Product Listing

GeForce GTX 960 performance review: NVIDIA targets the "sweet spot" GFX market (Updated)

By Wong Chung Wee & Vijay Anand - 28 Jan 2015
Launch SRP: S$369

The NVIDIA GeForce GTX 960 targets the "sweet spot" graphics market segment!

** Updated on 28th Jan - We've updated the article to include more references and performance figures of a Radeon R9 280 graphics card to provide an updating conclusion as to how the new NVIDIA GeForce GTX 960 fares. The original article was first published on 22nd Jan.

ASUS GeForce GTX 960 Strix 2GB GDDR5 OC Edition is one of the earliest GTX 960 graphics cards to reach our lab. (Image source: ASUS)

Maxwell goes mainstream

To kick off the new year for the discrete graphics market segment, NVIDIA has launched the GeForce GTX 960 graphics card. If you are wondering why we taken off the mantle to reveal an add-in card partner's ware, i.e., the ASUS GeForce GTX 960 Strix 2GB GDDR5 OC Edition, instead of a reference one from NVIDIA, it's because the mid-range series are given more leeway for partners to tweak their products. As such, a "reference GeForce GTX 960" may not give you the expectations of what's in the retail channel and NVIDIA has tasked its partners to send over their wares to relevant publications. Fortunately, ASUS is one of the earliest to send over their take of the GeForce GTX 960 to us and we'll be using it as our base comparison model of what you can expect out of the new GPU SKU.

The GeForce GTX 960 has a new GM206 GPU that is based on the same Maxwell architecture, like the earlier and more powerful GTX 980 and GTX 970 cards. Let's take a closer look at this new GPU.

The schematic of the GM206 chip (Image source: NVIDIA)

GM206: The newest kid on the block

According to NVIDIA, the GM206 GPU features "all the key architectural innovations" that were first introduced with the GeForce GTX 980. In essence, the Maxwell second generation architecture features a re-designed Streaming Multiprocessor (SMM) that touts efficiency on two fronts; power consumption and instructions processing. As a result, NVIDIA claims that the delivered performance per CUDA core has been improved by as much as 40% against previous-generation Kepler GK104 GPU.

The detailed block diagram of the Maxwell Streaming Multiprocessor (SMM) unit. A group of 8 of these form a Graphics Processing Cluster (GPC) and the GTX 960 GPU SKU has two GPC blocks as shown in the earlier GTX 960 schematic.

Speaking of CUDA cores, the Maxwell Streaming Multiprocessor (SMM) has housed them, and their accompanying components, in a quadrant design. There are 32 CUDA cores in each quadrant processing block along with its own dedicated warp scheduler. This translates to the SMM having a total of 128 CUDA cores along with eight texture units, Polymorph engine 3 and 96KB of shared memory that is in addition to its 48KB of texture/L1 cache. If you are curious, the stats and the genetics of the SMM are identical to the GTX 980's SMM block in its GM204 GPU. So there's nothing new here.

The GM206 GPU has a total of 8 SMM units for a total of 1024 CUDA cores, spread across to two Graphics Processing Clusters (GPCs). This is exactly half the processing power of the GM204 GPU used in the GTX 980 which boasts four GPC units with a total of 2048 CUDA cores. Since each GPC is accompanied by a 64-bit memory controller bus, the reduction would mean the GeForce GTX 960 will only have a 128-bit wide graphics memory bus

The so-called 'reference' GeForce GTX 960 is rated to perform at a base clock speed of 1126MHz, and a boost clock of 1178MHz. Its 2GB of GDDR5 memory modules have a clock speed of 7010MHz, with a total memory bandwidth of 112.16GB/s. Will the Maxwell architecture still shine in its reduced form that is the GM206? We'll soon find out.

Have a look at the table below to see how the GeForce GTX 960's specifications compare against the competition. Take note for our performance preview of the GeForce GTX 960 in the later page, we're actually using the ASUS GeForce GTX 960 Strix 2GB GDDR5 OC Edition. As such, its core and memory are clocked roughly 100MHz and 200MHz more (respectively) than a standard non overclocked counterpart.

NVIDIA GeForce GTX 960 2GB GDDR5 and other competitive GPUs compared
  NVIDIA GeForce GTX 960 ASUS Strix GeForce GTX 960 NVIDIA GeForce GTX 760 AMD Radeon R9 270X AMD Radeon R9 280
  NVIDIA GeForce GTX 960 ASUS Strix GeForce GTX 960 NVIDIA GeForce GTX 760 AMD Radeon R9 270X AMD Radeon R9 280
Core Code
  • GM206
  • GM206
  • GK104
  • Curacao XT
  • (Pitcairn XT variant)
  • Tahiti Pro
GPU Transistor Count
  • 2.94 billion
  • 2.94 billion
  • 3.54 billion
  • 2.8 billion
  • 4.3 billion
Manufacturing Process
  • 28nm
  • 28nm
  • 28nm
  • 28nm
  • 28nm
Core Clock
  • 1126MHz (Boost: 1178MHz)
  • Gaming mode: 1228MHz (Boost: 1291MHz)
  • OC mode: 1253MHz (Boost: 1317MHz)
  • 980MHz
  • 1050MHz
  • up to 933MHz
Stream Processors
  • 1024
  • 1024
  • 1152 CUDA cores
  • 1280
  • 1792
Stream Processor Clock
  • 1126MHz
  • 1228MHz
  • 980MHz
  • 1050MHz
  • up to 933MHz
Texture Mapping Units (TMUs)
  • 64
  • 64
  • 96
  • 80
  • 112
Raster Operator units (ROP)
  • 32
  • 32
  • 32
  • 32
  • 32
Memory Clock (DDR)
  • 7010MHz
  • 7200MHz
  • 6008MHz DDR
  • 5800MHz
  • 5000MHz
Memory Bus width
  • 128-bit
  • 128-bit
  • 256-bit
  • 256-bit
  • 384-bit
Memory Bandwidth
  • 112.16 GB/s
  • 115.20 GB/s
  • 192.3GB/s
  • 179.2GB/s
  • 240GB/s
PCI Express Interface
  • PCI Express 3.0
  • PCI Express 3.0
  • PCIe ver 3.0 x16
  • PCIe v3.0 x16
  • PCIe v3.0 x16
Power Connectors
  • 1 x 6-pin
  • 1 x 6-pin
  • 2 x 6-pin
  • 1 x 6-pin power cable
  • 1 x 6-pin power cable; 1 x 8-pin power cable
Multi GPU Technology
  • SLI
  • SLI
  • SLI
  • AMD CrossFire
  • CrossFireX
DVI Outputs
  • 1
  • 1
  • 1 x DVI-D, 1 x DVI-I
  • 1
  • 2
HDMI Outputs
  • 1
  • 1
  • 1
  • 1
  • 1
DisplayPort Outputs
  • 3
  • 3
  • 1
  • 1
  • 1
HDCP Output Support
  • Yes
  • Yes
  • Yes
  • Yes
  • Yes

Beyond raw specs of the GM206, it is capable of supporting the new technology fronts made possible by the Maxwell second generation architecture improvements. They are Dynamic Super Resolution (DSR), Multi-Pixel Programmable Sampling with Multi-Frame Sampled AA (MFAA), Voxel Global Illumination (VXGI), and VR Direct. Some promise to deliver better gaming experiences due to more realistic visuals while others focus on performance gains. For example, MFAA can net you higher visual quality at a lower performance overhead or you can trade quality for higher performance.

The GeForce GTX 960 has three DisplayPorts, one HDMI interface and a dual-link DVI-I port as seen on the ASUS GeForce GTX 960 Strix 2GB GDDR5 OC Edition.

The GTX 960 is also positioned as an Ultra 4K HD gaming card, just like the GTX 980 and GTX 970. But due to its low power envelope and reduced horsepower, the card is more ideal to double up as a component in a HTPC machine where high resolution output and some gaming caliber is necessary while maintaining a smaller form factor. Many add-in card partners offer custom cooling solutions where the fans turn off when the GPU temperature falls below a certain threshold. Such passive cooling measures translate to quiet operation; another feather in its hat as a HTPC discrete graphics component. In terms of video connectivity options, the card has three DisplayPorts, one HDMI interface and a dual-link DVI-I port.

To reiterate, the GeForce GTX 960 is aimed at the mass market graphics segment and its performance isn't the same as the top-end GTX 900 cards. At US$199, this is NVIDIA's new "sweet spot" graphics card to replace the GTX 760 whose original launch price was US$249 and has been further revised to US$219 when the GTX 980 debuted

 

Test Setup

** Updated on 28th Jan 2015 - We've now included the Radeon R9 280 graphics card as part of our test mix.

These are the specifications of our graphics testbed:

  • Intel Core i7-3960X (3.3GHz)
  • ASUS P9X79 Pro (Intel X79 chipset) Motherboard
  • 4 x 2GB DDR3-1600 G.Skill Ripjaws Memory
  • Seagate 7200.10 200GB SATA hard drive (OS)
  • Western Digital Caviar Black 7200 RPM 1TB SATA hard drive (Benchmarks + Games)
  • Windows 7 Ultimate SP1 64-bit

Below is the list of cards that we lined up for comparison:-. 

  • ASUS GeForce GTX 960 Strix 2GB GDDR5 (ForceWare 347.25)
  • NVIDIA GeForce GTX 760 2GB GDDR5 (ForceWare 347.09)
  • ASUS R9 280 Strix 3GB GDDR55 (AMD Catalyst 14.12)
  • Sapphire Radeon R9 270X Vapor-X 2GB GDDR55 (AMD Catalyst 14.12) 

The ASUS Strix GeForce GTX 960 was the card provided for our GTX 960 testing and it was tested at its rated clock speeds, which is a little faster than a reference GTX 960, but it should be more representative of the various overclocked GTX 960 graphics cards that will be populating the market. Take note that the ASUS Strix GeForce GTX 960 graphics card has a default boost clock of 1291MHz. It can however go up to 1317MHz if you install the ASUS GPU Tweak utility and set the graphics card to "OC game mode". For this article, we've left the ASUS card to operate in its default mode.

Naturally, we also included the Kepler-based GTX 760 for comparison as the incoming GTX 960 is its direct replacement. We also compared the new NVIDIA GPU against its closest AMD competitor, the Radeon R9 270X. If you take a close look at the GPU comparison table shown earlier, you would realize that the technical specifications of the Radeon R9 270X has the closest match to the GeForce GTX 960. This would make a great comparison from a technical point of view on where both these cards stack up. It is also the fastest card from AMD that officially works off just one 6-pin PCIe power input, just like the GTX 960. From a cost point of view, the reference Radeon R9 270X was initially launched for US$199 and the retail models are currently still being sold at a similar price point between US$179 to US$229. The representative test card for this class is the Sapphire Radeon R9 270X Vapor-X 2GB GDDR5, which we've tested at its rated clock speed.

Of course, you might also argue that the Radeon R9 280 is a better specced option at the same US retail price range that the R9 270X sells for, but we've run out of these cards for comparison during this article but we'll update it within the next few days once we get a unit and get the latest test results. For this, we've enrolled the ASUS R9 280 Strix graphics card, which is about 50MHz and 200MHz faster at the core and memory clocks over the reference Radeon R9 280. Then there's also an updated lateral variant of the R9 280 which is the Tonga GPU based R9 285 - but it goes at a much higher retail price bracket of US$220 to US$260, so we'll leave that out of direct comparison for the moment. For now, the Radeon R9 270X is still a decent competitive product.

Benchmarks

Here's the full list of benchmarks that we'll be using for our assessment:-

  • Futuremark 3DMark 2013
  • Crysis 3
  • Call of Duty: Ghosts
  • Thief
  • Hitman: Absolution

For our temperature, power consumption and overclocking testing, 3DMark 2013 was used.

Join HWZ's Telegram channel here and catch all the latest tech news!
8.0
  • Performance 8
  • Features 8
  • Value 7.5
The Good
Decent performance
The Bad
High operating temperature and power draw
ASUS GPU Tweak didn't work properly when setting performance profile
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.