GTX470 Performance Test Part 1: Windows XP

Introduction

After a long wait nVidia’s GF100 is here, and I decided to pick up a GTX470 as an upgrade for my GTX285 since I absolutely could not stomach the horrific noise levels from a GTX480. I have nothing against ATi, but I do prefer nVidia’s cards for gaming.

This article will conduct the tests on Windows XP, while a future article will retest everything on Windows 7 (64 bit), which I plan on installing soon.

But first, here are the specs for the two cards I’ll be comparing today:

Compared to the GTX285, we can see the GTX470 has a higher pixel fillrate and more shader performance, but it has less memory bandwidth, and seemingly less texturing performance too. In practice actual texturing performance should be better than the GTX285 due to improvements to the TMUs and caching. This was covered in our architectural analysis of the GF100, which I co-wrote.

With regards to estimated die sizes, the GTX470 is bigger than the already big GTX285, so it’s truly a beast. I think it’s safe to say that at least some of the delay in getting the GF100 to the market was because of manufacturing difficulties due to the chip’s extraordinary size and complexity.

Enough theory; let’s talk about the benchmarks. As usual my settings are somewhat unorthodox compared to the standard fare, because in most cases they’re the actual game settings I use when playing these games. There will be a total of 36 different games tested today.

Hardware

• Intel Core i5 750 (2.8 GHz, 21x multiplier, Turbo Boost on).
• 4 GB DDR3-1333 RAM (2×2 GB, dual-channel).
• Gigabyte GA-P55-UD3 (Intel P55 chipset, F6 BIOS).
• nVidia GeForce GTX285 (1 GB, reference clocks).
• nVidia GeForce GTX470 (1.28 GB, reference clocks)
• Creative X-Fi XtremeMusic.
• 30” HP LP3065.

Software

• Windows XP 32 bit SP3
• nVidia driver 197.13 & 197.41, high quality filtering, all optimizations off, LOD clamp enabled.
• DirectX February 2010.
• All games patched to their latest versions.

Settings

• 16xAF forced in the driver, vsync forced off in the driver.
• AA forced either through the driver or enabled in-game, whichever works better.
• Highest quality sound (stereo) used in all games.
• All results show an average framerate.

Hidden Super-Sampling

While this article was being written, a very interesting finding was made, that of hidden super-sampling present on the GTX470/GTX480.

Essentially by using TrAA (SS), the entire scene receives super-sampling, not just alpha textures. This works in OpenGL and in DX10/DX11. I noticed something was up in OpenGL anyway, so I disabled TrAA on the GTX470 to make it behave like the GTX285. But given I used TrMS/TrSS in every Direct3D title, I had to be sure it wasn’t happening in DX9 on XP.

Well after extensive testing, I can confirm all of the results are legitimate. SSAA is not being applied to the entire scene with TrMS/TrSS in DX9 on XP. That is to say, TrMS/TrSS behaves the same way it does on the GTX285, providing you keep the level in lock-step with the base MSAA level, which I did.

To double-check, I took some TrSS screenshots in some DX9 titles, and I can confirm super-sampling is only being applied to transparent textures, but nowhere else.

To triple-check, I disabled TrAA on the GTX470 and retested some of the games where it was slower (Call of Duty 2/4/5, UT 2004 and Bioshock 1), but the GTX470 was still unable to match the GTX285’s original score with TrAA.

So as far as these results go, I’m 100% confident they’re accurate as far as Windows XP goes.

Image quality in OpenGL titles is absolutely gorgeous with SSAA, and I’ll cover the hidden super-sampling (and other topics) in more depth in a future article concerning image quality.

2xAA

This is not a good start for the GTX470 as it racks up losses in every game, except two. Hell’s Highway receives a nice performance boost, and Jericho runs extremely well. I suspect the result in Jericho is what the card is capable of in other games if it had better drivers.

I’m very disappointed with the performance loss in Stalker Clear Sky because it’s one of the games where I needed more performance the most.

2560×1600

Here we see some reasonable performance gains in Stalker, Unreal Tournament 3, and Left 4 Dead 1. Unfortunately the rest of the games are about the same speed (or slower) on the GTX470.

8xMSAA

I cranked the settings to 8xMSAA (8xQ) in these older games. Serious Sam 2 sees an absolutely amazing performance gain from the GTX470, and Fear 1 gets a nice boost too. The rest of the games are the same or slower on the GTX470.

OpenGL

Clearly nVidia’s drivers have serious issues in OpenGL, at least as far as XP is concerned. The GTX470 actually posts an average loss overall over the GTX285 (9.42%), and there are big losses in Quake 1, Prey and Doom 3.

If you look here, the lowly 5770 (84.2 FPS) almost doubled the GTX470’s score when tested under the same conditions in Doom 3. The GTX470’s score is utterly unacceptable, especially in light of its improved ROP performance.

I’ll remind you that Quake 1 and Quake 2 are not the original versions, but are instead modernized source ports, and are vastly more demanding as a result.

4xAA vs 8xAA

Now I’ll test the performance hit of 4xMSAA vs 8xMSAA (8xQ). I almost didn’t bother completing this section, but I’m glad I did because there are some glimmers of hope for the GTX470.

There’s a much lower performance loss overall when moving from 4xAA to 8xAA compared to the GTX285. Of course any results where the GTX470’s 4xAA score is lower than the GTX285’s are hollow victories, but still, it shows the ROPs have potential, even when operating with subpar drivers.

In particular, we see a vastly lower performance hit in Call of Juarez 1 and Bioshock 1. The clear standout however is Clive Barker’s Jericho, where the GTX470 runs faster at 8xQ than the GTX285 does 4xAA. I’d almost suspect the GTX285 is running out of VRAM in some of these situations.

Yet again Prey has serious issues with the GTX470, coming off worse in every situation.

8xS & 16xS

Fortunately the combined mode super-sampling modes are still present in the driver, so I’ll test them in these classic titles.

Aside from the good performance gain in Half-Life 2, the GTX470 is absolutely woeful compared to the GTX285. It’s at best equal, and at worst 30.36% slower.

Conclusion

On XP the GTX470 is currently slower than the GTX285. It’s as simple as that. Moments of brilliance are vastly overshadowed by abysmal performance almost everywhere else.

Obviously the GTX285 doesn’t have a problem with either XP or my choice of settings, but the GTX470 does. So either XP’s driver isn’t optimal for the GTX470, or it’s a problem with nVidia’s GTX470 driver in general. Retesting on Windows 7 will answer that question soon enough.

The card is also much louder than the GTX285, even when sitting inside a solid steel Antec 902 with excellent cooling. My GTX285 often wouldn’t raise its fan much past idle speed during extended gaming sessions, but the GTX470 often spins up within seconds of entering a game, and it sounds like a hairdryer on a low setting. You will hear this card when gaming, even when wearing headphones like I do.

On the plus side, the GTX470 is about an inch shorter than the GTX285 which means it won’t overhang your motherboard, and hence might fit into smaller cases.

So that’s it for the XP results, but stay tuned for part two of this article where I’ll retest the GTX470 on Windows 7 (64 bit).

Pros

• Much faster than the GTX285 in a handful of situations.
• Best AF in consumer space.
• Rotated and sparse grid super-sampling in DX10/DX11 and OpenGL.
• TrAA can be decoupled from MSAA.
• Open and flexible profile system allows complete control over your gaming library.
• 1280 MB memory for demanding games.
• Relatively short for a high-end card.

Cons

• Current performance on XP is abysmal compared to the GTX285.
• Not always cost and/or performance competitive with ATi cards.
• Cooler is loud under load.

Please join us in our Forums

Follow us on Twitter

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

«1. Introduction2. Setup3. Hidden SSAA4. 2xAA5. 2560x16006. 8xAA7. OpenGL8. 4xAA vs 8xAA9. 8xS & 16xS10. Conclusion11. View All»