Every GPU launch cycle we go and get excited about the biggest and best cards in the market and the flagship cards like the RTX 3090 are exciting. But they are often beyond what people are looking to spend and beyond what a lot of people need if not gaming at high resolutions. So as Nvidia fills out their product stack, the cards that people are waiting for are farther down. Like the RTX 3060 Ti that is launching today. While the latest-gen cards, in general, are extremely hard to find still, Nvidia is finally starting to let loose the cards that fill the middle of the range that. Even right now four and a half years later, the GTX 1060 dominates the Steam hardware usage charts at 10.9% and the top 15-20 are almost all in that same xx50-xx70 range. I’m excited to see what Ampere does and to see what people looking to upgrade in that range can look forward to.

Product Name: Nvidia RTX 3060 Ti Founders Edition

Review Sample Provided by: Nvidia

Written by: Wes Compton

Amazon Affiliate Link: HERE

 

What’s the RTX 3060 Ti all about

The RTX 3060 Ti is based on the same GA104 GPU that the RTX 3070 used, which is based on the Ampere architecture which all of the 3000 series cards are based on and is the successor to Turing. It is built using an 8nm process from Samsung which is an improvement from the 12nm that Turing used. That alone is enough to make big improvements in performance, power usage, and cooling. But Ampere goes well beyond that. Ampere includes new third-generation Tensor Cores as well as new second-generation Ray Tracing Cores. They also changed how the normal programable shader which handles the 32-bit floating-point and 32-bit integer calculations are done. Ampere added a new data path that allows the integer32 portion to now also handle floating-point calculations if needed which can double the number of floating-point calculations possible. Graphics, as well as compute operations both, rely on the 32-bit floating-point as do modern shaders. Ray Tracing denoising shaders also benefit from the speedups as well.

The new second-generation ray-tracing cores have the ray-triangle intersection throughput doubled which improved the ray-tracing cores RT-TFLOPS from 21.7 on Turing to 31.6. Then on the Tensor core they now automatically identify and remove less important DNN (deep neural network) weights, processing the network at twice the rate of Turing.

So where does the RTX 3060 Ti land with current and past cards? Well, Nvidia is saying that it is faster than the RTX 2080 SUPER. So to compare specifications I included the 2080S along with the RTX 2060 SUPER which is the last generation equivalent. I also included the RTX 3070 which is the next step up on this generation and shares the same GPU, only the 3060 Ti is slightly cut down. The 3060 Ti has 38 SMs compared to the 46 on the 3070. This gets it 4864 CUDA cores which is a crazy number when you compare it with both the 2060S and 2080S or the GTX 1060 that I imagine at 4 and a half years old will have a lot of owners looking to upgrade to the RTX 3060 Ti. That GTX 1060 had 1280 CUDA cores.

The RTX 3060 Ti does change things up when compared to its GA104 based brother the RTX 3070. They have one full GPC turned off as well as one TPC which has a pair of SMs to reach the 38 SMs. The drop in total SMs matches up with the CUDA cores with both being a drop of 17%. But for the ray-tracing cores and the Tensor cores the RTX 3060 Ti has a larger drop of 44%, down to 152 tensor cores and 38 of the new gen 2 RT cores. As for clock speed, the RTX 3060 Ti matches closer with the old RTX 2060 SUPER than with the RTX 2080 SUPER and the RTX 3070 which were higher, it has a boost clock of 1665 MHz. It has the same 8GB of GDDR6 for VRAM, not the GDDR6X that the 3070 has and it runs at 7000 MHz as well but through a 256-bit interface where the RTX 3070 has a 320-bit interface. This makes a huge difference in total memory bandwidth, as you can see with the 3070 reaching 760 GB/s and the RTX 3060 Ti reaching 448 GB/s which matches the 2060 SUPER. The RTX 3060 Ti has a TGP of 200 watts which is a huge drop as well, which is nice to see but still more than the old 2060 SUPER.

 

Specifications	RTX 2060 SUPER FE	RTX 2080 SUPER	RTX 3060 Ti FE	RTX 3070 FE
SMs	34	48	38	46
CUDA Cores	2176	3072	4864	5888
Tensor Cores	272 (2nd Gen)	384 (2nd Gen)	152 (3rd Gen)	272 (3rd Gen)
RT Cores	34 (1st Gen)	48 (1st Gen)	38 (2nd Gen)	68 (2nd Gen)
Texture Units	136	192	152	272
ROPs	64	64	80	96
GPU Boost Clock	1650 MHz	1937 MHz	1665 MHz	1750 MHz
Memory Clock	7000 MHz	7748 MHz	7000 MHz	7000 MHz
Total Video Memory	8192 MB GDDR6	8192 MB GDDR6	8192 MB GDDR6	10240 MB GDDR6X
Memory Interface	256-bit	256-bit	256-bit	320-bit
Memory Bandwidth	448 GB/s	496 GB/sec	448 GB/s	760 GB/s
TGP	175 Watts	250 Watts	200 Watts	320 Watts
Launch MSRP	$399	$699	$399	$699

 

For every review, I always double check the specifications listed against our sample using GPUz. This is just to confirm that we don’t have any inconsistencies which for the RTX 3060 Ti Founders Edition GPUz has the boost clock at 1665 MHz which matches Nvidia’s specifications. I also do this to document the driver used, which in this case was the pre-launch driver provided by Nvidia and is listed as a beta driver. Our BIOS version is also listed, for reference if problems come up in the future.