Introduction

Today we have with us our first NVIDIA GeForce RTX 5060 Ti 8 GB graphics card to comprehensively review. We couldn't get a hold of an 8 GB RTX 5060 Ti ahead of the review posting date, and as far as we can tell, no other publication could, so we had to buy one off the shelf and put it through our test bench. We picked the card priced closest to the NVIDIA MSRP, the Gainward RTX 5060 Ti 8 GB Ghost.



NVIDIA launched the RTX 5060 Ti with two memory variants, 16 GB and 8 GB, with both variants being launched simultaneously. This is different from the RTX 4060 Ti, which launched as an 8 GB variant across a narrow 128-bit wide GDDR6 memory bus, and felt like a step backward from the previous-gen RTX 3060 Ti with its 256-bit memory bus, and the RTX 3060, with its 12 GB across a 192-bit memory bus. NVIDIA tried to appease the pushback from the market with the introduction of the RTX 4060 Ti 16 GB variant, which improved high-resolution performance, but was still hamstrung by memory bandwidth due to that narrow memory bus. With the new RTX 5060 Ti, NVIDIA breathed life back into the memory situation by implementing the new GDDR7 memory standard, and giving the RTX 5060 Ti 28 Gbps GDDR7 speeds--a significant 55% increase in memory bandwidth over the RTX 4060 Ti and its 18 Gbps GDDR6.

With the bandwidth situation sorted, the attention turns back to memory size. Here, NVIDIA has been smart with its approach to the market, by simultaneously launching the 16 GB and 8 GB variants. Both use 28 Gbps GDDR7 for the same 448 GB/s of memory bandwidth, so those wanting the future-proofing of 16 GB, or want to explore new use-cases such as 1440p with DLSS can opt for that card, whereas those purely gaming on 1080p can save a chunk of cash by opting for the 8 GB variant.

On paper, the 8 GB variant is supposed to be priced at $380, and the 16 GB variant at $430, but we now know that these prices are pure fiction. The RTX 5060 Ti 16 GB is realistically available from $500 mark, $70 higher than NVIDIA's starting price; while the RTX 5060 Ti 8 GB is priced closer to its natural MSRP. We bought our card for 400 Euros including VAT, close to the NVIDIA MSRP, so the real world price difference between the 8 GB and 16 GB variants is closer to $120 than the $50 NVIDIA wants you to believe.

Besides memory, both variants have an identical hardware configuration. They're both based on the GB206 silicon, which they max out, enabling all 36 streaming multiprocessors (SM) present, which works out to 4,608 CUDA cores, 144 Tensor cores, 36 RT cores, 144 TMUs, and 48 ROPs. Besides updating the memory sub-system to GDDR7, the PCIe host interface has been updated to PCI-Express Gen 5, and although the interface is still x8, Gen 5 brings bandwidth back up to what it was with the RTX 3060 Ti with its PCI-Express Gen 4 x16.

The GeForce RTX 5060 Ti is powered by the Blackwell graphics architecture, which introduces Neural Rendering, a new concept in consumer 3D graphics, which leverages generative AI to create richly detailed visual assets in real time, combining it with the rest of the conventionally rendered game. This relies on the GPU being able to accelerate AI models and render 3D graphics in tandem, and making it possible is the AI Management Processor (AMP), a new hardware scheduler that's being introduced with Blackwell. DLSS 4 replaces the convoluted neural networks (CNN) based AI models driving the various components of DLSS with newer Transformer-based models, which are more accurate, and improve image quality at every performance tier. Multi Frame Generation is a new Blackwell-exclusive feature that lets the GPU create up to 3 frames succeeding every conventionally rendered frame using AI, effectively quadrupling frame-rates. This relies on hardware flip-metering, which is being introduced with the new updated display engine.

NVIDIA is building the RTX 50-series GPUs on the same NVIDIA 4N foundry node as the RTX 40-series Ada generation, so all efficiency improvements you see are purely a function of the architecture. The new Blackwell SM comes with concurrent FP32+INT32 execution capability on all CUDA cores in the SM, not just half of them. The new CUDA cores and shader execution reordering components come with awareness for Neural Shaders. The new 5th Gen Tensor core comes with FP4 capability for even more throughput. The new 4th Gen RT core improves performance and comes with even more fixed-function hardware, which enables Mega Geometry, or the concept of increasing the poly counts of ray traced objects using hierarchical memory structures, just like with Mega Textures. The new display engine supports DisplayPort 2.1 with UHBR20, while the new NVENC and NVDEC video accelerators come with support for 4:2:2 color formats.

The Gainward RTX 5060 Ti 8 GB Ghost comes with a simple, no-frills board design that uses an aluminium fin-stack heatsink that's ventilated by a pair of fans. The card has all the basics covered, such as running the RTX 5060 Ti at reference clock speeds. It even comes with a legacy 8-pin PCIe power connector that some gamers with older PSUs might find convenient. As we mentioned earlier, we got our card for 400 Euros including VAT, which would put it close to the NVIDIA MSRP of $375.