PCIe connector, however, NVIDIA made 8-pin the standard. The typical use-case for the RTX 3050 would be 1080p gaming with RTX ray tracing enabled, but with DLSS also added to the mix. Without ray tracing, it should be able to handle AAA titles at 1080p with fairly high settings. Available from January 27, 2022, the RTX 3050 starts at $249.
The GeForce RTX 3090 Ti is the company's new "BFGPU," a halo product. Much like the RTX 3090, though, it will be available both as an NVIDIA-designed "Founders Edition" product, as well as custom-design card. In its Founders Edition avatar, the card debuts a new 16-pin power connector, that's capable of an enormous amount of power delivery. This is needed because the RTX 3090 Ti has a typical board power of 450 W, which is 100 W more than its predecessor.
The RTX 3090 Ti maxes out NVIDIA's largest "Ampere" GeForce silicon, codenamed "GA102." This means all 10,752 CUDA cores, 84 RT cores, 336 Tensor cores, are enabled. On closer look of the specs, you'll notice that the CUDA core count isn't that much higher than the 10.496 of the RTX 3090 to warrant the massive increase in power. It's because NVIDIA also upgraded the memory sub-system. While the memory amount is unchanged at 24 GB, the company is using faster 21 Gbps-rated GDDR6X memory chips, while the RTX 3090 uses 19.5 Gbps-rated ones. The third area of development, which we believe is the largest contributor to the power draw, is the GPU Boost frequency, which is now 1860 MHz, compared to 1695 MHz on the RTX 3090. NVIDIA is expected to launch the RTX 3090 Ti within January 2022, the company is yet to finalize pricing.
The reference-spec sees the RTX 3050 GPU clock boost up to 1780 MHz. With a typical board power if 130 W, the card can make do with even a 6-pin PCIe connector, however, NVIDIA made 8-pin the standard. The typical use-case for the RTX 3050 would be 1080p gaming with RTX ray tracing enabled, but with DLSS also added to the mix. Without ray tracing, it should be able to handle AAA titles at 1080p with fairly high settings. Available from January 27, 2022, the RTX 3050 starts at $249.
The RTX 3090 Ti maxes out NVIDIA's largest "Ampere" GeForce silicon, codenamed "GA102." This means all 10,752 CUDA cores, 84 RT cores, 336 Tensor cores, are enabled. On closer look of the specs, you'll notice that the CUDA core count isn't that much higher than the 10.496 of the RTX 3090 to warrant the massive increase in power. It's because NVIDIA also upgraded the memory sub-system. While the memory amount is unchanged at 24 GB, the company is using faster 21 Gbps-rated GDDR6X memory chips, while the RTX 3090 uses 19.5 Gbps-rated ones. The third area of development, which we believe is the largest contributor to the power draw, is the GPU Boost frequency, which is now 1860 MHz, compared to 1695 MHz on the RTX 3090. NVIDIA is expected to launch the RTX 3090 Ti within January 2022, the company is yet to finalize pricing.
33 Comments on NVIDIA Launches GeForce RTX 3050 and RTX 3090 Ti
This is just free market at work: limited supply and a category of users (miners) willing to pay 5x the MSRP or more. It's called inelastic supply.
I for one, don't like the trend of PCs getting bigger and hungrier. It screams lazy chip design. The advancement of CPUs and GPUs is not measured by pure performance, but performance per Watt, in my opinion. Looking at it like that, there has been absolutely no advancement from nvidia since Pascal, apart from RT and DLSS.