The GTX 680 Arrives! – Part Two, Maximum Overclocking vs. the HD 7970

Nvidia has released its long awaited GeForce GTX 680 based on its brand new Kepler DX11.1 GK104 architecture two weeks ago.  On release day, we brought you Part One of our GTX 680 series, Nvidia’s GTX 680 arrives!  “Faster, Smoother, Richer” – is it enough to take the Performance Crown?    We found that the GTX 680 beats the HD 7970 in every metric including performance – and it is priced $50 less.

We now have had two more weeks hand’s on experience testing the GTX 680 versus the HD 7970 and we have learned quite a bit more that we would like to share with you.  In Part One, we learned that the GTX 680 does indeed overclock.   We were able to get it stably up from its stock base clocks (1006/3004MHz) to a decent overclock (1156/3029MHz) which compared favorably with our overclock of the HD from stock (925/1375MHz) to overclocked (1100/1400MHz).  However, we did not adjust the voltage nor the fan speed,  nor did we find our maximum overclock for each card’s memory.

For Part Two, we will compare the performance of our overclocked GTX 680 to our overclocked HD 7970 as a starting point as we overclock each of them to the edge of stability by maxing out the voltage and fan speed to give you an idea of what future Super Overclocked versions of each of these cards may be like and how they may related to each other in terms of performance as faster and faster overclocked versions of each card appear.

In Part One, we compared the relative performance of the GTX 680, GTX 580 and GTX 690, as well as the HD 7970, HD 6970 and HD 6990.  The dual-GPU cards of the last generation were generally faster although overclocked, the GTX 680 and the HD 7970 approached their performance in some games.  The HD 6970 and to a lesser extent, the GTX 580 were left in the dust.  This time, we will only be comparing the  HD 7970 to the GTX 680 and we shall overclock them each as far as we can.  This time we shall adjust the voltage as far as it is useful and as far as our overclocking tools – MSI’s Afterburner for Radeon and EVGA’s Precision X for GeForce – will allow us.

We are also going to revisit the “Power Usage” section as befits overclocking in this review.  We noticed that the HD 7970 drew about 44W more than the GTX 680 when both were overclocked without any voltage adjustment and we look forward to seeing what happens to each cards’ power requirements when the clocks go up with additional voltage.

AMD has already launched most of its 28nm DX11.1 series – $550 for their flagship single-GPU HD 7970  – as well as their HD 7950, HD 7870 and HD 7850, down to their entry-level enthusiast HD 7770 and HD 7750 cards. To compete, Nvidia has launched its new Kepler architecture 28nm DX11.1 GeForce lineup with its GTX 680 flagship heralding the rest of the series as forthcoming. The reference GTX 680 is an overall better value than the reference HD 7970 as it comes with a MSRP of $499 which is $50 less than the slower HD 7970. However, we still need to answer the question: Is an overclocked HD 7970 worth the $50 premium over the GTX 680?

That question is important because we expect that Nvidia will shortly launch it’s own entire 28nm DX11.1 line-up based on their GK104 “Kepler” architecture.  We expect to see the GTX 670 and GTX 660 as well as the GTX 550 launch within a few months and we need to see what this new 28nm Kepler architecture brings over their GF110/100 series besides DX11.1 and a smaller process. Overclocking our GTX 680 to its maximum may give us some idea of future GK104 core scalability as Nvidia refines their process.  And we shall get an idea if AMD’s “Tahiti” architecture will be able to scale enough for highly overclocked versions of the HD 7970 to catch up to the GTX 680.  AMD is loathe to drop pricing on the HD 7970 as it would cause a cascade effect on their entire line up which has been premium-priced over the last generation.

To properly bring you this review, we purchased a PowerColor HD 7970 from NewEgg and put it through its paces with the very latest AMD performance drivers – Catalyst 12-3 which is even newer that the set we tested for the GTX 680 launch.  We are also using the very latest WHQL GeForce drivers 300.10 which is an upgrade over the launch drivers but appears to have no performance differences in our games.

Today you will see us pit our PowerColor HD 7970, which is now further overclocked from 1100/1400MHz to 1200/1575MHz, against the new GTX 680 which is further overclocked from 1156/3029MHz to 1181/3604MHz.   Sometimes it is a bit confusing when talking about the GTX 680’s “stock” clocks because there is a dynamic GPU Boost which further automatically increases the clocks by +50MHz to +100MHz over the ‘base clock’ of 1006/3004MHz  There is another way to look at it; we ultimately boosted the GTX 680’s clocks by +175MHz on the core and +600MHz on the memory while we boosted the HD 7970’s clocks by +275MHz on the core and +200MHz (+400MHz, the way Nvidia reports it) on the memory clocks.

We continue to benchmark with 21 modern games and with 3 synthetic benchmark at 1920×1200 and 2560×1600 resolutions and with details fully maxed and with 4x or 8xAA/16xAF applied whenever possible.

 

Is the overclocked HD 7970 worth $50 more than its rival, Nvidia’s GTX 680?

We declared the GTX 680 the performance winner and could find nothing compelling to recommend the HD 7970 at fifty dollars more, two weeks ago.  We believe that AMD is immediately going to respond to Nvidia’s GTX 680 launch by allowing their partners to overclock the current HD 7970 and we expect to see super-overclocked “halo” cards, appear at the tech sites so as to give the impression that the HD 7970 is not so far behind the GTX 680.  One of these such cards is the Sapphire “Toxic” HD 7970 and it is expected to be clocked at 1150/1500MHz.  Since our own PowerColor HD 7970 can supass this overclock (1200/1575MHz), we believe that we have a very fair representation of what is coming in the near future for our evaluation today.

Widespread e-tail availability of the GeForce GTX 680 has not happened yet. So one still has a little time to decide what to do and this review is designed to help with an important potential upgrade.  Since we do not want any chance of our CPU “bottlenecking” our graphics, we continue testing both of our graphics cards with our Intel Core i7-920 at 4.20GHz (approximately the same performance as a mildly overclocked Core i7-2700K), 6 GB Kingston DDR3 using a Gigabyte X58 16x + 16x PCIe CrossFire/SLI motherboard.

Before we take a look at overclocking and power draw, let’s check out the testing platform.

 

Test Configuration – Hardware

• Intel Core i7 920 reference 2.66 GHz, overclocked to 4.2 GHz using the 21x multiplier; Turbo and HyperThreading are on.
• Gigabyte GA-MA790X-UD4P (latest BIOS, PCIe 2.0 specification; CrossFire/SLI 16x+16x).
• 6 GB OCZ DDR3 PC 1800 Kingston RAM (3×2 GB, tri-channel at 1200MHz; 2×2 GB supplied by Kingston)
• GeForce GTX 680, 2 GB (1006/3000MHz and also overclocked), supplied by Nvidia
• PowerColor Radeon HD 7970, 3 GB with custom cooling at stock clocks (925/1375MHz) and also overclocked.
• Onboard Realtek Audio
• 2 x 500 GB Seagate Barracuda 7200.12 hard drives; one for AMD and one for Nvidia
  
• Thermaltake ToughPowerXT 775W power supply unit supplied by Thermaltake
• Thermaltake Element G Case supplied by Thermaltake
• Frio OCK CPU cooler, supplied by Thermaltake
• Philips DVD SATA writer
• HP LP3065 2560×1600 thirty inch LCD.

Test Configuration – Software

• Nvidia GeForce 300.10 WHQL drivers for the GTX 680; High Quality
• AMD 12.3 WHQL Catalyst drivers; 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 forced 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 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

• Wolfenstein

• Left 4 Dead 2
• Serious Sam 3 BFE

• Crysis
• Far Cry 2
• Just Cause 2
• World-in-Conflict
• Resident Evil 5

• BattleForge
• Alien vs. Predator
• STALKER, Call of Pripyat
• Metro 2033
• F1 2010
• H.A.W.X. 2
• Lost Planet 2
• Civilization V
• Total War: Shogun II
• 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 & Temperatures

In our charts we use the memory clocks as Nvidia and AMD each express them.  You can multiply the AMD clocks by 2X to get the equivalent data rates if you wish; or cut the Nvidia clocks in half.  At any rate, Nvidia has actually clocked its card’s memory much higher than AMD’s.

Here are the settings (base clocks before GPU Boost are used for the GTX 680):

 

GTX 680 (1006/3004MHz  stock)

1156/3029MHz   >>>  1181/3579MHz) – 1181/3604MHz

 

HD 7970 (925/1375MHz stock)

1100/1400MHz  >>> 1125/1575MHz  >>> 1200/1575MHz

 

This evaluation is based on overclocking each card in three steps as far as they will go; first at stock and then with a substantial core overclock, followed by a memory overclock which is finally followed by a voltage increase.  This appears to be representative of each card’s performance  scaling by overclocking these cards.

Each of our competing video cards was overclocked as far as it could go without exceeding the core voltage or thermals so as to throttle the respective GPUs.  We got 100% stability from each of our GPUs at the noise-expense of running their cooling fans each at 85%.

We used MSI’s Afterburner to overclock our PowerColor HD 7970 and EVGA’s Precision X for our GTX 680.  Both utilities allowed for quite a bit of control over each card’s overclock and we were also able to adjust the voltage.  Overclocking the GTX 680 beyond the stock overclock was easy. No one should let the boost clocks intimidate them. Using EVGA Precision tool, simply push the sliders to the right as far as you dare and set the clocks.  We finally settled on +175MHz on the core and 600MHz on the core.

 

We did exactly the same thing with our PowerColor HD 7970. Its stock clocks are 925/1375 and we pushed them beyond the fastest AMD partner-overclocked Radeon HD 7970’s clocks to 1200MHz/1575MHz.

We always use EVGA’s Precision overclocking tool for Nvidia cards. They recently added integrated GPU voltage adjustments. In addition to the EVGA GeForce GTX 680, EVGA has also introduced a new version of its popular “Precision” overclocking software, EVGA Precision X. It allows you to fine tune your GeForce graphics card, including GPU Clock speed, GPU Voltage, Memory Clock speed and Fan speed.

 

EVGA Precision X features many new and key features:

• · Brand New GUI – Built from the ground up for the next generation of graphics accelerators.
• · Advanced Hardware Monitoring – See GPU vitals in real time.
• · In Game Monitoring – See your GPU vitals, without leaving your game.
• · Frame Rate Target – GeForce GTX 680 will adjust power/clock dynamically to match set framerate.
• · Fan Curve – Setup and advanced fan profile.

Learn More about EVGA Precision X

Temperature

Our ambient (room temperatures) were fairly warm – 74-78F – so as to approximate a warm Spring day. Be aware that we used our Thermaltake Element G case which has excellent airflow for an oversized midtower. The GTX 680 runs quite cool at even our first mild overclock, in the low to mid-50s C even under load. However, once the core speed increased, so did the temperatures until we were idling in the low 60s when we hit near the 1300MHz peak on the core from boost and temperatures would rise into the low-80sC.  In comparison, the PowerColor HD 7970 tolerated higher temperatures into the low 90sC.

You can see from the performance charts what effects increasing the core speed has on the GTX 680 – from the reference speed of 1006/3006MHz , past adding +150MHz to finally settling on +175MHz.

Power Draw

We did a rough comparison by using our entire system running identical stressful benchmarks while monitoring with Kill-A-Watt. Previously, our GTX 680 drew 195W at idle before any voltage adjustments.  Here is the GTX 680 in our system at idle when it is fully overclocked with the maximum voltage boost.  It uses ten watts more power at idle when overclocked.

Now we ran our benches continually and looked for the peak usage of power with our GTX 680 overclocked as far as it would go stably with the fan at 85% and with the maximum voltage increase. At stcock, the absolute peak power usage was a blip at 451W. Now let’s see how far the total systen power draw increases when we overclock and overvolt our GTX 680.

Our system draws 44W more with the overclocked/overvolted GTX 680 over the stock GTX 680 – the very same amount that the HD 7970 draws on the stock fan profile and stock voltage.  Now let’s crank up the voltage of the HD 7970 and fan to 85% and see how far over 495W it climbs.

The PowerColor HD 7970 when overclocked and overvolted draws 102W more than overclocked on the stock fan profile and with the stock voltage.  It also uses over 100W more than the GTX 680 in the same situation and in the same system!

 

Noise

Both cards are extraordinarily quiet for high-end flagship cards. It appears that Nvidia has especially tuned the GTX 680 to be quiet somewhat at the expense of cooling. It will be interesting to see what cooling designs their partners implement.  However, when a fan is turned up to 85% it becomes annoying.  The GTX 680 might be a little quieter than the PowerColor HD 7970 but it is subjective and neither are ideal over 60%.  We use 85% to keep the cards cool and stable only for benches.  When we play games for ourself, we move the clocks to maximum for the stock voltage and fan profile.

Let’s head to our performance charts to see how the Super Overclocked HD 7970 fairs against a highly overclocked reference GTX 680 and peer a bit into the future performance of these cards.

Performance summary charts & graphs

Here are the summary charts of 21 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 end of this page. The benches are run at 1920×1200 and 2560×1600 with separate charts devoted to dividing games by their DX pathway to keep the graphs relatively short.

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. Let’s break down our master chart into several smaller charts and graphs. First up, let’s look at OpenGL and DX9 together, then DX10 and fnally DX11 games separately.

Futuremark & Heaven synthetic tests

3DMark11 is Futuremark’s latest DX11-only benchmark and here are the default benchmarks detailed results of the GTX 680 at maximum overclock.  Notice the voltage slider is pushed all the way to the maximum at right.

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 the detailed score:

We do not graph the synthetic benchmarks but they are presented in the Main Summary Charts at the end of this section.  Synthetic tests are interesting but they are not necessarily indicative of real world gaming performance. In all three cases, the GTX 680 “wins” over the HD 7970 and overclocking makes no difference to the ranking. Next up, let’s look at DX9 and OpenGL games.

DX9 and OpenGL

Wolfenstein is our only example of OpenGL and it is listed on our chart together with DX9 games without being mentioned as OGL. Unfortunately Rage is hard to benchmark so it will have to do until Doom IV is released. Left 4 Dead 2 represents the very popular if aging Source Engine and Serious Sam 3 BF3 is a very new and demanding DX9 game released late last year on the Serious 3.5 engine that provides impressive visuals and on ultra settings and high resolution, can slow the fastest video cards.

Here is our Open GL and two DX9 games at 1920×1200:

Overclocking actually catches the GTX 680 up to the HD 7970 in Left 4 Dead 2.  And now at 256ox16oo.

Wolfenstein is much faster on the HD 7970 although it is quite playable on the GTX 680. Both Left 4 Dead 2 and Serious Sam 3 BFE are faster on the HD 7970 than the GTX 680 although there is no practical difference in the gameplay experience. Let’s check out DX10 games

DX10 Games

We test five DX10 games. Here are the results at 1920×1200

And now at 2560×1600.

Out of these five DX10 games, only World-in-Conflict is marginally faster on the HD 7970 over the GTX 680; in Crysis the situation reverses and the GTX 680 is faster. However, the other 3 games are noticeably faster on the GTX 680. At 1920×1200 the performance gap is wider than at 2560×1600.

DX11 Games

Most of our testing emphasizes DX11 games and we bench 13.  Unfortunately, a chart with 13 games is much too long and we break our DX11 titles into “older” and “newer” which is mostly based on the date the games were released.  Here are our older DX11 games, first at 1920×1200.

And now at 2560×1600:

And now the “newer” DX11 games at 1920×1200

And now the same newer DX11 games at 2560×1600

From the above 13 DX11 games, the HD 7970 is faster in 3 games – BattleForge, Aliens vs. Predator and Metro 2033. In three games, they trade blows – Call of Pripyat, F1 2010 and Deus Ex: Human Revolution, depending on the resolution. Finally, the GTX 680 is faster in seven DX11 games that we tested: Lost Planet 2, H.A.W.X. 2, Crysis 2, Civilization 5, Shogun II, DiRT 3 and Batman: Arkham City.

And when we overclock, the position doesn’t change in favor of the HD 7970 and the performance gap with the GTX 680  is rarely narrowed by more than a percent or two with overclocking.  Even at 1200MHz, the HD 7970 fails to overtake an overclocked GTX 680.  It also generally appears that the GTX 680 continues to show its strength at 1920×1200 which diminishes a bit compared to the HD 7970 as the resolution goes up.

Main Overall Summary chart

AMD vs. Nvidia Memory Clocks

In our charts we are expressing the memory clocks as Nvidia and AMD expresses them.  You can multiply the AMD clocks by 2X to get the equivalent data rates if you wish; or cut the Nvidia clocks in half.  At any rate, Nvidia has clocked its card’s memory much higher than AMD’s. One can even multiply Nvidia’s by 2x and AMD’s by 4x if you want the double data rate expressed so they are both equal.

This is the master chart and it has not been made into a graph as there would be too much information to include and view.  All of the preceding charts and graphs are based on this two-part chart.

These same two charts are now presented so the GTX 680 is pitted directly against the HD 7970 at each step of their similar corresponding overclock, from stock right up to their individual maximum clocks.

No matter how you add it up, the GTX 680 is generally around five percent faster overall than the HD 7970; a little faster at stock and a tiny bit less when both cards are overclocked to their maximums. Both cards overclock superbly so neither card has much of an advantage over the other with headroom on stock clocks and with the stock fan profile.  We note, as the clockspeeds increase, the performance scales nicely with both architectures although the Radeon narrowed the gap very slightly when both cards were overclocked.

From our testing, 1200MHz/1575MHz is not enough for a super overclocked HD 7970 to catch the overclocked reference GTX 680.   The Sapphire “Toxic” will evidently not be fast enough to take back the performance crown for AMD.  They would have to significantly increase the clocks of the HD 7970 to catch the GTX 680.  And no matter how you look at it, the GTX 680 has a real advantage by the lower amount of power it uses to achieve the same or slightly better performance.

Let’s head for our conclusion.

Conclusion

This has been quite an enjoyable now four-week exploration for us in evaluating our new GTX 680 since Kepler Editor’s Day in San Francisco. We are totally impressed with the cool-running Kepler chip that has such outstanding overclockability for a good price. It slots right above the HD 7970 and it offers more advantages than just price.

We see good overclockability with quietness at stock voltage and fan profile from even the regularly-cooled reference design GTX 680 as well as from the PowerColor custom-cooled HD 7970.  The GTX 680 is a better value overall than the HD 7970 if you are looking at performance and especially since the HD 7970 is still priced fifty dollars more than the GTX 680.  From this evaluation, we see that the GTX 680 will not take a back seat to the HD 7970 in overclocking and we do not see this changing in the immediate future.

The competition is hot and AMD offers their own set of features including Eyefinity and HD3D with the competing Radeon HD 7970. However, we still expect that AMD will be forced to drop pricing on the reference and stock-clocked HD 7970s and we also expect that they will feature more overclocked and super-overclocked versions as their process continues to mature. Of course, Nvidia’s partners will do the same and we can expect more overclocked GTX 680s.

Stay tuned, there is a lot more coming from ABT.  You can expect more great reviews from our Mobile Tech guys and you can expect a Genius product review also this week.  And don’t forget to check our forums!  Our tech discussions are outstanding.

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. Test Configuration3. O/C'ing. Power Draw & Temps4. Performance Summary Charts5. Conclusion & The Verdict6. View All»