Introduction

We have with us the MSI GeForce RTX 5070 Ti Gaming Trio OC+. The Gaming Trio brand of graphics card almost single-handedly built the MSI Gaming brand for PC gamers, going back to the early 2010s. Since then, it has represented a solid combination of a factory overclock, low noise, slick looks when installed, and a well-executed RGB LED setup. Over the years, MSI topped the Gaming Trio with tiers such as Gaming X, the more premium Gaming Z, and more recently, the new SUPRIM and Vanguard series to be positioned above, but Gaming Trio remains the iconic brand-defining model. The RTX 5070 Ti Gaming Trio OC+ is the first Gaming Trio series card from the RTX 50 Series. The GeForce RTX 5070 Ti is NVIDIA’s third GPU launch from this generation, and occupies an interesting performance-price proposition that puts it in the gray area between the performance and enthusiast market segments, with a starting price of $750.

The GeForce Blackwell graphics architecture introduces Neural Rendering, allowing a generative AI model to participate in graphics rendering by blending AI-generated objects, materials, and textures with rasterized 3D. AI now goes beyond DLSS detail reconstruction. The new DLSS 4 and Multi Frame Generation (MFG) are major upgrades. DLSS 4 replaces CNN-based AI with a more accurate Transformers-based model, enhancing super resolution, ray reconstruction, and frame generation. MFG, enabled by Blackwell’s Flip Metering, lets the GPU predict up to three future frames, quadrupling frame rates. A single rendered pixel can generate 15 more.

The new Blackwell SM (streaming multiprocessor) comes with concurrent FP32 and INT32 capability for all 128 of its CUDA cores, unlike on Ada, where only half the cores were INT32 capable. It also comes with a redesigned shader execution reordering (SER) that’s aware of neural shader objects. The new 5th Generation Tensor core comes with FP4 data format support, stepping up throughput, trading in precision. The 4th Generation RT core has new hardware capabilities, enabling Mega Geometry—the ability for ray traced objects to have exponentially higher polygon count using a technique not all that different from Mega Textures—all those polygons come with the cost of intersecting each of them with rays.

The GeForce RTX 5070 Ti is carved out from the GB203 silicon as the RTX 5080. It enables 70 out of 84 streaming multiprocessors (SM), which works out to 8,960 CUDA cores, 280 Tensor cores, 70 RT cores, 280 TMUs, and 96 ROPs. This reduced ROP count is because an entire GPC has been disabled to lower the SM count. The RTX 5070 Ti also gets 48 MB out of the available 64 MB. The memory subsystem is almost identical to the RTX 5080 spare for slightly lower speed, but it is a massive upgrade over the previous-generation RTX 4070 Ti. The new RTX 5070 Ti gets 16 GB of memory across a 256-bit wide memory bus, compared to the 12 GB over 192-bit of the RTX 4070 Ti. This memory is also significantly faster at 28 Gbps GDDR7, resulting in 896 GB/s of memory bandwidth, a massive 77% increase in bandwidth over that of the RTX 4070 Ti. This should prove essential given the memory-sensitive AI models these GPUs will be running.

The MSI RTX 5070 Ti Gaming Trio OC+ features a slightly toned down version of the Tri Frozr 4 cooling solution that we’ve seen in the Vanguard series. The vapor chamber plate from the Vanguard is replaced by a nickel-plated copper baseplate. There are five heatpipes instead of eight and the heatsink is lighter as a result. The StormForce axial airflow fans, however, are carried over. MSI is giving the RTX 5070 Ti factory overclocked speeds of 2572 MHz compared to 2452 MHz reference. MSI is expected to price this card at $980, which is still a 30% premium over the NVIDIA baseline.