While it’s definitely nice to have the pixel pushing power of a GTX 1080 Ti, or RX Vega 64, one can’t overlook just how much power you can get for a considerably smaller investment. Graphics cards have come a long way in a relatively short amount of time since the first DX11 GPUs arrived. Today, one can have 2010-flagship beating performance for under a hundred dollars. So, what do the Green and Red teams have to offer in the $100~ price bracket? Today I’m putting NVIDIA’s second-smallest Pascal-based Graphics card, the GeForce GTX 1050 2GB, up against AMD’s Polaris-based Radeon RX 560, also in 2GB flavour. Before we start, let’s take a look at the GPUs powering these budget-friendly cards.

NVIDIA 1050 vs RX 560

The GPUs

NVIDIA GeForce GTX 1050

  • 132mm² die size
  • 3.3 billion transistors
  • 14nm FinFET manufacturing process
  • 640 CUDA cores
  • 40 TMUs
  • 32 ROPs
  • 128-bit memory interface
  • 1MB L2 cache
  • 2GB GDDR5
  • 7Gbps memory data rate
  • 1354MHz base clock
  • 1455 MHz Turbo clock

A simplified block diagram of the GP107 GPU. Note: Each memory controller represents a 32-bit channel.

The GeForce GTX 1050 is one of two graphics cards powered by the second-smallest implementation of the Pascal micro-architecture, the other being the GTX 1050 Ti. The only hardware difference between the two cards is the fact that the former has one of the GP107’s 6 SM clusters fused off, dropping the CUDA core count down to 640, from 768. This also reduces the TMU count from 48 to 40, and means GTX 1050 has to make do with five tessellation engines. An interesting fact about the GP107 chip is that it is the only Pascal GPU to be fabricated on a process other than TSMC’s 16nm; this processor is in fact fabbed on Samsung’s 14nm process. Despite having the same number of CUDA cores and the same width memory interface as the older GTX 750 Ti, GTX 1050 has significantly higher clock speeds and twice as many ROPs to boot, so performance should be quite a lot higher. NVIDIA specifies this part with a TDP of just 75 Watts, this means that the GPU can be powered from the PCI Express slot alone, but the card I am testing today has an additional six-pin for greater power delivery. The actual clock speed under load using its GPU boost technology for my GTX 1050 is 1695 MHz.

AMD Radeon RX 560 2GB

  • 123mm² die size
  • 3 billion transistors
  • 14nm FinFET manufacturing process
  • 1024 stream processors
  • 64TMUs
  • 16 ROPs
  • 128-bit memory interface
  • 1MB L2 cache
  • 2GB GDDR5
  • 7Gbps memory data rate
  • 1175MHz base clock
  • 1275 MHz Turbo clock

Simplified block diagram of thePolaris 21 GPU used in the RX 560

AMD’s Radeon RX 560 is based on a refined, likely better binned variant of the same Polaris 11 GPU core used in the RX 460. It is however renamed to Polaris 21. The main difference is that the 560 (at least one variant) has all of its 16 Compute Units enabled, giving the chip 1024 Shader processors and 64 Texture Mapping Units. There is actually a variant of the RX 560 that has the same CU count as the older 460, which is known as the ‘560D’. But today I am testing the fully enabled silicon. Compared to the Gp107 chip in the GTX 1050, Polaris 21 can process half as many pixels per clock (16 vs 32) but has significantly more shader ALU and ‘int8’ texture filtering power. (GCN does Fp16 texture filtering at half speed, as far as I know). Memory bandwidth is the same at 112GB/s, and both GPUs have a 1MB L2 cache. AMD rates this card with a typical board power varying between 60 and 80 watts. This means that some variants will come equipped with an additional 6-pin power connector. The card I am testing today has such a power connector.