Tuesday, August 3, 2010

Two GeForce GTX 460s Beat One GTX 480?

Two GeForce GTX 460s Beat One GTX 480?

The GeForce GTX 460 has already proven itself an excellent value as a single card, but can two of them offer superior performance at similar cost to Nvidia’s flagship GTX 480? Let's just say that there's a good reason to buy an SLI-compatible motherboard.
Graphics giants Nvidia and AMD continuously run up against the limits of manufacturing technology in their effort to release revolutionary products every eighteen months.
Note to AMD enthusiasts: The focus of today’s article is on GeForce GTX 460 SLI performance relative to the GeForce GTX 480. For a detailed analysis of how the GeForce GTX 480 compares to multi-GPU Radeon configurations, including the Radeon HD 5970 and Radeon HD 5870 in CrossFire, please consider Don Woligroski’s Asus ARES review.

Sparkle was kind enough to supply one GeForce GTX 480 and two GeForce GTX 460 graphics cards. Each card includes a mini-HDMI to standard HDMI data cable, a DVI-I to VGA adapter, driver CD with various freeware, installation guide, and two power adapter cables.

Support for up to seven graphics cards makes the X58A-UD9 this reviewer’s choice for multi-card graphics tests. Less expensive X58-based solutions would have also been adequate for two-way SLI testing, since the chipset supports two x16 cards at full bandwidth.

We brought back Intel’s Core i7-980X to satisfy popular demand, even though none of today’s games can properly access more than four CPU threads. SLI does allow some games to push a few CPU cores to their limit however, so we overclocked ours to 4.00 GHz.

Diamond’s Radeon HD 5870 provides a second reference point for anyone who would like to compare Don Woligroski’s comprehensive CrossFire analysis to today’s limited SLI tests.
Benchmark Configuration
3D Games
Aliens Vs. Predator Benchmark Alien vs Predator Benchmark Tool
Test Set 1: Highest Settings, No AA
Test Set 2: Highest Settings, 4x AA
Call of Duty: Modern Warfare 2 Campaign, Act III, Second Sun (45 sec. FRAPS)
Test Set 1: Highest Settings, No AA
Test Set 2: Highest Settings, 4x AA
Crysis Patch 1.2.1, DirectX 10, 64-bit executable, benchmark tool
Test Set 1: Highest Quality, No AA
Test Set 2: Highest Quality, 4x AA
DiRT 2 Run with -benchmark example_benchmark.xml
Test Set 1: Highest Settings, No AA
Test Set 2: Highest Settings, 4x AA
S.T.A.L.K.E.R.: Call Of Pripyat Call Of Pripyat Benchmark version
Test Set 1: Highest Settings, No AA
Test Set 2: Highest Settings, 4x MSAA
Synthetic Benchmarks and Settings
3DMark Vantage Version: 1.0.1, GPU and CPU scores

3DMark adds a PhysX Processing Unit score by default, invalidating any test results where ATI cards have been included. We disabled it for the following tests.

Two GeForce GTX 460 graphics cards put a beating on the GeForce GTX 480 at the default 1280x1024 resolution, a place where the Radeon 5870 also does well. Higher resolution tests will show just how unimportant the default resolution should be to high-end gamers, though.

Maintaining the same chart scale for all resolutions gives a visual representation of how far performance really drops as settings are increased. Two GeForce GTX 460 graphics cards continue to show up the single GTX 480 at 1680x1050.

The GeForce GTX 460’s lead begins to fall off a little at 1920x1200. Even still, the difference between two GeForce GTX 460s and a single GTX 480 is almost as big as the difference between a GeForce GTX 480 and a single GTX 460. The real question is whether this synthetic's results will be repeated in real-world games?

Any GeForce GTX 480 buyers who thought solid performance might be limited to DX11-based games will be shocked to see the GeForce GTX 460 SLI configuration leaping out of the gate in MW2, tearing a 25% lead away from the GeForce GTX 480.

Can SLI scaling ever reach 100%? It can with the GeForce GTX 460, but only at MW2’s highest settings. It’s at this point that we really must remind readers that this particular game is completely playable at all settings, even using a single GeForce GTX 460.

The GeForce GTX 460 SLI solution continues its rampage in Crysis, beginning with an unbelievable 95% improvement over a single card when using medium resolutions at the game’s highest settings. Equally unfathomable is its 27% lead over the single GeForce GTX 480.

Yes, it can play Crysis, even at the 1920x1080 pixel resolution native to mid-budget monitors, and even with all the eye-candy enabled. Adding AA to the GeForce GTX 480, however, causes it to drop below our target 40 FPS average frame rate.

The AA bug we normally see in SLI systems at Crysis’ highest settings appears to be gone completely using Nvidia’s latest drivers with the GeForce GTX 460. We even retested at 8x to make sure the bug was gone, and confirmed that it is no longer a problem. Nobody will be playing this game at these low frame rates, but at least the performance scaling is appropriate.

With the GeForce GTX 460 SLI configuration’s pattern of performance domination clearly established, further superlatives add nothing to its top score in Call of Pripyat benchmark.

We finally get to the point where the GeForce GTX 460’s performance is actually needed by increasing this benchmark’s resolution to 2560x1600.

It’s not always easy to keep track of which card wins and loses in individual benchmarks, or by how much. While today’s performance numbers were far more consistent than those of most graphics reviews, we thought there might be a few readers who could still use a breakdown of average performance.

At 1680x1050, the GeForce GTX 460 SLI configuration outpaces the GTX 480 by a wide enough margin that it wins even with anti-aliasing enabled.

1920x1080 results give the GeForce GTX 480 its second beating, yet even the single GeForce GTX 460 produces adequate frame rates, on average.

If you can afford a 2560x1600 display, you’ll want nothing less than a GeForce GTX 460 SLI configuration. The average doesn’t show how it stumbled in a few benchmarks, but it absolutely slams the single-card GeForce GTX 480. Of course, we have a pair of GeForce GTX 480s to show off in this quarter's System Builder Marathon, and that is a performance segment unto itself.

Power consumption numbers are measured for the entire system by a meter gauging use from the wall outlet. Our OCZ-Z1000 power supply is 90% efficient (±1%) at the “Full GPU Load” power consumption levels charted below. So, anyone who wants to know the actual consumption of system components need only multiply the “Full GPU Load” readings by 0.9.

Efficiency is a comparison of work done to energy used. There is no metric to determine 100% efficiency using bits-per-watt or even FPS-per-watt, since bits and rendered frames are intangible. The best we can hope for is a comparison of how much more or less efficient one system is compared to another. To do that, we must first see how much more energy each system used than the lowest-consuming system, which in this case is the single GeForce GTX 460-based configuration.

Next, we need to know how much more performance the three top systems have over the fourth. We averaged the frames per second of each system and divided the result by that of the lowest system.

Dividing the performance difference (above) by the power consumption difference (two charts above) gives us a “full load” performance-per-watt difference, with the lowest system representing 100%. On the other hand, since the work produced by idle systems is invariable, comparing the efficiency of idle systems requires simply dividing each configuration’s idle power by that of the most miserly configuration. Because there is no such thing as 100%-efficient, subtracting the 100% baseline from each result allows us to show the difference between configurations.

Adding a second GeForce GTX 460 gives us a 90% performance boost, while requiring only 53% more total system power, yielding a performance-per-watt increase of 24% (1.90 / 1.53). The same system, at idle, consumes 13.1% more power than the single GeForce GTX 460 baseline.

And now for heat:

The temperature numbers above are for the first card in the system. In our SLI configuration, with two cards spaced less than one expansion slot apart, the bottom card actually runs cooler than a single card. The is likely due to the system increasing the speed of both fans simultaneously. Our motherboard also supports spreading the cards several spaces apart, and a test with two spaces between cards revealed identical temperatures in SLI and single-card configurations.

With an MSRP of $250 and a Web price around $230, GeForce GTX 460 1GB graphics cards were already known to be great mid-priced performers. Doubling these up in SLI gives us the same MSRP and Web price as a single GeForce GTX 480. But does this give us more performance? You bet!

The GeForce GTX 460 SLI configuration absolutely obliterates the GeForce GTX 480’s performance scores, landing a 26% performance coup de grace upon its big brother after continuously battering it with wins in every benchmark at every setting. That would put it in the same performance class as a certain $700 dual-GPU card, according to Don Woligroski’s recent review. But—at less than $500—it doesn't even need to compete there. While far-more-expensive solutions do have their own particular strengths, what we really wanted to know was where our $460 would be best spent, and today’s test revealed that answer.

Because SLI scales so well, giving two GeForce GTX 460 1GB graphics cards a 90% performance boost over a single card, the value for two cards together is very similar to what earned the single GeForce GTX 460 1GB its previous Recommended Buy award. Today’s award applies in the same way to the $500 graphics category, where the only reason we can think of to purchase a GeForce GTX 480 right now is to use it in SLI for the $1000 price class.

While today’s award applies to all GeForce GTX 460 1GB models priced between $230 and $250, we’d still like to express our special thanks to Sparkle Computer Corp. for supplying the units we used to determine today’s results.

No comments:

Post a Comment