Nvidia GeForce GTX 960: Maxwell In The Middle

Test System And Benchmarks

Gaming

This article marks the first international use of Tom's Hardware's reference platform, a forward-looking system based on Intel's Core i7-5930K. We overclocked it to 4.2GHz, which should be suitable to alleviate platform bottlenecks from the game benchmark results.

As always, we strive to represent results across a wide range of relevant graphics hardware. For this launch, that means we're including the GeForce GTX getting replaced and the end-of-lifed GeForce GTX 770. From AMD, we selected the GeForce GTX 960's main competition, the Radeon R9 285, in addition to the Radeon R9 280X. Finally, we included the GeForce GTX 970 and Radeon R9 290X to show what the next tier of graphics cards is capable of.

Again, Nvidia did not supply us with a reference board for testing. We were forced to rely on the factory-overclocked Asus Strix GeForce GTX 960 instead. While we're disappointed that we aren't able to provide baseline performance results, the selling point of many third-party models with aggressive factory overclocks is only $210, just $10 higher than the base MSRP. As such, Asus' Strix give us an acceptable representation of well-priced factory overclocked cards.

We tested some of the most popular games at 1920x1080 and 3840x2160. The 4K resolution is equivalent to four 1080p monitors. Despite that massive number of pixels, 4K displays are becoming more popular every day thanks to sub-$600 options like Asus' PB298Q:

Unlike older models that require splitting a single video stream into two HDMI inputs, this 28" display is capable of 3840x2160 video at 60Hz over a single DisplayPort 1.2 cable. You can read more about the screen in Asus PB287Q 28-Inch 4K Monitor Review: Ultra HD For $650.

We tried to use game settings that the GeForce GTX 960 can handle at those resolutions. In many cases, the 960 simply isn't able to cope with 4K at the lowest possible detail levels. But in all cases it facilitates attractive detail settings at 1080p.


Test System
CPU
Intel Core i7-5930K (Haswell-E), 3.5/3.7GHz, Six Cores, LGA 2011-v3, 15MB Shared L3 Cache, Hyper-Threading enabled, Overclocked to 4.2GHz
Motherboard
MSI X99S Xpower AC (LGA 2011-v3) Chipset: Intel X99 Express, BIOS v1.5
Networking
On-Board Gigabit LAN controller
Memory
Crucial Ballistix DDR4-2400, 4 x 4GB, 1200MHz, CL 16-16-16-39 2T
Graphics
EVGA GeForce GTX 970
1050/1178MHz GPU, 4GB GDDR5 at 1753MHz (7012 MT/s)

Nvidia GeForce GTX 770

1046/1085MHz GPU, 2GB GDDR5 at 1752MHz (7008MT/s)

Asus Strix GeForce GTX 960
1253/1317MHz GPU, 2GB GDDR5 at 1800MHz (7400MT/s)

Nvidia GeForce GTX 760
980/1033MHz GPU, 2GB GDDR5 at 1502MHz (6008MT/s)

AMD Radeon R9 290X
1000MHz GPU, 4GB GDDR5 at 1250MHz (5000MT/s)

AMD Radeon R9 280X

850/1000MHz GPU, 3GB GDDR5 at 1500MHz (6000MT/s)

AMD Radeon R9 285
918MHz GPU, 2GB GDDR5 at 1375MHz (5500MT/s)
SSD
Samsung 840 Pro, 256GB SSD, SATA 6Gb/s
Power
Be Quiet! Dark Power Pro 10, 850W, ATX12V, EPS12V
Software and Drivers
Operating System
Microsoft Windows 8 Pro x64
DirectX
DirectX 11
Graphics Drivers
All GeForce Cards: Nvidia 347.25 Beta Driver
All Radeon Cards: AMD Catalyst Omega14.12
Benchmarks
Middle-earth: Shadow of Mordor
Version 1.0.1806.18, Built-In Benchmark
Arma 3
Version 1.38, 30-sec. Fraps "Infantry Showcase"
Battlefield 4
Version 1.3.2.3825, Custom THG Benchmark, 90-Sec
Assassin's Creed Unity
Custom THG Benchmark, 60-Sec
ThiefVersion 1.7, Built-In Benchmark
Alien: Isolation
Built-In Benchmark
Far Cry 4
Version 1.01, Custom THG Benchmark, 30-sec Fraps

Power Consumption

Nvidia's Maxwell architecture presents us with a new set of challenges for measuring power consumption. If all four rails are to be measured exactly (to figure out Maxwell’s efficiency secrets), then a total of eight analog oscilloscope channels are needed. This is because voltage and current need to be recorded concurrently at each rail in real-time. If the voltages are measured and then used later, the result may be inaccurate. So, how do we solve this problem?

We enlisted the help of HAMEG (Rohde & Schwarz). In the end, we had to use two oscilloscopes in parallel (a master-slave triggered setup), allowing us to accurately measure and record a total of eight voltages or currents at the same time with a temporal resolution down to the microsecond.

The measurement intervals need to be adjusted depending on the application in question, of course, in order to avoid drowning in massive amounts of data. For instance, when we generate the one-minute graphs for graphics card power consumption with a temporal resolution of 1ms, we have the oscilloscope average the microsecond measurements for us first.

We use a riser card on the PCIe slot (PEG) to measure power consumption directly on the motherboard for the 3.3 and 12V rails. The riser card was built specifically for this purpose.

In addition, we separately measure the voltage and current at each of the two individual PCIe power connectors.

Test MethodologyNo-contact current measurement at all rails
Direct voltage measurement
IR real-time monitoring
Test Equipment2 x HAMEG HMO3054, 500MHz Four-Channel Oscilloscope with Data Logger
4 x HAMEG HZO50 Current Probe
4 x HAMEG HZ355 (10:1 Probe, 500MHz)
1 x HAMEG HMC8012 DSO with Data Logger
1 x Optris PI450 80Hz Infrared Camera + PI Connect
Test SystemIntel Core i7-5960X, 4.2GHz
16 GB G.Skill Ripjaws DDR4-2666 (4 x 4GB)
MSI X99 Gaming 7
2 x Transcend SSD370 (System, Applications + Data, Storage)
be quiet! Dark Power Pro 1200W
Microcool Banchetto 101

Temperatures

Lower power consumption should result in lower GPU temperatures under load, which in turn should necessitate a simpler and quieter cooling solution. We’re focusing on the tested cards behavior in a closed case, given the number of reader requests we've received about this topic.

We’ve identified a method to confirm what our sensors tell us and to spice up our usual temperature graphs a bit in the form of the PI450 by Optris. This piece of equipment is an infrared camera that was developed specifically for process monitoring.

It supplies real-time thermal images at a rate of 80Hz. The pictures are sent via USB to a separate system, where they can be recorded as video. The PI450’s thermal sensitivity is 40mK, making it ideal for assessing small gradients.

Noise

Our noise measurements are performed using a calibrated high-end studio microphone positioned perpendicular to the middle of the graphics card in question at a distance of 50cm. This distance, in conjunction with the microphone's strong cardioid directionality, represents a compromise between avoiding noise due to fan turbulence and avoiding ambient noise, which can never be completely eliminated. We like to perform our noise measurements at night for this reason.

This thread is closed for comments
226 comments
    Your comment
  • Novuake
    This seems meh... Impressive but not phenomenal power consumption to performance numbers. Especially compared the GTX970/980.

    Would have liked to see two more things.

    1. More extensive AA. post processing and memory bandwidth testing. Pretty sure Nvidia hamstrung the card a bit in some scenarios with a 1280bit interface. I had to read it 4 times before I believed it and still am skeptical.

    2. Overclocking benchies.


    So otherwise I guess we are back to the "old" ti-designation setup where the GTX960ti SHOULD be based on GM206 and vanilla GTX960 is not.
  • sconzen
    I may be blind, but I don't see the Zotax Amp! edition in the temperature and noise tests. Confirm?
  • damric
    The R9 280 is the fast and cheap elephant in the room that was never mentioned in this review,
  • Grognak
    Well, I'm not saying a 10% improvement on top of a reduced power consumption isn't nice, because it really is, however we're still quite far away from the 770. I suppose Nvidia has a card planned to fill the massive performance gap between the 960 and 970, one at 4Gb of VRAM maybe?
  • ykki
    The damn arrows are STILL blocking the charts!
  • sconzen
    I may be blind, but I don't see the Zotax Amp! edition in the temperature and noise tests. Confirm?
  • ykki
    Great review.
    Now AMD, time to bust out the 270x!
  • ykki
    Lol I meant the 370x!
  • Novuake
    1427918 said:
    The damn arrows are STILL blocking the charts!


    I am not the only one! Thank you!
  • ykki
    Novoake, I am very sorry but I stole your comment from an earlier review.
    But seriously, those arrows can block out the sun if tom's put 'em right.
  • ImDaBaron
    The Gainward version of this card has to be the ugliest video card I've ever seen
  • maestro0428
    Yes, the arrows on the charts bug me too. I am a bit let down by the performance here. Sure, the efficiency is great, but bandwidth is just to low to play above 1080p. I was hoping to put two of these in SLI, but I am afraid it won't do my Surround set up justice. Looks like I may be going with a single 980 or eventually two 970s. Bummer.
  • Novuake
    1427918 said:
    Novoake, I am very sorry but I stole your comment from an earlier review. But seriously, those arrows can block out the sun if tom's put 'em right.


    I do not recall posting such a comment? But I may have been frustrated that day. Hehe I am quite outspoken.
  • silverblue
    Not bad at all. R9 285 or better performance for 2/3 the power (or better), and for a little cheaper as well - a great 1080p card.
  • xXComputer_Nerd1625Xx
    Honestly I'm a little let down by the specs. I know specs aren't everything, but I really would've liked to see a beefier GPU compared to the last-gen 760 (which on paper looks better) and that also surpassed the last-gen 770, which this card can hardly do. I've got to admit though, at $200 it still is a great card, and maybe I'm asking for too much.
  • gudomlig
    power charts seem questionable to me. 10 minutes running furmark drew an average of 280 watts with my sapphire 7950 boost and draws about 330 watts with my MSI gaming gtx 970. the power consumption torture charts show GTX 970 running at less watts than a 7950...I call BS. Even the older reviews here on Tom's show GTX 970 draws more power than 7950 boost so not sure where they pulled these numbers and makes me question the integrity of the entire article
  • Agera One
    Why don't you always put the Anti-aliasing GIF on a screen with no moving objects? That would be more accurate to see what changes !!
  • cmi86
    I appreciate what this card was able to do in regards to the lower TDP and slightly higher performance than the 760. That said I don't think this card did enough to win my vote. I am currently shopping for a new GPU and was really looking forward to this release thinking it would be a huge game changer like the 760 but it just wasn't. Now I'll be waiting for the 3XX release before making a decision.
  • mapesdhs
    Where are the EVGA ACX 2.0 960 results?

    Ian.
  • elbert
    It requires an overclock to match the 280/285 in 1080p and totally gets crushed in 4k. While costing more than 280. Most of the gtx960 8 pin overclocking versions are more than the 285. The power saving may offset one disadvantage but costing more while under performing lol.

    I was planing on buy two but this so bad ill wait and check out GTX960ti 1280sp.
  • salgado18
    And why did you use the Asus Strix overclocked card for performance benchmarks, but for power you use Gainwards, which can be downclocked to reference specs? That's heavy bias on the numbers, and saying "A reference-clocked GeForce GTX 960 would have been slower than Asus' specimen..." is no excuse, when you did make one! Come on, great review and writing, but that is one major slip!
  • logainofhades
    Really disappointed with this card. Hopefully a 960ti will be released to be what this card should have been. Guess I am going to buy a GTX 770, from a friend of mine instead. Odd to see that all the sudden the HD 7970/R9 280x is faster than the GTX 770 now. First time I have really seen that.
  • cleeve
    120171 said:
    And why did you use the Asus Strix overclocked card for performance benchmarks, but for power you use Gainwards, which can be downclocked to reference specs? That's heavy bias on the numbers, and saying "A reference-clocked GeForce GTX 960 would have been slower than Asus' specimen..." is no excuse, when you did make one! Come on, great review and writing, but that is one major slip!


    The answer is timing.

    Game benchmarks were don in our Canadian lab. The Asus card was the only one sent early enough for us to get game benchmarks in time.

    Power tests are done at the German lab, where the Asus card arrived later than the other samples. Unfortunately, it was unavoidable.
  • fw1374
    At page 9 three times it is written GeForce GTX 690 instead of 960. Just a typo but it is there :)