Review of the Nvidia GeForce GTX 1070

ZAH Computers Tech Editorial

ZAH Computers Hardware Research Desk

On May 17th, Nvidia introduced the GeForce GTX 1080, marking the launch of its latest flagship graphics card. In conjunction with this release, Nvidia discreetly unveiled the GeForce GTX 1070 as well. However, the specifications provided have been somewhat restricted. Today, we are pleased to present a review of the reference card, commonly referred to as the founders edition. While the two cards may appear visually similar, there are indeed several notable differences. In this review, we will guide you through all the details. The long-awaited moment has finally arrived! Both leading manufacturers of desktop dedicated GPUs are nearing the launch of their latest graphics processing units. This anticipation has been building for a considerable time, and for good reason. The GPU industry has experienced a standstill, awaiting the emergence of a more efficient GPU fabrication process. Previous generation GPUs were produced using a 28 nm fabrication process, and an intermediate transition to 20 nm was expected to address the current demands; however, this technology proved to be problematic. Aside from a few smaller ASICs, the 20 nm node has not met expectations. Consequently, the industry has been in a holding pattern until a newer and smaller fabrication process became available, which would enable a reduction in die size, leading to lower voltage consumption, decreased transistor gate leakage, and, importantly, an increase in the number of transistors within a GPU. The solution has been found in the recent advancements of 14/15/16 nm fabrication processes, utilizing the now-familiar FinFET + VLSI technology, which essentially adds "wings" to transistors. While Intel has been employing this technology for some time, both Nvidia and AMD are now transitioning to these advanced nodes. Nvidia has taken the lead by announcing their new products based on TSMC's 16 nm fabrication process, introducing the Pascal GPU architecture, named in honor of the mathematician, similar to previous architectures such as Kepler, Maxwell, and Fermi. Pascal has arrived, and today marks the release of reviews for the second product utilizing the GP104 GPU, the GeForce GTX 1070.

NVIDIA GeForce GTX 1070

16 nm GP104 silicon "GP104-200-A1" GPU
1920 CUDA cores
15 out of 20 streaming multiprocessors enabled
120 TMUs
64 ROPs
256-bit GDDR5 memory, 8 GB standard memory amount
Maximum GPU Boost frequency ~1,700 MHz
6.45 TFLOP/s single-precision floating point performance
150 W TDP, single 8-pin PCIe power connector
3x DisplayPort 1.4, 1x HDMI 2.0b

The GeForce GTX 1070 is fundamentally based on the Pascal GP104 GPU, although it has been intentionally downscaled for various reasons. While the GTX 1070 shares the same GP104 GPU architecture as the GTX 1080, it is a reduced version, as Nvidia has disabled certain segments. Specifically, the GTX 1080 features 2560 shader processors, whereas the GTX 1070 is equipped with 1920 shader processors. This configuration results in 15 out of the 20 streaming multiprocessors being active (15 SMs multiplied by 128 shader cores). Another significant alteration is the type of memory utilized; the GTX 1070 employs standard GDDR5 memory instead of the more advanced GDDR5X. The memory operates at a frequency of 2,000 MHz, translating to an effective GDDR5 speed of 8 GHz, with a memory bandwidth of 256 GB/s. The two distinctions contribute to a decrease in performance from 9 TFLOP/s in single-precision floating point calculations for the GeForce GTX 1080 to 6.45 TFLOP/s for the GeForce GTX 1070. This indicates that the latter still achieves approximately two-thirds of its potential. However, this performance disparity is reflected in the pricing, with the GeForce GTX 1070 starting at 379 USD, while the founders edition (reference card) is priced at 449 USD.