ASUS GeForce RTX 5070 Ti TUF OC Review 79

ASUS GeForce RTX 5070 Ti TUF OC Review

Architecture »

Introduction

ASUS Logo

ASUS TUF Gaming GeForce RTX 5070 Ti OC is a premium custom design graphics card based on NVIDIA's latest performance-segment graphics card. For the RTX 5070 Ti, the TUF Gaming line of graphics cards are positioned between the company's SFF-Ready Prime series, and the high-end ROG Strix series. In its latest generation, the ASUS TUF Gaming carries forward its rustic industrial product styling from its past two generations. The Vented Exoskeleton is built on the idea of exposing as much of the heatsink underneath outside as possible, with the cooler shroud providing structural support. The cooler also debuts innovations such as the company's latest Axial-Tech fans that it shares between the TUF Gaming and ROG Strix cards (albeit with slightly different impeller shape), a new phase-change thermal pad for the GPU, a moisture-resistant PCB surface treatment, and dual-BIOS.



The new GeForce RTX 5070 Ti is located in a gray area between the performance and enthusiast segments. The RTX 5080 from last month was squarely enthusiast, the RTX 5070 Ti is a slightly different equation—the GPU isn't explicitly marketed for 4K Ultra HD gaming, instead putting it in the rather broad 1440p segment that has GPUs priced $400 and above. With a starting price of $750, the RTX 5070 Ti is expected to be a lot more than a 1440p-class GPU. The RTX 5070 Ti should provide plenty of performance for 4K Ultra HD at native resolution, if you know your way around game settings, or can use features such as DLSS.

The new Blackwell graphics architecture debuts a revolutionary new technology called Neural Rendering. It brings the awesome power of generative AI directly into the graphics rendering stack, combining objects created by a generative AI model with conventional raster 3D graphics. This is possible because the GPU is able to accelerate AI models in tandem with rendering graphics, thanks to a hardware scheduling component call the AI Management Processor (AMP). NVIDIA hasn't refreshed the foundry node on which these GPUs are built, they use the exact same NVIDIA 4N foundry node as the previous RTX 40-series, and so all performance-per-Watt improvements you see are purely a function of the architecture and the new power management technologies being introduced with it.

The new Blackwell streaming multiprocessor (SM) comes with concurrent FP32 and INT32 math capability across all its CUDA cores—on the older Ada SM, only half the cores had INT32 capability. The new shader execution reordering engine is aware of neural shaders and neural objects. The 5th Gen Tensor core comes with FP4 data format support, to increase throughput by trading in precision. The 4th generation RT core comes with even more fixed-function hardware, this time enabling Mega Geometry—a technique to increase the triangle count of ray traced objects by incorporating hierarchical techniques similar to Mega Textures.

DLSS 4 being introduced with Blackwell introduces new Transformer-based AI models replacing the convoluted neural networks (CNNs) of older generations of DLSS. These models handle super resolution (upscaling), ray-reconstruction, and frame generation. With DLSS 4, NVIDIA is introducing Multi Frame Generation (MFG), a technology with which the GPU generates up to three frames following a conventionally rendered one, effectively quadrupling framerates. It's exclusive to Blackwell because of the new hardware flip-metering component that's part of the Display Engine.

The RTX 5070 Ti is carved out from the same GB203 silicon powering the RTX 5080, but with 70 out of the 84 SM present on the silicon being enabled, besides 48 MB out of the 64 MB L2 cache present. This works out to 8,960 CUDA cores, 280 Tensor cores, 70 RT cores, 280 TMUs, and 96 ROPs. The memory sub-system is a massive upgrade over the older RTX 4070 Ti. The memory size has been increased to 16 GB, up from 12 GB; the memory bus to 256-bit wide, from 192-bit; and the memory speed is 28 Gbps over the 21 Gbps of the RTX 4070 Ti, resulting in 896 GB/s of memory bandwidth, or a whopping 77% increase in bandwidth. The GPU needs this to run its memory-sensitive AI models.

With its default P-mode (Performance BIOS), the ASUS TUF Gaming RTX 5070 Ti OC runs the GPU at 2588 MHz boost, compared to 2452 MHz reference. There is a software-based OC mode that can drive this further up to 2610 MHz boost. ASUS is pricing the card at $1,000, a massive 33% premium over the $750 NVIDIA baseline.

NVIDIA GeForce RTX 5070 Ti Market Segment Analysis
 PriceCoresROPsCore
Clock
Boost
Clock
Memory
Clock
GPUTransistorsMemory
RTX 3080$4208704961440 MHz1710 MHz1188 MHzGA10228000M10 GB, GDDR6X, 320-bit
RTX 4070$4905888641920 MHz2475 MHz1313 MHzAD10435800M12 GB, GDDR6X, 192-bit
RX 7800 XT$4403840962124 MHz2430 MHz2425 MHzNavi 3228100M16 GB, GDDR6, 256-bit
RX 6900 XT$45051201282015 MHz2250 MHz2000 MHzNavi 2126800M16 GB, GDDR6, 256-bit
RX 6950 XT$63051201282100 MHz2310 MHz2250 MHzNavi 2126800M16 GB, GDDR6, 256-bit
RTX 3090$900104961121395 MHz1695 MHz1219 MHzGA10228000M24 GB, GDDR6X, 384-bit
RTX 4070 Super$5907168801980 MHz2475 MHz1313 MHzAD10435800M12 GB, GDDR6X, 192-bit
RX 7900 GRE$53051201601880 MHz2245 MHz2250 MHzNavi 3157700M16 GB, GDDR6, 256-bit
RTX 4070 Ti$7007680802310 MHz2610 MHz1313 MHzAD10435800M12 GB, GDDR6X, 192-bit
RTX 4070 Ti Super$75084481122340 MHz2610 MHz1313 MHzAD10345900M16 GB, GDDR6X, 256-bit
RX 7900 XT$62053761922000 MHz2400 MHz2500 MHzNavi 3157700M20 GB, GDDR6, 320-bit
RTX 3090 Ti$1000107521121560 MHz1950 MHz1313 MHzGA10228000M24 GB, GDDR6X, 384-bit
RTX 4080$94097281122205 MHz2505 MHz1400 MHzAD10345900M16 GB, GDDR6X, 256-bit
RTX 4080 Super$990102401122295 MHz2550 MHz1438 MHzAD10345900M16 GB, GDDR6X, 256-bit
RX 7900 XTX$82061441922300 MHz2500 MHz2500 MHzNavi 3157700M24 GB, GDDR6, 384-bit
RTX 5070 Ti$7508960962295 MHz2452 MHz1750 MHzGB20345600M16 GB, GDDR7, 256-bit
ASUS RTX 5070 Ti
TUF OC
$10008960962295 MHz2588 MHz
(+136 MHz)
1750 MHzGB20345600M16 GB, GDDR7, 256-bit
RTX 5080$1000107521122295 MHz2617 MHz1875 MHzGB20345600M16 GB, GDDR7, 256-bit
RTX 4090$2400163841762235 MHz2520 MHz1313 MHzAD10276300M24 GB, GDDR6X, 384-bit
RTX 5090$2000217601762017 MHz2407 MHz1750 MHzGB20292200M32 GB, GDDR7, 512-bit
Next Page »Architecture
View as single page
Feb 21st, 2025 15:06 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts