The GTX 670 Arrives – is it a game changer?

Nvidia has just released the $399 GeForce GTX 670 based on its brand new 28nm Kepler DX11.1 architecture today. This new GPU is the third in Nvidia’s lineup, after the dual-GPU $1000 Flagship GTX 690 and the $500 GTX 680. To properly bring you this review we are going to cover the history of Kepler and why Nvidia is competing with GK104 instead of GK110, and we’ll give you an insider behind-the-scenes look at what probably went on inside AMD and inside Nvidia that set up for today’s launch. And of course, we are focusing on the performance of the GTX 680 and the HD 7970 to see where the new GTX 670 fits in.

The GTX 670 is Nvidia’s replacement for the GTX 570 which launched at the beginning of 2011 at $379. This time, Nvidia is aiming for slightly less than GTX 680 performance in a “value” package to directly compete with AMD’s top cards, the HD 7970 and the HD 7950. The reference HD 7970 is now selling for $480 after a major price drop down from $550 in response to the GTX 680. This is the third time in less than two months that Nvidia is again bringing out a faster card than or nearly equal in performance to AMD’s flagship HD 7970 – and this time the GTX 670 is selling for $80 less than the HD 7970.

 

 

So you will see us pit the reference GTX 670, both stock and overclocked, against our PowerColor reference design HD 7970 (above right) and also against the stock GTX 680 (above left) using 18 modern games and 3 synthetic benchmarks mostly using 1920×1200 and 2560×1600 resolutions.

We are also comparing the performance of our last generation reference top single-GPU video cards, HD 6970 and the GTX 580 as they are the fastest video cards of AMD’s and Nvidia’s last 40nm generation.  We are also tossing into the mix, the GTX 570 as it is being replaced by the GTX 670.  The only difference is that the GTX 570 launched at $20 less, for $379 and it was positioned mostly against the AMD flagship HD 6970 also at the same price; and it was also positioned against the HD 6950. This time the $399 GTX 670 is positioned against the (now discounted to) $480 AMD flagship, HD 7970 and also against the HD 7950.  Naturally we want to know if the HD 7970 is worth $80 more as most of them also now come with game bundles.

We will use 18 games and three synthetic benchmarks to compare our cards at 1920×1200 and 2560×1600. We shall also compare Nvidia’s 3-panel Surround working now off of a single GTX 670 at 5760×1080 resolution versus the HD 7970 running Eyefinity at the same resolution. Lastly, we also bench 3D Vision 2 and PhysX, ‘on’ versus ‘off’ at the very popular 1920×1080 resolution.

 

History of Nvidia’s Kepler and why it is based on GF104 and not GF110

There are a lot of misconceptions about why Nvidia brought out a GK104 256-bit bus high end card built on GF104 to replace the 384-bit flagship card, the GTX 580, built on Fermi’s GF110 design.  Last Autumn, it was not well known outside the industry, but well-known to AMD that Nvidia was complaining that TSMC simply could not build a 450nm “large die” on the 28nm process in time for them to compete with AMD’s upcoming Tahiti.

AMD expected that Nvidia would be very late as they were certain that the large die “Big Kepler” was impossible to produce until Q2 or Q3.  What they did not realize was that Nvidia was also working on GK104 to replace the very popular GF104 that featured the highly efficient GTX 560 Ti.  Evidently, Nvidia was originally planning to meet AMD with a “GTX 570 Ti” type card, so as to at least be close in performance to what they expected the HD 7970 to be. There was some quiet celebration at the end of last year that insiders noted as Nvidia apparently could not believe that AMD had brought out Tahiti under-performing compared to what they were expecting.  Nvidia may have really thought at the time of launch that perhaps part of Tahiti was disabled in AMD’s rush to be first as “aggressor” and “predator” as their CEO had boasted not long before.

Well, Nvidia spent the next few weeks refocusing their efforts to make Kepler as cool and as efficient as possible while still beating the HD 7970’s performance.  And because AMD was so certain that Nvidia would be many more months later with Kepler, they priced their 28nm cards at record highs for AMD pricing and they completely abandoned their “sweet spot” pricing in favor of “premium pricing.”

For example, AMD replaced the $379 HD 6970 with the $550 HD 7970 with barely a 33% performance increase – which was low enough for Nvidia to beat with their midrange-now-turned-flagship Kepler GK104 as the reincarnated GTX 680.  Because Nvidia launched the GTX 680 at $500, undercutting AMD’s pricing by $50 and beating AMD’s performance across the board, AMD was forced to discount their reference cards to about $480 from $550.

It is obvious that AMD is still trying to hold on to the $500 pricing with overclocked HD 7970s by increasing the base clocks to about 1000MHz, something easily attainable to anyone with an original reference 925MHz card, but now with less power draw from a more mature process.  This is the same strategy that AMD has employed with Cypress to meet the GTX 480 and with Cayman to meet the GTX 580.  However, this time things are very different and their strategy is much more likely to fail.

Unless AMD has a respin waiting – which means they would have been working on it since December (and this is highly unlikely now, as evidently they really believed that Nvidia would be many months later with Kepler than they were) – they may have lost this round as overclocking simply won’t cut it.  Kepler is more efficient than Tahiti.

 

What’s New with Kepler’s GTX 670?

Nvidia’s marketing buzzwords for the GTX 680 and GTX 690 launches were, “Faster. Smoother. Richer.” The GTX 670 is also designed for extreme efficiency and high performance.

Faster

The GTX 680’s Kepler architecture has 8 SMX units and 1536 CUDA cores.   In comparison, the more affordable GeForce GTX 670 ships with 1344 CUDA Cores and 7 SMX units.  The memory subsystem of the GeForce GTX 670 is identical to the GTX 680, consisting of four 64-bit memory controllers (256-bit) with 2GB of GDDR5 memory.  The base clock speed of the GeForce GTX 670 is 915MHz. The typical Boost Clock speed is 980MHz.  GeForce GTX 670’s memory speed is 6008MHz data rate.

The Kepler GTX 670 promises better geometry and texture processing than Fermi’s GTX 570 thanks to its improved instruction throughput and redesign. In addition, Nvidia brings “GPU Boost”, a dynamic way to boost clocks speeds and maximize performance for each game.

Smoother

New kinds of anti-aliasing – FXAA and TXAA – now compete with MSAA in terms of IQ while not sacrificing performance. Also, there is a new “Adaptive VSync” that is reduces tearing and stuttering associated with regular VSync. Great hardware needs great software to support it and Nvidia is also a software company.

Richer

For the first time, it is now possible to play games spanning 3 displays in Surround off of a single GeForce GTX 670.  And this time, the GTX 670 brings two dual-link DVI connectors, one HDMI and one DisplayPort connector (for a 4th accessor display) so that only one adapter is needed for any DVI-enabled display for 3-panel Surround.   In contrast, you will generally need two adapters for the HD 7970 to run the competing 3-panel Eyefinity and you may also experience more tearing with the AMD solution.

PhysX has also been improved.  The GeForce GTX 670 reference board measures 9.5″ in length. Two 6-pin PCIe power connectors are required for operation and if you forget to connect them, you will get an on-screen reminder.

How does the $399 GTX 670 compare with its rival, AMD’s HD 7970 at $480, down from $550 last month

This evaluation attempts to analyze and compare GTX 670, GTX 680 and GTX 570 performance. We also include HD 7970 performance as well as AMD’s fastest card of the last generation, their single-GPU HD 6970 and we will probably announce a price to performance winner. We will also look at the details to see what this new Nvidia Kepler enthusiast value GPU brings to the table for nearly four hundred dollars.

The five competing cards

The GTX 670, GTX 680, the GTX 570, the HD 7970, and the HD 6970 are the top single-GPU cards from Nvidia and AMD of this and of the last generation and we will see where they sit in relation to each other.  We believe that it is important to see how much performance increase the GTX 670 has brought over the GTX 570 so as to see if it is a worthy upgrade.

Since we do not want any chance of our CPU “bottlenecking” our graphics, we are testing all of our graphics cards by using our brand new Intel Core i7-3770K at 4.60GHz, 4 GB Kingston DDR3 and a Gigabyte Z77 motherboard. This new motherboard features an 8x + 8x PCIe 3.0 specification for CrossFire/SLI, but we are testing the GTX 670 using a 16x PCIe 3.0 slot. This new Core i7-3770K at 4.8GHz is more than enough to differentiate high end video cards at high resolution and high detail settings.

Before we do performance testing, let’s take a look at the GTX 670 and quickly recap its new Kepler DX11.1 architecture and features.

 

 

Architecture and Features

We have covered Fermi’s GK104 architecture in a lot of detail previously. You can can read our GTX 680 introductory article and and its follow-up. The new Kepler architecture builds on Fermi architecture with some important improvements and refinements that we will briefly cover before we get into performance testing.

 

SMX architecture

As Nvidia’s slide indicates, the new architecture is called SMX and it emphasizes 2x the performance per Watt of Fermi. Their multi-threaded engine handles all of the information using four graphics processing clusters including the raster engine and two streaming multi-processors.

The SM is now called the SMX cluster. Each SMX cluster includes a Polymorph 2.0 engine, 192 CUDA cores, 16 texture units and a lot of high-level cache. Four raster units and 128 Texture units comprise 32 ROPs; eight geometry units each have a tessellation unit, and more lower-level cache.

Nvidia has really improved their memory controller over last generation as there is a 256-bit wide GDDR5 memory interface at 6Gbps declared throughput for each of the two GPUs. An onboard PLX bridge chip provides independent PCI Express 3.0 x16 access to both GPUs for maximum multi-GPU throughput.

The memory subsystem of the GeForce GTX 670 is identical to the GTX 680, consisting of four 64-bit memory controllers (256-bit) with 2GB of GDDR5 memory.  GeForce GTX 690’s memory speed is 6008MHz data rate.  The base clock speed of the GeForce GTX 670 is 915MHz. The typical Boost Clock speed is 980MHz.

The GeForce GTX 680 reference board measures 11″ in length whereas the GTX 670 is 9.5″. Display outputs include two dual-link DVIs, HDMI and one mini-DisplayPort connector. Two 6-pin PCIe power connectors are required for its operation.

This is a very brief overview of Kepler architecture as presented to the press at Kepler Editor’s Day in San Francisco a few weeks ago. When we attend Nvidia’s upcoming GPU Technology Conference (GTC) in less than two weeks time, you can expect a lot more details about the architecture.

GPU Boost

GPU Boost was invented by Nvidia to improve efficiency and to raise the GTX 690 clocks automatically in response to dynamically changing power requirements. Up until now, Nvidia engineers had to select clock speeds on a specific “worst case” power target – often a benchmark.  Unfortunately, all apps are not equal in their power requirements and some applications are far more power-hungry than others. That means that in some games with lower power requirements, the game is not optimized for higher core frequency because it is limited by a global power target. With GPU Boost, there is real time dynamic clocking with polling every millisecond. In this way, clocks can be ramped up to meet the power target of each application – not held back by the most stressful application, which is usually a benchmark, not a game.

As we found with the GTX 680, GPU Boost goes hand-in-hand with overclocking and it delivers additional frequency in addition to the clocks set by the end user. GPU Boost continues to work while overclocking to the maximum allowed by the ever-changing power envelope.

Moving the voltage higher also moves the frequency and boost higher. In practice, if you monitor the frequencies, they constantly change up and down.

Adaptive VSync

Traditional VSync is great for eliminating tearing until the frame rate drops below the target – then there is a severe drop from usually 60 fps down to 30 fps if it cannot meet exactly 60. When that happens, there is a noticeable stutter.

Nvidia’s solution is to dynamically adjust VSync – to turn it on and off instantaneously. In this way VSync continues to prevent tearing but when it drops below 60 fps, it shuts off VSync to reduce stuttering instead of drastically dropping frame rates from 60 to 30 fps or even lower. When the minimum target is again met, VSync kicks back in. In gaming, you never notice Adaptive VSync is happening; you just notice less stutter (especially in demanding games).

Adaptive VSync is a good solution that works well in practice. We spent more time with Adaptive VSync in playing games and it is very helpful although we never use it when benching.

 

FXAA & TXAA

TXAA

There is a need for new kinds of anti-aliasing as many of the modern engines use differed lighting which suffers a heavy performance penalty when traditional MSAA is applied. The alternative, to have jaggies is unacceptable. TXAA – Temporal Anti-Aliasing is a mix of hardware mult-sampling with a custom high quality AA resolve that use temporal components (samples that are gathered over micro-seconds are compared to give a better AA solution).

There is TXAA 1 which extracts a performance cost similar to 2xAA which under ideal circumstances give similar results to 8xMSAA. Of course, from what little time we have spent with it, it appears to be not quite as consistent as MSAA but works well in areas of high contrast. TXAA 2 is supposed to have a similar performance penalty to 4xMSAA but with higher quality than 8xMSAA.

TXAA will be the subject of an IQ analysis in a forthcoming article and we are told that we shall see games that support it natively this year. For now, it appears to be a great option for the situations where MSAA doesn’t work efficiently.

FXAA

Nvidia has already implemented FXAA – Fast Approximate Anti-Aliasing. In practice, it works well in some games (Duke Nukem Forever), while in other games text may be a bit blurry. FXAA is a great option to have when MSAA kills performance. We plan to devote a entire evaluation to comparing IQ between the HD 7000 series and the GTX 600 series as well as comparisons with the older series video cards.

 

Specifications

Here are the specifications for the GTX 680:

Here are the specifications of the GTX 670:

We see everything is very similar to the GTX 680.  The GeForce GTX 670 was designed from the ground up to deliver exceptional tessellation performance which Nvidia claims is about 8 times the HD 7950’s tessellation performance. Tessellation is the key component of Microsoft’s DirectX 11 development platform for PC games.

Tessellation allows game developers to take advantage of both GeForce GTX 670 GPU’s tessellation ability to increase the geometric complexity of models and characters to deliver far more realistic and visually rich gaming environments. Needless to say, the new GTX 670 brings a lot of features to the table that current Nvidia’s customers will appreciate, including improved CUDA’s PhysX, 2D and 3D Surround plus the ability to drive up to 3 LCDs plus a 4th accessory display from a single GTX 670 ; superb tessellation capabilities and a really fast and power efficient GPU in comparison to their previous GTX 570.

Surround plus an Accessory display from a single card

One of the criticisms that Kepler has addressed from Fermi was that two video cards in SLI are required to run 3-panel Surround or 3D Vision Surround. From a single card, the GTX 670, 680 and the GTX 690 now can run three displays plus an accessory display.  Interestingly, Nvidia has changed their taskbar from the left side to the center screen. We now prefer the taskbar in the center; it might be more convenient for some users rather than clicking all the way over to the left for the start menu as with Eyefinity.

One thing that we did notice. Suround and 3D Vision Surround are now just as easy to configure as AMD’s Eyefinity. And AMD has no real answer to 3D Vision or 3D Vision Surround – HD3D lacks basic support in comparison.

One new option with the GTX 670/680/690 is in the bezel corrections. In the past, the in-game menus would get occluded by the bezels and it was annoying if you use the correction. Now with Bezel Peek, you can use hotkeys to instantly see the menus hidden by the bezel. However, this editor does not ever use bezel correction in gaming.

One thing that we did note – Surround suffers from less tearing than Eyefinity.  The only true solution to tearing in Eyefinity is to have all native DisplayPort displays or opt for the much more expensive active adapters.  And you will need two adapters for Eyefinity for most HD 7970s to run Eyefinity, whereas you only need one for Surround and the GTX 670 and the GTX 680.

Nvidia claims a faster experience with the custom resolutions because of a faster center display acceleration. Of course, we tested Surround’s 5760×1080 resolution and even 3D Vision Surround. Check out the results in the Performance Summary chart and the 3D Vision/Surround section.

A look at the GTX 670

Nvidia has redesigned their GEFORCE logo and the GTX 670 is on a short PCB compared to the GTX 680.  With the GeForce GTX 670, Nvidia’s board partners have the option to produce custom GTX 670 boards on launch day. To get the GeForce GTX 670 into smaller form factor chassis, Nvidia made a number of adjustments to the reference board to save space by moving the GTX 670’s power supply much closer to GPU.

With the GTX 670’s power circuitry moved to the other side of the board, the area on the right side of the PCB was removed to save board space.  The same cooling fan used on the GeForce GTX 680 is adapted for the GTX 670 and it is fitted with acoustic dampening material to minimize unwanted tones in the fan noise.  It is a pretty quiet card although not as quiet as the GTX 680.

The GTX 670’s blower fan exhausts hot air from the GPU outside the system chassis helping to reduce temperature inside the PC. This feature is particularly useful for small form factor PCs including possibilities for home theater PC (HTPC).

.

Here you can see the GTX 670 with its cover removed.

And here is the bare PCB.

 

SLI

The GTX 670 is set up for SLI by using two GTX 570s. We hope to bring you a follow-up evaluation comparing GTX 670 SLI performance to the GTX 690.

The specifications look extraordinary with solid improvements over the Fermi-based GTX 570. Let’s check out performance after we look at our test configuration on the next page.

Test Configuration – Hardware

• Intel Core i7-3770K reference 3.50 GHz/Turbo to 3.9GHz, overclocked to 4.6 GHz; Turbo is off and HyperThreading is on.
• Gigabyte Z77MX-D3H (latest BIOS, USB/PCIe 3.0 specification; CrossFire/SLI 8x+8x).
• 4 GB OCZ DDR3 PC 1800 Kingston RAM (2×2 GB, tri-channel at 1200MHz; supplied by Kingston)
• GeForce GTX 670, 2 GB (base clocks of 915/3000MHz and also overclocked, +150/+400MHz), supplied by Nvidia
• GeForce GTX 680, 3 GB reference clocks, supplied by Nvidia.
• GeForce GTX 570, 1.25 GB reference design and clocks, supplied by Nvidia
• GeForce GTX 580, 1.5GHZ reference design and clocks, supplied by Nvidia
• PowerColor Radeon HD 7970, 3 GB with custom cooling at stock clocks (925/1375MHz)
• AMD Radeon HD 6970, 2GB reference design and stock clocked, supplied by AMD
• Onboard Realtek Audio
• 2 x 240 GB Kingston HyperX SSDs; one for AMD and one for Nvidia, supplied by Kingston
• Thermaltake ToughPowerXT 775W power supply unit supplied by Thermaltake
• Cooler Master Elite mid-Tower case, supplied by Cooler Master
• Noctua NH-DH14 CPU cooler, supplied by Noctua
• Philips DVD SATA writer
• HP LP3065 2560×1600 thirty inch LCD.
• Three ASUS VG236 23-inch 1920×1080 120Hz LCDs supplied by ASUS/Nvidia and used for Surround/Eyefinity 5760×1080 resolution.
• Asus VG278 27″- 120Hz 1080p display and 3D Vision 2 Glasses supplied by Nvidia/ASUS.

Test Configuration – Software

• Nvidia GeForce 301.24 Beta drivers for GTX 570 and GTX 580; for the GTX 680/670, 301.33 release drivers. High Quality
• AMD 12.4 WHQL Catalyst drivers; latest CAPs. High Quality – optimizations off; use application settings
• Windows 7 64-bit; very latest updates
• Latest DirectX
• All games are patched to their latest versions.
• VSync is off in the control panel.
• AA enabled as noted in games; all in-game settings are specified with 16xAF always applied; 16xAF forced in control panel for Crysis.
• All results show average, minimum and maximum frame rates except as noted.
• Highest quality sound (stereo) used in all games.
• Windows 7 64, all DX10 titles were run under DX10 render paths; DX11 titles under DX11 render paths.

The Benchmarks

Synthetic

• Vantage
• 3DMark 11
• Heaven 3.0

• Left 4 Dead 2
• Serious Sam 3 BFE

• Crysis
• Far Cry 2
• Just Cause 2
• World-in-Conflict

• BattleForge
• Alien vs. Predator
• STALKER, Call of Pripyat
• Metro 2033
• F1 2010
• H.A.W.X. 2
• Lost Planet 2
• Civilization V
• Crysis 2
• Dirt 3
• Deus Ex: Human Revolution
• Batman: Arkham City

Before we get to the performance charts, let’s look at overclocking, power draw and temperatures.

 

Overclocking, Power Draw and Temperatures

Overclocking the GTX 670 is just as easy as overclocking the GTX 680 and 690. What is surprising is that we matched the GTX 680 overclock on the core exactly, +150MHz, although we got only +400MHz on the memory compared to +550MHz on the GTX 680. We did not adjust the GTX 670’s voltage. Temperatures were never an issue and the fan profile remained at stock which meant that the GTX 670 is also extraordinarily quiet at maximum load!

We decided to compare our GTX 670 power draw with that of the HD 7970 with both fully overclocked and in the same identical system; the only difference being the video cards. Both systems idled below 100W testifying to these cards’s excellent power management at idle and Ivy Bridge’s extreme efficiency compared to our older Bloomfield i7-920 system.

Here is the power draw of the entire system with a GTX 670 overclocked as far as we could push it and under full load but with no voltage adjustments:

This is very respectable for a video card card of this incredibly high performance level! Now let’s look at an overclocked HD 7970 in the same system under the exact same conditions; again, no overvolting is involved.

(The original image wrongly pictured the overvolted HD 7970; this one is overclocked but not overvolted)

Radical. Kepler is extraordinarily power efficient – much more so than Tahiti.  It will be interesting to see how the new editions of the HD 7970 will improve on power draw as their clocks are increased by +75MHz to 1000MHz in the upcoming “GHz” editions which replace the reference versions that we are testing with and is current now.

Let’s head to the performance charts and graphs to see how the GTX 670 compares with the last generation GTX 570 and GTX 580 as well as against the GTX 680 and against the AMD HD 7970 and HD 6970.

Performance summary charts & graphs

Here are the summary charts of 18 games and 3 synthetic tests. The highest settings are always chosen and it is DX11 when there is a choice; DX10 is picked above DX9, and the settings are ultra or maxed. Specific settings are listed on the Main Performance chart at the beginning of this page. The benches are run at 1920×1200 and 2560×1600 with separate charts devoted to overclocking, PhysX, 3D Vision, 5760×1080 Surround (including 3D Vision Surround), as well as dividing games up into easy to read charts by their DX pathway and by resolution.

Main Overall Summary chart

In the first three columns of the main performance summary chart, the GTX 670 is tested at stock and overclocked with a single stock HD 7970 in  between for comparison; next is a single GTX 680 and then the GTX 580 is next and then is the 3GB HD 7970 and finally the the GTX 570. This is the master chart and although it has not been made into a graph as there is too much information to put onto a single graph, there are many other charts and sub graphs that are based on it.

 

All results, except for Vantage and 3DMark11, show average framerates and higher is always better. In-game settings are fully maxed out and they are identically high or ultra across all platforms. As usual, we begin with the synthetics.

Futuremark & Heaven synthetic tests

3DMark11 is Futuremark’s latest DX11-only benchmark and Vantage is DX10. Unfortunately, scores are completely meaningless when they are presented in this way but they do offer supporting data to accompany our game benches. Here is the chart with Vantage and 3D Mark11:

The GTX 670 comes close to the GTX 680 and the HD 7970 and pulls way ahead of the other competing cards; and when it is overclocked +150/+4000MHz, it is a beast! Heaven 3.0 is a very demanding benchmark and here it is expressed in a chart.

Again, synthetic tests are interesting but they are not necessarily indicative of real world gaming performance. In all three cases, the overclocked GTX 670 “wins” over everything else except for the GTX 680 by a fair margin. Next up, let’s look at DX9 games.

DX9 Games

We test the popular Source Engine represented by Left 4 Dead 2 and also a demanding DX9 game, Serious Sam 3, BFE with both at completely maxed out settings. First up is 2560×1600:

 

Now we see 1920×1200

 

Both Left 4 Dead 2 and Serious Sam 3 BFE are faster on the overclocked GTX 670 than any of the other cards although when both are at stock, the HD 7970 pulls ahead.  The GTX 670 improves strongly over the GTX 570 and even the GTX 580.  Let’s check out DX10 games

DX10 Games

We test four DX10 games – Just Cause 3, Far Cry 2, Crysis and World in Conflict, Soviet Assault. Here is 2560×1600 resolution.

Now at 1920×1200:

Out of these four DX10 games, the GTX 670 goes toe to toe with the HD 7970, beating it in Just Cause 2 and in Far Cry 2 and trading blows in Crysis and World in Conflict, depending on the resolution.

DX11 Games

Most of our testing emphasizes DX11 games and we bench 12. Since the charts get too long, we break them up into charts of 6 games each.

First up are the older DX11 games at 2560×1600

In most of these games, the overclocked GTX 670 beats the stock HD 7970 although the Radeon is faster when both are at stock clocks.  Now those same games at 1920×1200:

Although the gap narrows at lower resolution, the HD 7970 is faster when clocks are stock.  Now the newer DX11 games at 2560×1600:

This time, the GTX 670 at stock clocks is generally faster than the HD 7970 except for Deus Ex and Batman.  Now the same DX11 games at 1920×1200:

Again, at the lower resolution, the GTX 670 nearly catches up to the HD 7970 in Batman and it narrows the gap in Deus Ex, and it pulls further away in the other 4 games. .

Super-Widescreen 5760 x1080, Surround, 3D Vision Surround, and PhysX

Here is the main chart that gives the details for the tests:

Let’s look at 3D Vision at 120Hz versus the same settings with 3D Vision at the popular 1920×1080 resolution:

Here we see some strange results with 3D Vision as the drivers do not appear to lock Kepler to 60Hz as they did with Fermi and you can see the exact performance penalty that 3D Vision requires per each game.

Surround vs Eyefinity

Now the GTX 670 benchmarked in 3-panel Surround with slightly lesser settings (see the main chart above).

This time, the GTX 670 is stronger in Batman (partly due to using PhysX on ‘normal’), Just Cause 2 and in Far Cry 2 while the HD 7970 is stronger in the other 4 games.

Eyefinity has more noticeable tearing in some games on one side screen and sometimes in both and VSync is not applied because the framerates are usually below 60 at 5760×1080. Tearing is more noticeable on AMD Eyefinity than it is on Nvidia’s Surround, and it is especially noticeable with AMD’s more recent drivers.

For 3D Vision and for Surround, several  games need to have their settings reduced. Just remember that you are playing across three screens and and are also rendering each scene twice for 3D Vision!!

One thing that we found really strange and it may be a driver bug with Kepler – the frame rates appear to be no longer locked to 60Hz in the 3D Vision drivers as we previously tested. Previously, frame rates would be capped at 30 or 60 fps if possible. However, now it is very convenient to see exactly what performance penalty 3D Vision takes – sometimes it is more than 50 percent; other times less.

Next up, let’s look at PhysX

PhysX

We test PhysX in two games. Batman: Arkham City makes great use of PhyxX and it is a real shame to play the game without it. In both cases, turning on PhysX, although affecting the frame rate, it is enough to play the game with fully maxed out details and FXAA or AAA with our GTX 670.  

Let’s check out Overclocking:

Overclocking

We overclocked our GTX 670 +150MHz on the core and +400MHz on the memory. This is a very good overclock on stock voltage and stock fan profile and it falls only a bit short of the overclock on our GTX 680 of +150/+550MHz.

Here are all of our games compared at 2560×1600 – stock versus overclocked:

 Here is 1920×1200

As you can see the GTX 670 scales extremely well and in every case, the overclocked GTX 670 is able to beat the GTX 680, although the GTX 680 can itself be overclocked a bit further than our GTX 670.  The GTX 680 also appears to be a stronger performer at the highest resolutions, so as to justify it’s $100 higher price tag.

The GTX 670 performs very close to its big brother the GTX 680, significantly closer than 10%; and it comes very close in performance to the stock HD 7970, matching or beating it in some games but losing to it in a few more in our benching suite.  However, the two cards are within a percent or two of each other and are very evenly matched so as to provide near-identical performance, just depending on which game is being played or benched.  Let’s head for our conclusion.

 

Conclusion

This has been quite an enjoyable, if far too short, 1-week exploration for us in evaluating our new GTX 670. It did extraordinarily well performance-wise comparing it to the the GTX 580 and GTX 570 where it brings higher performance and better efficiency than the older cards.  We are totally impressed with this cool-running “value” enthusiast card which is able to go nearly head to head against the more expensive HD 7970 and the GTX 680.

We see good overclockability with extreme quietness at stock voltage and fan profile even when the GTX 670 is highly overclocked.  It uses very little power compared to the HD 7970 and the difference widens as the overclock goes up. For its $399 price, it gives more than 90% of the performance of the GTX 680 which is priced at $500 and the 570 comes right into the $480 HD 7970’s performance territory.   For the second time, the reference HD 7970’s $480 price needs to be adjusted downward.  In this way, the GTX 570 is a game changer.

Pros

• The GTX 670 provides excellent performance for a $399 card and makes the HD 7970 again overpriced.
• TDP and power draw is superb. Performance per watt is better than its competitor’s flagship 28nm HD 7970 (to say nothing about beating the last generation) and it is quiet in comparison to the older cards.
• Overclockability is excellent – GPU Boost works as advertised.
• The reference design cooling is quiet and efficient; the card and well-ventilated case stay cool even well-overclocked.
• It is possible to use two of these cards for SLI performance without needing a massive PSU
• 3D Vision 2 and PhysX enhance gaming immersion and both are improved using the GTX 670 compared to the last generation.
• Surround and 3D Vision Surround plus an accessory display can now be driven off of a single GTX 670 by requiring a single passive adapter;  HD 7970 requires two, and tearing is noticeably better with Surround, compared to Eyefinity
• New AA allows for high performance without jaggies in deferred shading lighting engines
• Adaptive VSync reduces stuttering while retaining the advantages of minimizing tearing.

Cons

• None

 

The Verdict:

• If you are buying a powerful video card right now and looking for the highest performance as well as bang for buck, the GTX 670 is an easy choice nearly matching the HD 7970 performance for about $80 less. We feel it deserves ABT’s Great Value award.

 

We do not know what the future will bring, but the GTX 670 brings good value and a great performer to the GeForce family. With great features like PhysX and the second generation of 3D Vision, you can be assured of immersive gaming by picking this card for 1920×1080, 2560×1600 or even higher resolutions, including for 3-panel Surround.

If you currently game on an older generation video card, you will do yourself a big favor by upgrading. The move to a GTX 670 will give you better visuals on the DX11.1 pathway especially over DX10, and you are no doubt thinking of SLI if you want to get nearly the ultimate in gaming performance without breaking the bank with a GTX 690. We expect that many enthusiasts will – like us – upgrade to Intel’s Ivy Bridge – and this is an excellent video card to compliment their fastest processors.

The competition is hot and AMD offers their own set of features including Eyefinity and HD3D.  We are interested to see if there will be any further downward adjustments in AMD’s pricing and how they will counter this new GGTX 670 release.

Stay tuned, there is a lot coming from us at ABT. Next up is an evaluation of the flagship Noctua NH-DH14 CPU cooler which has allowed us to increase our Core i7-3770K to 4.6GHz just for this review – and we find we can even clock it higher to 4.8GHz. And you can expect more great reviews from our Mobile Tech guys to put into our new section; expect a Genius product review also this week! And don’t forget to check our forums! Our tech discussions are becoming among the best to be found anywhere!!

Mark Poppin

ABT Senior Editor

Please join us in our Forums

Become a Fan on Facebook

Follow us on Twitter

For the latest updates from ABT, please join our RSS News Feed

Join our Distributed Computing teams

• Folding@Home – Team AlienBabelTech – 164304
• SETI@Home – Team AlienBabelTech – 138705
• World Community Grid – Team AlienBabelTech

«1. Introduction2. Architecture recap & specifications3. Test Configuration, Power Draw & Overclocking4. Performance Summary Charts5. The Conclusion6. View All»