Techgage logo

NVIDIA’s PhysX: Performance and Status Report – Part 2

Date: August 6, 2008
Author(s): Rob Williams

It took a little longer than expected, but NVIDIA is soon to unveil new drivers that will open up PhysX support on all 8, 9 and GTX-series of GPUs. We’ve decided to follow-up on our previous article and see where PhysX stands today, and also pit seven GPUs against the new drivers.


In late June, I published the first-look we’ve had at PhysX since the acquisition by NVIDIA. Though I didn’t know it at the time, it was the first part of the series, with three parts likely to be in total. The first article happened rather quickly, because NVIDIA released a new GPU driver out of nowhere that opened up PhysX acceleration on select GPUs. I couldn’t help but test it out immediately.

Although that article happened just over a month ago, lots has happened since in the PhysX world, and we’ll be taking a look through all of it here.

One thing that impressed me with the last article is how well it performed. It quickly became one of our most popular articles this year, and that’s saying quite a bit given we were half-way through 2008. It became obvious to me just how interested gamers have become in PhysX, which is a stark contrast to how things were when AGEIA was at the helm.

Towards the end of that article, I promised to follow-up on my testing once new drivers became available, so here I am, and I’m sure you are to be pleased with what you’ll see throughout the reports. I should mention that this article was completed in a very short amount of time, so there are certain features I had to omit, which is one reason I see a part three of this article being necessary in the future.

So, what’s new with PhysX?

Gamers are not the only ones who’ve become interested in seeing PhysX progress, because as we’ll find out in this article, NVIDIA is taking the technology very seriously. Although things seemed slow post-acquisition, the company has been putting forth a lot of energy into their newfound physics accelerator, and I expect the rest of the year will be very interesting where PhysX is concerned.

For proof of their interest, look no further than the ‘NVIDIA PhysX Pack #1’ that will be released next Tuesday (ironically the same day as a certain competitor’s embargo lift). It will contain a slew of PhysX-related goodies that will set gamers off on the right foot. Here is what you’ll be able to expect:

We were told that a PhysX Pack #2 was en route, and I’m sure they’ll continue to be released as time goes on, depending on how much new PhysX content is out there. Aside from the Pack #1 itself, also released next Tuesday will be the 177.79 driver and also the PhysX software version 8.07.18. What this means is that you’ll be able to enable PhysX support on your 8, 9 or GTX-series card, no matter which model.

This is the patch we’ve all been waiting for, because it essentially blows the PhysX doors wide open for gained developer support. If all 8, 9 and GTX-series cards include PhysX support, it gives game developers all the more reason to support it, which to me, is a very good thing.

In the last article, I did a small amount of performance comparison testing, using 3DMark Vantage and Unreal Tournament III. This time I’ll be doing the same, but while also adding Ghost Recon Advanced Warfighter 2 to the mix, in addition to NVIDIA’s own Fluid techdemo and the Metal Knight Zero automated benchmark.

In this article, I will not be including certain types of testing like I did in the first one, namely due to lack of time. So, there will not be PPU comparison testing (using the AGEIA add-in card), nor will I be testing at varying game resolutions. These inclusions are not necessary, as the main focus of this article will be how each GPU scales with each other, when PhysX is brought into the picture.

One feature we still need to wait for is the ability to dedicate an installed GPU strictly for PhysX acceleration. For example, the scenario could include your SLI setup of dual 9800 GTX’s and then a single 9600 GT, which would be the card to handle PhysX computations. This would essentially ease off your big card/s to allow maximum average FPS, while leaving the minor work for the smaller GPU.

Before we jump into our performance results, please take a minute to review our system setup, as seen below. Note that we’ve tested using seven different GPUs, with another in SLI mode, which is why nothing specific is listed under video.

For tests that don’t involve PhysX acceleration, the 174.74 driver will be used, except for the 9800 GX2 and GTX 280, which both use the 177.41 driver. For all PhysX accelerated tests, the upcoming 177.79 driver is used across all cards, along with the PhysX 8.07.18 software package. That combination is what allows PhysX support on the 8, 9 and GTX-series of GPUs.

With that all taken care of, let’s start off our testing with some 3DMark Vantage.

3DMark Vantage Performance

Around the same time I published the first part to this series, NVIDIA’s latest GPU driver at the time was getting a lot of attention, but for all the wrong reasons. I won’t delve into the entire backstory here, but I do recommend checking out that article for more information if interested.

The basic reason was due to the fact that their latest driver caused over-inflated 3DMark Vantage CPU scores, which was true. But, I’ll still stand by my reasoning that NVIDIA did nothing wrong, as they simply pushed the PhysX computation from the CPU to the GPU, which is the exact same thing that happens when you play a PhysX-enabled game with the same drivers.

If there was blame to be thrown at anyone, it would be Futuremark. After all, it’s their benchmarking tool that obviously focuses way too much on the CPU aspect. But on the other hand, the over-inflated scores are not really over-inflated at all. We all knew that GPUs excelled in certain areas where CPUs don’t, and PhysX-specific computation is apparently one of those examples.

The fact of the matter is, comparing the CPU result of an Intel Core 2 Extreme QX9650 to a GeForce 9800 GTX, for example, only goes to show just how much more efficient a GPU is in that certain situation. In the end, I don’t blame anyone, because there’s no blame to be had. 3DMark Vantage showcases how the GPU can excel at PhysX when compared to the CPU, and that in itself gives us some interesting metric to work with.

In the first article, I ran 3DMark Vantage in its entirety, but due to time constraints and the lack of a need, I’m going to focus on just the CPU results this time around. Included will be the overall CPU score (which includes results from the PhysX acceleration) and also the individual results from CPU test #1 and #2. Below are the descriptions of the CPU-specific tests:

If no PhysX accelerator is present, then the benchmark will fall back on the CPU. All results are displayed as ‘operations per second’, although the results screen itself calls it ‘steps/s’. Also, regardless of the mode selected, the CPU tests run in a fixed resolution, so scaling between GPUs is kept accurate.

Let’s start off by taking a look at the overall CPU scores, both with PhysX acceleration disabled and also with it enabled. The differences are… well, undoubtedly obvious:

You can fine-tune the differences by seeing which test focuses on the CPU itself, and which targets PhysX:

It goes without saying that a faster GPU will allow greater PhysX capabilities, but that’s not much of a surprise. The rule of thumb when it comes to gaming with PhysX is the same rule of thumb for gaming in general: the bigger the card, the better the experience. The more performance a card can dish out, the better capable it will be to simultaneously handle both the graphics and PhysX at the same time. Small GPUs struggle.

But let’s face it, 3DMark Vantage is the least fun benchmark out there, so let’s move onto a real game, and a good one at that.

Unreal Tournament III Performance

It likely comes as no surprise that Unreal Tournament III utilizes the Unreal Engine III game engine, and it’s completely appropriate to use in our testing since it’s the same game engine that most PhysX-capable games are currently using. In the last article, I used the Heat Ray PhysX level for testing, and I’ll be doing the same here, but with a twist.

NVIDIA provided a recorded demo for us to play back that really stresses PhysX to the limit. For regular GPU reviews, I shun timedemos for reasons I’ve laid out clearly before, but when it comes to CPU or GPU computational testing, timedemos are completely relevant, especially when wanting to see the real scaling of multiple GPUs.

The level itself is great, even aside from the PhysX aspects. It looks fantastic, taking place in a part of a gritty town, and has many hiding spots. Also fun is a massive creature that can literally kill you just by looking in your general direction. The other players are far from being the only thing you will have to worry about.

Throughout the level are numerous gas canisters waiting to feel a bullet. When struck, they’ll explode and warp the surrounding air, but they don’t do much to the environment like other ‘techdemo’ games out there, like Warmonger and Metal Knight Zero.

Below, you can see the clear-cut results of PhysX vs. non-PhysX, and like the 3DMark Vantage scores, the differences are quite obvious. Using a bigger GPU will of course increase both the graphics performance along with the PhysX performance, but UT III is one game that really doesn’t need a ‘hardcore’ GPU, so anything bigger will essentially help you out with PhysX more than anything.

During this fresh round of testing, I realized a mistake I made in the first article: not testing out the other PhysX levels. This time around, I decided to load up Lighthouse, and it sure didn’t take long before I realized just how much better that level is for testing.

Throughout, walls can be blown out, gas canisters blown up and floors can be shot through. Doesn’t sound too intense, but just try playing it without PhysX-acceleration and you are sure to cry. The level begins out just fine, but within thirty seconds you begin to realize the mistake you made. As you progress, the game decreases in performance, and reaches a crawl of about 1 frame-per-second. Yes, horrible.

So this level also made a perfect test for real-world testing, and below, you’ll see the results of my runs:

You might notice the omission of not using the PhysX acceleration, and it’s for the simple reason that the results proved all to be the same, since the CPU was being targeted more than the GPU. It really wasn’t a worthy set of results to include. Plus, I tend to not enjoy games that run at 1FPS.

Levels like this are a perfect example of what a PhysX game is all about, though. Most of what is included might be eye candy, but it really can add a lot to the experience. After all, half the fun in games is being able to blow things up, and when it’s more realistic, it’s more fun.

Next up, Ghost Recon Advanced Warfighter 2 results.

Ghost Recon: AW 2 Performance

When the last article was posted, I received a few questions on why I didn’t include GRAW 2 in the testing, and the truth is, I forgot all about it. I’ve never been a huge fan of the series and it somehow completely left my mind before I sat down to work on the article. But it’s here now, and I admit, it was a rather fun one to test.

This game was benchmarked a lot differently than the others, because each run literally took a minute. This run of course begins out on AGEIA Island, and each run I make begins at the entrance to the camp, with gas canisters in front of me.

At the beginning of each run, I simply back up a few steps, then shoot the canisters to deliver one nice physics-accelerated explosion, and then I proceed up the hill until I reach two small huts. I then throw a grenade into the left one and watch it go boom, and then toss another right underneath the northern fence, and also see it fall to pieces.

Lastly, I backtrack and head down a hill to apply an explosive to the side of a gas tank, then head back up the hill. I then face the tank, blow it up and wave to the two idiots who were dumb enough to stick around. Overall, this entire run takes around one minute to complete, but throughout, the physics exhibited put the GPU to good use.

Below are the results from the run. Because the level cannot be run without some sort of PhysX accelerator, no CPU results are shown.

Once again, the results scale according to the GPU installed, and that’s not the least bit surprising. As mentioned earlier, the better the GPU, the better the experience. Depending on the GPU you have, you may either have to disable physics entirely (obviously not in this particular title), or lower your graphic settings in order to support the physics. It’s a trade off, but those with larger GPUs won’t have to worry as much.

Techdemos: Great Kulu, Fluids & Nurien

Along with the new PhysX Pack #1 comes a few new tech demos, all of which we’ll tackle here in some detail. I’ll briefly benchmark the ‘Fluid’ and ‘MKZ’ demos, since it’s easy to do and gives us an accurate representation of how the GPUs scale. Generally, techdemos are needless to benchmark as they don’t represent real-world gameplay, like UT III and GRAW, but that doesn’t make them any less cool to toy around with.

The demos available with the Pack #1 are ‘Great Kulu’, ‘Metal Knight Zero’, ‘Fluid’ and also ‘Nurien’. Kulu and Fluid are both interactive, but the other two are not. The Great Kulu demo is the most interesting of the bunch, mainly because you get to deal with a freaking mutant squid.

The Great Kulu Tech Demo

The Great Kulu is a techdemo built presumably in-house at NVIDIA and utilizes the Unreal Engine III game engine (aka the same engine many other games use). Since one of UE III’s features is PhysX support, it makes all the sense in the world to build a techdemo around it, to showcase just what PhysX is capable of in a real title.

You start off the demo inside a small room with one heck of an ugly squid that I’m sure will end up giving me nightmares. The goal of the demo is of course to show off PhysX, but it goes beyond what most games and techdemos do, which usually confine you to blowing stuff up. Here, the squid is part of the PhysX action and reacts to your gun’s electric shocks, and it itself moves in a realistic fashion.

As the demo continues, you’ll see the squid push his way through a few doorways, which is actually quite fun to watch. You can see his blubber move through the door, and he has to give it more than one go just to make it through. Towards the end of the demo, you acquire a gun that shoots a grenade-like explosive into his skin. Moments later, boom.

This part was tons of fun, because when it ‘sploded all over the place, you can see pieces of the body still jello-ing on the ground, and it actually looks quite realistic. Well, as realistic as a mutant squid blown to bits would be, I suppose.

Fluids Tech Demo

We’ve all seen fluid-based physics demos, but these two are actually somewhat interesting. The ‘Outdoor Scene’ showcases a small base with a massive pipe situated on top of one building. Out spurts a constant stream of dark water, clear water, oil or spheres, based on your choice.

The other demo shows a see-thru NVIDIA logo that contains the same substances as the first test, allowing you to rotate to your heart’s content. You can judge by the images just how addictive these tests can be, and you’d be right to assume so. Both are fun to play with, and further show what PhysX is capable of.

Did I mention they’re addictive?

Nurien Demo v0.7

Given that there is a demand for social networking tools and barely any exist, Nurien deserves the right to show us what it’s made of. This application is a little different than the rest though, as the graphics are actually good, unlike something akin to Second Life that people put up with for no reason, despite the fact that its graphics would have been considered very lackluster ten years ago.

I digress. Similar to Second Life, Nurien will allow users to create unique avatars, own the latest fashions, in-game objects, handle photos and video, and of course chat to each other. “How in the world is PhysX used here?????“, you ask?

The demo features a runway with very attractive fake women walking along showing-off their latest fashions. Surprisingly enough, the demo is based on UE III as well, so clothing utilizes realistic cloth simulations, as well as the curtains. Hair moves realistically and smoke particles also fill the air.

Sure, this is one weird way to show off PhysX, but it definitely gets the point across. Imagine such realism on the avatars in any one of your other games. If they moved this realistically, it would certainly add to the fun-factor.

Techdemos: Warmonger, Metal Knight Zero

Warmonger – Operation: Downtown Destruction

If you frequent our site, then this title should be of no secret to you, as we’ve talked about it quite a bit in the past. The first ‘techdemo’ released with the help of AGEIA was Cell Factor, a game that really lacked in gameplay but did a fine job of showing off what PhysX could do. The follow-up techdemo was of course Warmonger, a game that was delayed over, and over again, before its release late last year.

Like Cell Factor, Warmonger was released as a full game, but was really an over-glorified PhysX techdemo in disguise. Because of this, gameplay lacked, just like in CF, but this title really pushed the boundaries of ‘blowing s*** up’.

This multi-player-only game gives you four levels, each which takes place in a desolate city, and allows you to kill your friends and also anything else. You can tear through cloth sheets with your bullets, blow the tiles off columns and walls, actually shoot a sad-looking truck enough to bend the metal, and so forth.

While Warmonger isn’t exactly the best game out there, it’s free, and that in itself makes it a good deal. Plus, now that anyone with an NVIDIA 8, 9 or GTX-series card will be able to play it in its full PhysX glory, it’s going to see a lot more attention than it did when AGEIA was handling things.

Metal Knight Zero

This is one game that’s still shrouded in mystery, but as it stands, it appears to have the same goals as both Cell Factor and Warmonger… it’s a free title where PhysX is really allowed to strut its stuff. It’s an FPS, like the others, and allows you to blow up pretty-much whatever you feel like.

The game adds certain social-networking features, such as the ability to purchase items (with real cash or not, I’m unsure) and trade with other players. The game features a social-networking area where you can customize your avatar with certain accessories, such as clothing, helmets, equipment, weapons and panties. Alright, I made that last one up.

Throughout the techdemo, there are many opportunities to blow up gas containers, which result in dust particles, rubble being flung about and of course, lots of smoke. NVIDIA boasts the fact that 2,000 particles can be unleashed at once, so if this game were food, it’d be rice.

Like Cell Factor, Warmonger and others, this game includes cloth that can be torn with your gun. In addition, many other objects can be absolutely mutilated, such as trash cans, billboards, benches, boxes, vehicles, among others.

Like Warmonger, this game will be free on release, and is included as a non-playable techdemo with the PhysX Pack #1, so you have no reason to ignore it.

Final Thoughts

If there was ever a time that PhysX was exciting, it’d be right now. Never before have I felt so confident in the technology, because it’s obvious NVIDIA is taking it very seriously, as evidenced by the many new techdemos and upcoming games as illustrated throughout the article. From hereon out, things should only get better, as adoption by game developers heats up.

The problem right now is the same one shared by AGEIA a while ago, in that the main benefit of the technology can only really be seen in techdemos. There are a few games out there to support the technology, but as far as I’m aware, there is no game that supports the physics the entire way through, except for Cell Factor and other small ‘games’.

On the other hand, games like GRAW 2 and UT III show exactly what PhysX is capable of, in addition to the techdemos that are far too much fun to goof around with. After playing with the new demos, I’d be impressed if adoption of the technology didn’t pick up far more readily in the months ahead.

Of course it’s easy to get excited when everything is currently free. The ‘PhysX Pack #1’ features lots of goodies to get you started, and since many of you out there already have an NVIDIA 8, 9 or GTX card, there’s little reason to not download it once it becomes available (unless you have a stingy ISP). Despite the fact that some of the included items have been available for a while, the new techdemos are worthy of a download by themselves.

Upcoming support is a bit slim, but as I mentioned, I wouldn’t be surprised to see things speed up. There is Metal Knight Zero, which is sure to be boring (judging from what I see in the techdemo), but it will still deliever the same goods that Cell Factor and Warmonger did. Then there is Backbreaker, an upcoming American football game that actually goes beyond PhysX for graphics and begins to handle the player AI, and from what I saw at CES 2007 (seriously), it’s could very-well be the PhysX killer-app.

Beyond that, there is Empire: Total War and Mirror’s Edge, two games that actually look quite promising, so we can only hope that they live up to the hype. With NVISION ’08 right around the corner, I wouldn’t be surprised if we learned of more developers who’ve jumped on the bandwagan either. At least we can hope.

So where’s ATI in this? Nowhere right now, although I’ve been told by NVIDIA before that nothing stops other GPU vendors from supporting it. Of course, the problem isn’t so much deciding to support it as it is the fact that they’ve already signed over to Intel to support Havok. It’s all a very sticky situation. Intel doesn’t currently have Havok functioning via hardware acceleration, but I’d be hard pressed if that wasn’t one of the features of Larrabee when it’s released in 2009/10.

Is PhysX now worth getting excited about? Of course, especially given that everything up to this point has been free. If you already own a capable card, there’s little reason to not give the new demos a try, and if you are not a gamer, how in the world did you make it so far into this article?!

I mentioned briefly on the starting page of this article that there will be a part three to this series, partly due to the fact that this particular article was put together in very short time. There are a few things I’d like to investigate further, such as memory use, and whether or not more memory will be of any benefit, CPU usage with various GPUs, more real-world benchmarking and a few other minor things. PhysX is hopefully here to stay, so such testing should go to good use.

NVIDIA’s first conference happens in two weeks, called NVISION ’08, and it’s being held in San Jose. If you are interested in attending, tickets are available on their web site (different prices depending on what you want to do). We’ll be there and reporting whatever is of interest, so definitely stay tuned. If there are any PhysX updates, we’ll make sure you are the first to know about it.

Discuss in our forums!

If you have a comment you wish to make on this review, feel free to head on into our forums! There is no need to register in order to reply to such threads.

Copyright © 2005-2021 Techgage Networks Inc. - All Rights Reserved.