AGEIA’s PhysX PPU represents a new era in computing. For the first time ever, dedicated physics processing is becoming a reality. We take a look at the ASUS PhysX P1 to see if physics processing is all it’s cracked up to be.

{multithumb}
{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}
Introduction

If the universe is a strange place, our interpretations of the universe are even stranger. Our planet alone has gone from being a flat board floating in space with a big star revolving around it, to a slightly oblong sphere, itself revolving around the same star. I blame physics. Our attempts at applying the scientific method to the phenomena we observe in nature, a practice we know as “Physics,” is inherently flawed. Not so much flawed in the sense that the laws it produces are illegal, but flawed in the sense that the laws it produces are not always upheld. I suppose it is a natural human instinct to be skeptical of taking the word of a bunch of guys draped in white lab coats as the truth.

Whatever the case, physics has been and will always be one of the driving forces of our technological progression as a society. We see physics being emphasized in school classrooms today, now more than ever, in an attempt to catch up with the more scientifically astute minds on the other side of the globe. While a more rigorous approach to physics as a way of advancing the world technologically is a very worthy endeavor, the most important characteristic of the subject is its ability to enforce and uphold the principles of reality.

{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}It is these principles of reality that an imaginative bunch of game developers are most interested in. Over the last three years we have seen in-game physics effects grow by leaps and bounds. We’ve gone from holy unrealistic interactions between objects in a world in games such as DOOM and even Counter-Strike, to games like Half-Life 2 and the upcoming Unreal Tournament 2007, where realistic physics take the main stage. It is upon this reflection that I can’t help but sit back and be very proud of the game developers who, in about seven years, have been able to go from cartoon-looking worlds to some of the most realistic environments ever produced on screen. I blame physics.

Unfortunately for the end consumer, the physics in video games is often limited by modern technology, and performance in games that utilize these realistic physics engines usually gets hit pretty hard. That is why, not too long ago, a company called AGEIA began development on what they were calling a Physics Processing Unit (PPU). This unit, designed to be implemented as an add-in discrete board, would be able to handle all of the complex physics calculations that were traditionally carried out by a system’s CPU. This would potentially allow the CPU to do its own thing, completing what would be called the “Gaming Power Triangle,” whose vertices are the CPU, GPU and PPU. Now all of a sudden in games where you would have seen the same explosion time after time, you see highly dynamic and unique explosions with individual particle collisions on a scale never before imagined, well, except for maybe in real life. This is all provided, of course, that the game you are playing supports the AGEIA’s PhysX PPU.

{mospagebreak heading=Introduction&title=Theory}

Theory

As recently as one year ago you would be very hard-pressed to find significant backing for the need for hardware-accelerated physics. Early this year, however, developers started to realize that no matter how complex they made the physics algorithms in their game engines, they were unable to produce the advanced gaming physics necessary to bring the level of immersion to game play that was expected in order to complement the astounding visual effects starting to emerge in the latest titles. This “advanced gaming physics” would come to be defined as the simultaneous ability to produce physics effects on these four dimensions: Fidelity, Scale, Interaction, and Sophistication.

It should be noted that the complex nature of CPUs and even the parallel nature of modern GPUs can more than adequately produce physics effects in any one of those four dimensions, but they are inherently limited to only the one. The CPU is responsible for calculating and giving out orders. The GPU is responsible for receiving information from the CPU, rendering it, and displaying it on the screen. Taking away time from these responsibilities to handle the immensely unique physical calculations that would be needed to produce physics effects true to the four mentioned dimensions would not only be tremendously difficult, but it would reduce the overall performance of the computer for obvious reasons. Hence the need for a dedicated PPU to handle the calculations that allow the user to interact and perceive their surroundings as realistically as possible.

{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}What exactly are Fidelity, Scale, Interaction, and Sophistication? Furthermore, why are they important to gameplay?

Fidelity, defined as simply as possible, is the accuracy with which something is reproduced in regards to its true self. For instance, in audio terms, the sound you hear coming from your headphones is in high fidelity if it sounds exactly the same as did in the studio on the day it was recorded. This applies to the world of gaming by describing the realistic deformations and dynamic movements that are the result of collisions during game play. An example of this would be in a fighting game: when one boxer gives the other a right hook to the jaw, the face of the victim would be deformed not only on the level of skin movement, but the jaw itself would be shifted; the whole body would react in the opposite direction in a way that is equal and opposite to the reaction in the first boxer’s glove.

Scale, in this sense, would be the ability of the effect to increase or decrease in magnitude in progression. If that doesn’t make sense, then think of scale in terms of the boxing scene we have just described. Once the boxer’s glove hits his opponent’s jaw, the beads of sweat on the impacted face are sure to be displaced. The ability of these beads to both be displaced, and perhaps fly from the impact in in the form of mist and drops of sweat like you would see in a real boxing match, represent the dimension of scale.

{mospagebreak title=Theory (cont)}

Interaction is probably the easiest of the four dimensions to describe. Simply stated, interaction is the effect that the characters have on their environment, and effect the environment has on the characters. Again in terms of the boxing scene, the sweat from the boxer’s face hitting the floor would both splash and cause the floor to become slightly more slippery. The boxer may be knocked off balance and run into the ropes around the ring, which in turn cause him to spring forward, back towards his opponent.

Sophistication then, is essentially the magnitude at which all of this can occur. The boxer will be built on a life-like skeletal frame rather than a series of boxes and spheres. The boxing glove will be a boxing glove instead of a sphere. In a game where you can shoot steel drums, the drum would deform from the bullet impact as expected, rather than a simple material change. In the event of an explosion causing this drum to be blasted away, the deformed part would slide with increased friction along the ground surface, whereas the non-deformed part would roll as expected, but be interrupted by the deformed part. Highly realistic, flexible, and unique models add to the physics effects and realism.

You may have noticed that these four dimensions seem to overlap quite significantly. That is the beauty of the whole thing. The effect of the full implementation of two or even three of these dimensions would be nothing compared with the realism and immersive effect that is created through the synergy of all four dimensions.

{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}However, processing these four dimensions to their utmost degree is not an easy task. Not only would the chip need a highly parallel architecture to accommodate for the large amounts of information that need to be processed to produce the required scale, but the architecture itself would have to be custom tailored to physics calculations. A parallel architecture is more effective in this scenario because it can take these large chunks of data and compute them at once, as opposed to a more serial architecture that computes smaller bits of data simultaneously. Furthermore, the data required to process these four dimensions is traditionally placed in the system RAM before it is calculated by the CPU. Should this data be placed in a memory bank that is dedicated to the parallel processor discussed above, the speed at which this data could be processed increases, as does the efficiency of the system’s RAM, which is no longer having to store all of these bits of data.

More observant readers will note that there already exists a processor in gaming computers that acts in this manner, the GPU. However, a GPU is a highly specialized processor that is designed to render and display frames on the screen. Due to its parallel nature, it would likely be fairly effective at physics calculations, but just doing this would take away from the graphics processing capabilities of the card, and it would still not be as effective as a processor that is specifically designed to handle these physics calculations. That is where AGEIA’s PhysX PPU comes in.

{mospagebreak title=Features}

Features

Here are the features and benefits of the AGEIA PhysX processor, taken directly from the ASUS PhysX P1 box:
{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}
These features can all be summed up by stating the four main advantages the AGEIA PhysX PPU offers.

1. Memory Bandwidth – The PhysX PPU provides for a HUGE amount of memory bandwidth, somewhere to the order of a maximum possible 2 terabits per second. This is several times more than what we see on modern CPUs. All of the data stored in the dedicated GDDR3 memory for each particle can now be collected and sent through the processor much more quickly than on a CPU. This accommodates for the massive amount of scale necessary for producing that extra level of realism.
   
2. Optimized Physics Processing – Those of you that have studied physics and dynamics will note that most of the calculations in the field involve the use of the use of Newtonian mechanics. Newtonian mechanics utilizes several basic formulas that are then specialized for whatever application. The calculations used to evaluate these formulas are primarily based on linear algebra. The PhysX PPU is optimized for geometric and linear algebraic calculations. This allows objects to uphold their physical shape and react in true dynamic fashion as we would see in real life. This attributes to the Fidelity involved in producing realistic effects.

{multithumb popup_type=none enable_thumbs=0}

[ArticleAdvert]

{multithumb popup_type=default enable_thumbs=default}

3. Multi-Core Processing Complex – Details of this aspect of the PPU are not very clear, expectedly because AGEIA would not benefit from giving away their secrets. Apparently there are many individual processing elements on the PhysX PPU that can be used to compute the different physics calculations simultaneously. Almost like a bunch of small parallel cores making up an overall serial processing unit, kind of reminiscent of the architecture of a Cell processor.
   
4. Memory Architecture – Having an extremely high potential memory bandwidth is not enough for a processor that hopes to be most efficient at physics calculations. Objects in physics are inherently moving and changing. Likewise, data for these objects, including shape, size, and surface friction, are also constantly changing. The memory for the PhysX PPU had to allow for this dynamic data characteristic. The solution was to rework the memory architecture to accommodate for increasingly random and spread-out memory reads from the bank.

As we have stated time and time again in our other articles, theoretical features and potential performance is all very interesting stuff and makes for a good conversation, but if the realized performance of the card does not live up the expectations that that doesn’t do anybody any good.

{mospagebreak title=ASUS PhysX}

ASUS PhysX P1

There are currently only two companies producing retail parts for AGEIA’s PhysX PPU: ASUS and BFG. Today we will be looking at the ASUS PhysX P1 Physics Card. A quick glance over at our forums will make it quite clear that products from ASUS are quite popular among the gaming community. Their motherboards, which are known for their quality, stability, and above-average overclocking ability are well-regarded amongst gamers who prefer to build their own PCs. Let’s take a look at the specs of the ASUS PhysX P1 physics card.
{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}
In the graphics card world, a company called NVIDIA lets OEM partners stray from the reference design of its graphics boards. For instance if you look up the NVIDIA GeForce 7900GT graphics card on Newegg, you will see different-looking graphics cards, with varying specifications. This is done by the OEM manufacturers because they want consumers to buy their version of the card, for obvious reasons. AGEIA on the other hand has apparently not allowed ASUS and BFG to alter the operating frequencies of the parts on the graphics board. This could be for a lot of different reasons, including an optimized clock ratio, power consumption, and heat concerns. That is all speculation though, and as ATI has learned, it is not necessarily a good or bad decision in the first place. Nevertheless this results in PhysX cards with identical specs from both ASUS and BFG. The only distinguishable difference between the cards is the cooler designs and the software packages.

{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}Before getting started with the tests, we thought we would address some of the specs that you might have questions about. One of the things that stands out to us when we look at these specs is that the processor frequency is not revealed. This could be the case for several reasons, none of which we have any idea about. However, if the processing world has taught us anything in recent years, it is that clock speeds really don’t count for much. We can look at the CPU scene and see a 2.8GHz Athlon64 FX-62 beating out a 3.46GHz Intel Pentium 4 EE, and then turn right around and see a 2.4GHz Intel Core 2 Duo E6600 beating out that same FX-62. Similarly in the graphics scene, certain cores at low clock speeds beat out their higher-clocked competition. So when it really comes down to it, not knowing the core clock speed of AGEIA’s PhysX PPU is not very important.

Moving on, the next listed specification that stands out is that the PhysX PPU utilizes the PCI bus. This came as a surprise to us on our first look; why would a company debut this new device on the oldest of the currently used add-in card interfaces? However, the reason they probably did this becomes clear when you realize that out of every computer currently out there, a good 95%+ of them have PCI slots. This maximizes the potential consumer base for AGEIA. There is no means of testing the effectiveness of this card on the PCI interface compared with the PCI-Express interface, but we wouldn’t anticipate any tangible differences.

The final thing on the list that catches our eye is the value of 28W for peak power consumption. This number is pretty low for a chip like this and as a result, we wouldn’t expect the heat output of the card to be very high at all. However, it is still high enough that it will need to make use of a supplementary power cable, because the PCI bus can supply a maximum of 15W. For comparison purposes, ATI’s Radeon X1900 XTX under heavy 3D load can consume as much as 120W.

{mospagebreak title=Package}

Package

The ASUS PhysX P1 comes bundled with quite a good amount of software. We received the “GRAW Edition” of the card, so we got a full copy of Ghost Recon: Advanced Warfighter along with it.
{multithumb}

The package also includes demo CDs for a PhysX optimized game called CellFactor, and an interesting Marble Madness-esque game called Switchball. In terms of hardware, the package includes the PhysX P1 card itself, a molex power connector cable, and a lovely CD wallet.

{mospagebreak title=Test Methods}

Test Methods

There is something we need to make very clear before we continue with this part of the review. AGEIA’s PhysX PPU is a very young technology, and as such, support for it at this time is quite limited. There are very few games currently out that support the technology, and even fewer still that can support it or not support it. Of these games, there is only one major label that makes use of the technology: Tom Clancy’s Ghost Recon: Advanced Warfighter. (Note: City of Villains also supports the PhysX processor, but only in a very specific part of the game that, due to time constraints, was not reachable for this review.)

As a result of this limited software support, it is extremely difficult to effectively test the PhysX processor’s performance benefits. It gets even harder when you take into consideration that GRAW has no demo playback capability or built-in benchmarking tool. This means that in order to run benchmarks on the PhysX processor, you must devise customized testing methods. As a result, it is very unlikely that you will see similar results posted in two different reviews of the card. That said, we installed the ASUS PhysX P1 into our test bed and tried our best to get some meaningful results.

Test Setup

• Case: Antec Sonata II
• Power Supply: Antec NeoHE 550W
• Motherboard: Intel Desktop Board D975XBX (Bad Axe)
• Processor: Intel Core 2 Extreme X6800, 2933MHz
• Hard Drive: Western Digital 2500KS 7200RPM, 250GB with 16MB Cache Buffer
• Video: eVGA NVIDIA GeForce 7900GT KO Superclocked (580/1580)
• Memory: 1024MB (2×512MB) Corsair XMS2 PC2 6400 (TWIN2X1024A-6400)
• Optical Drive: Lite-ON SHW160P6S05
• Cooling: Thermaltake Big Typhoon
• Audio: Creative SB Live! 24-bit
• Physics card: ASUS PhysX P1

Software Configuration

• Motherboard BIOS: D975XBX Express BIOS Update (Rev. 1209, BETA)
• Operating System: Windows XP Professional with Service Pack 2
• Video Driver: NVIDIA ForceWare Version 91.36 WHQL Certified (August 15 release)
• Physics Driver: AGEIA PhysX version 2.5.0 (August 2 release)
• Game Version: GRAW version 1.10

{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}As mentioned above, there is no really great way to test the benefits of this card. This is due to the lack of being able to test an identical set of physics calculations with and without the ASUS PhysX P1 card installed. Hopefully in the near future we will see more of the in-development titles released, and some of them might include a proper way to attain some performance numbers (note to developers: pretty please?).

Also due to the less than impressive software support for the card, we will only be testing performance on one game, GRAW. Our reasoning for this is that of the games that currently support the PhysX processor, GRAW is far and away the most mainstream and likely to be played by hardcore gamers such as those currently reading this article. This is not to say that there will never be a game that supports the PhysX processor that will be any bigger than GRAW, indeed this is not the case at all. A quick visit to the AGEIA site reveals a list of games that will support the technology, but are currently in development. There are quite a few games on that list that impressed us from early screenshots and demos, most noticeably Joint Task Force and, of course, Unreal Tournament 2007.
{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}
Our custom test method for GRAW consisted of the most easily-repeatable complex physics interaction we could think of. In the game, you are dropped from an airplane into the heart of Mexico City, and your first task is to rendezvous with your team of soldiers (Ghosts). The first soldier you should meet with is named Kirkland. Near Kirkland there is a building that has two pallets of boxes in front of it by a loading ramp. What’s that? Boxes you say? Yes indeed, a whole bunch of undisturbed, identically shaped, and ready to be messed with boxes. What better place to throw a grenade? This is exactly what we did. We threw three grenades into these boxes and recorded the frame rates we achieved when doing so. We repeated this process as consistently as possible for the duration of 10 tests. This process was carried out with and without the PhysX processor, to gain a real idea of the performance differences. This process will be shown visually in the following section.

{mospagebreak title=Testing}

Testing

As stated above, our custom testing method of GRAW consisted of tossing three grenades into a big stack of boxes. We saved the game after throwing the first grenade in order to ensure all of our tests started in exactly the same manner. As a result, approximately 3 seconds after our tests start, a grenade blows up in the stack of boxes, causing them to blow up and fly every which direction. This would be what we classify as the “Fidelity” dimension of the test. After the initial scattering of boxes, we threw another grenade into the main pile at 15 seconds into the test. This grenade then exploded at approximately 19-20 seconds into the test. We did the same thing again at 30 seconds, and saw the explosion at about the 34th second. We ran the test without the ASUS PhysX PPU installed first, to get a “control” set of data as it were. FPS was plotted against time with the aid of Fraps. The test duration was held at a constant of 45 seconds from level load. The results of our first tests are displayed graphically below.
{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}
{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}You can clearly see that the FPS drops at instances where the grenades exploded and caused the massive physics extravaganza to take place. Also please keep in mind that the first 4 seconds of these results should be disregarded because they occur while the level is still partially loading and the frame rates in Fraps are just ramping up. One thing you will notice is that our test bed manages to keep the frame rates well above 100, even during the explosions. This was interesting to us considering GRAW is such a pretty looking game. Quantitatively, it appears that the explosions consistently caused a drop of 40 FPS. After running this test, we installed the ASUS PhysX P1 with the latest driver version, 2.50. There are no options within GRAW that allow you toggle on or off physics acceleration, so we had to assume the game automatically detects the PhysX card (note to developers: If we are pushing for the advancement of gaming physics, how can we not have customizable options for the game physics?). Brace yourself - here are the results of our GRAW test with the ASUS PhysX P1 PPU installed.
{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}
Obviously the first thing you will notice here is that there are again FPS drops when the grenade explosions occur. The next thing you will notice is that these FPS drops are far more drastic than in the previous test. This almost seems counter-intuitive: Why would the FPS drop more with the PhysX card, a card that is designed to offload physics calculations from the CPU, than without the PhysX card? We were kind of confused with this at first too, but then we realized what exactly the PhysX card is supposed to do.

{mospagebreak title=Analysis, GRAW}

Analysis: Ghost Recon Advanced Warfighter

You see, the physics card is somewhat like a preparation tool. It takes the event that will shortly be produced on the screen and does all the calculations that will increase the fidelity, scale, interaction, and sophistication of the scene, as stated above. That, however, is the limit of the PPU’s involvement. The PPU does not aid in the visual production of the event on the screen; that is a task that is still handled 100% by the video card. Think of it like this: the physics card has just created a far more realistic explosion, but the video card still has to render all of those extra particles. The end result is a much more believable visual effect, at the cost of system performance due to the increased graphics load.

To view videos of the explosions with and without the PhysX effects, download the following file: shoes_physx_grawvideos.zip (17.3MB .WMV)

It is a tradeoff then; a more realistic visual effect for a significant drop in FPS. This, again, might seem counter-intuitive. Why not just crank up the visual effects in the game and lose the same amount of FPS and save yourself the cost of a roughly $250 ASUS PhysX P1 card? Well, the fact of the matter is that cranking up the visual effects in the game will result in added video card load, and that load will be generated by enhanced texture detail, shading, and shadows – no physics. If there was an option to increase game physics, the reality is that the impending drop in performance due to increased CPU load would likely cause a larger FPS drop than we have witnessed here with the PhysX card.

Curious as to exactly what amount of work is being offloaded from the CPU to the PhysX card, we ran a performance monitor log during these scenes as well. The results of these tests indicated an insignificant difference in terms of CPU usage time. See for yourself:
{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}
{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}The difference between the CPU usage with and without the PhysX processor was less than one tenth of a percent (The usages being 53.09% and 53.18%, respectively). Even with the tremendously powerful processors that we have nowadays, this CPU load would not even account for an increase by 1 frame per second. These results really left us with a sense of confusion. It does not make sense that there would not be a higher physics load on the CPU without the PhysX card. The only possible explanation for this is that the Physics effects in GRAW are relatively insignificant, or that they are “pre-loaded” so to speak in the constant processor load, meaning that there will be no spike in CPU usage when a physics-intense event occurs. If this is the case, the GRAW might simply not be a very good indication at all of the abilities of the ASUS PhysX P1 physics card.

As stated before, however, there really is no excellent way to test the abilities of these add-in physics cards due to the lack of sufficient software support. The largest scale of physics use that we know about can be found in a game called CellFactor. In this game you are an elite soldier-type guy who has weird psy-op abilities and a futuristic machinegun/grenade launcher. It is a fun game insofar as you get to make stuff blow up on a scale never before seen in any game. You can launch a grenade into a mound of oil drilling equipment stuff, and watch it be blown to smithereens. We would LOVE to see the difference in performance that the PhysX card affords in this game, simply because it is so intensely based on physics. However, without program code modifications, there is no way to play this game without the PhysX card installed. Therefore we are not in a much better position than when we started.

{mospagebreak title=Conclusion}

Conclusion

We’ve seen some very interesting results from the ASUS PhysX P1. Well, not so much the ASUS PhysX P1 as the AGEIA PhysX PPU. What we can say about the ASUS PhysX P1 is that it is a very nice card, has a more than adequate cooler that is also relatively quiet, and does not consume much extra power at all. We also liked the AGEIA by PhysX logo they slapped onto the front of the faceplate; it adds a very nice touch to the rear of the computer. We don’t have anything to compare it to, so we can’t really say if it is better or worse than the competition. But it does have high quality capacitors which is something we like!

AGEIA’s PhysX PPU is a singular product that we feel is still very much in its infancy. What we have seen from our GRAW tests is that the technology, while promising, is not implemented in mainstream games to the extent it could be. In our custom explosion test it was readily evident that the visual splendor of the scene increased somewhat dramatically, but it came at the expense of frame rates. This is due undoubtedly to the video card having to foot the extra rendering load created by all of the additional particles. There was also the confusing result of the insignificant difference in CPU load from running our test with and without the PhysX PPU installed, an anomaly we still don’t have an answer for.

It is unfortunate really, that the PhysX PPU has such mediocre software support at this stage. The technology in the card, in our opinion, is clearly powerful and has a tremendous amount of potential. However, there are no games, at least no games that anybody is really going to want to play, that support the PhysX technology. The games are coming though, they are definitely coming. Games like “Joint Task Force” look like they will be able to make use of the PhysX technology on a grand scale, as do games like Unreal Tournament 2007. The upcoming PhysX patch for Rise of Nations: Rise of Legends also promises to bring another level of realism with the increased detail that the PhysX PPU will afford. There is no doubt in our mind that hardware accelerated physics is going to absolutely take off over the next two years, but at this point, everything just seems… not ready.

{multithumb popup_type=none enable_thumbs=0}

{multithumb popup_type=default enable_thumbs=default}Right now, it seems that the only real competition for AGEIA is time. The time it takes for these killer applications to come out that will make the PhysX card worth its weight and some. We see this problem as two-fold. On one hand, if these games take too long to come out, the general consumer base that AGEIA is targeting might lose interest in the product, and the PhysX PPU might slip from the limelight and end up going the way of a certain product from a company that starts with an “R” and rhymes with “Wrambus.” We would hate to see this happen, as we have already stated we think that the PhysX PPU, the ASUS PhysX P1 in particular, is a great product with tons of potential. However, it is a possibility. On the other hand, there are two quite viable alternatives to the PhysX PPU in terms of hardware acceleration solutions currently in the works. One of these alternatives is, or at least appears to be, a far more viable option than the other. The first, and what we consider a far less-likely option, is that the advent and implementation of multi-core CPUs, not just the quad core behemoths that will be upon us shortly, but processors like the Cell that will use around nine different, specialized cores to complete different tasks, will handle the physics calculations. We see a couple of problems with this. The first problem is that modern CPUs, while absolutely incredible at general computing tasks, are less than spectacular at specialized calculations like video rendering and physics (hence the implementation of dedicated GPUs some 15-ish years ago). A processor that is specialized to the extent of a PPU will always be superior at performing the linear algebraic functions necessary to do physics calculations. The second potential problem with this solution is that while hardware grows, software grows alongside. Applications and even operating systems in the future will be largely multi-threaded. Taking away from the CPU time of these applications could have adverse affects on the performance of your computer beyond what we have seen today.

The second, more viable solution, is the use of a technology called Havok FX. Havok, the undisputed heavyweight champion of software physics for many years now, announced back in November of 2005 that they would actively be developing a technology to compete with AGEIA’s recently announced PhysX PPU. Game developers, and indeed gamers, are quite familiar with Havok physics engines. It is, after all, a modified Havok 2 engine that makes up the lion’s share of a game engine we have come to know as “Source” (Half-Life 2). Havok FX technology is rumored to offload physics calculations from the CPU and make use of the idle time experienced by modern GPUs. While most modern GPUs are actually pushed to their limits by current games, there are features on almost all graphics cards that are largely unused. While the specifics of how Havok plans to balance the physics load with the graphics load are not yet known, it is clear that a multiple-GPU solution would be far more suited for such a task than a single video card. In fact, ideas have been flying around in the minds of NVIDIA and ATI that involve crazy things like 3 dedicated graphics cards (one being for physics) and multi-card, multi-GPU setups. While it would seem that a specialized physics processor would be superior to a GPU in terms of efficiency, there is no way of verifying this assumption. If a multi-GPU solution does indeed turn out to be a viable alternative AGEIA’s PhysX PPU, we will have a hard time advising people to disregard the obvious flexibility advantages that such a solution would have over the PhysX.

It is because of all the reasons stated above that we simply cannot recommend AGEIA’s PhysX PPU to the mainstream gaming audience. For the ultra-enthusiasts that seem to have bag fulls of loot hidden under their beds, go for it. The enhanced visual affects afforded by the PhysX PPU will not be matched for at least another few months, and by that time you should see a considerably larger amount of hardware support for the card. But as it stands, the average consumer and even most enthusiasts stand to gain very little from the AGEIA PhysX PPU. About four months from now it could and most likely will be a completely different story, as games will have come out that benefit insanely from a dedicated physics processor. AGEIA’s PhysX PPU is a product that just seems to have been launched before its time, the rest of the world just needs to catch up.

Read the addendum to this article, which includes performance numbers from CellFactor, here.

Popularity: 6% [?]

You Should Also Check Out These Post:

• Happy Holidays!
• NVIDIA Spikes Previews of GTX 295 With Limitations
• UnitedKeys Taking Pre-orders of its OLED Gaming Keyboard and Keypad
• BFG Will Show Off It's First Full Gaming PC at CES 2009
• Logitech Announces... You Got To Be Kidding...

More Active Posts:

• Some Updates (8)
• NVIDIA has an overcomplicated lineup (8)
• What happened... (8)
• Welcome To FPSLabs V3 With Added Awesome! (7)
• NVIDIA GeForce 9800GTX on April 1st (7)
• Razer Piranha Review (7)
• Microsoft to sell XP until 2010 (6)
• Weekly Hardware Recap (6)
• PureTrak Talent Review (6)
• The Weekly Hardware Recap (5)