Slashdot Mirror
Ageia PhysX Tested
MojoKid writes "When Mountain View California
start-up Ageia announced a new co-processor architecture for Desktop 3D Graphics that off-loaded the heavy burden physics places on the CPU-GPU rendering pipeline, the industry applauded what looked like the enabling of a new era of PC Gaming realism. Of course, on paper and in PowerPoint, things always look impressive, so many waited with baited breath for hardware to ship. That day has come and HotHardware has fully tested a new card shipped from BFG Tech, built
on Ageia's new PPU. But is this technology evolutionary or revolutionary? "
Well, since there hasn't been anything like it before, it would be Revolutionary by definition. However, I think that it will be a little while before we can really make any intelligent conclusion on the matter as it is still way to early in the development cycle for any kind of "review" to be valid. What, with one game and one demo as all that is available? Too soon.
It's bated breath, not baited.
Bated as in masturbated.
...they could use a card dedicated to keeping their server up when Slashdot finds it. It's already down for me.
picpix image polls. create - share - vote. fun!
Since Mainframes, I've always thought it makes more sense to modularize hardware.
While studying for my EE, I often wondered what the purpose of having a clock was, since so much of the individual chips often had finished their calculations before the next clock cycle came around.
I think we are going to see the clock go away, replaced with "Data Ready" lines, which will also help heavily in determining the bottlenecks in a given system (Hint: it's the system that is taking the longest to put up the "Data Ready" flag).
I also think that optics will be the way of the future. Quantum will be like Mechanical Television: cute idea, but impractical for mass production.
Optics. Think of it this way: Imagine a bus that can address individual I/O cards with full duplex, simply by using different colors for the lasers. Motherboards are going to get a lot smaller.
That's my opinion, anyway.
Joe
---
Q:Why couldn't Helen Keller drive?
A:Because she was a woman.
The link: http://www.anandtech.com/video/showdoc.aspx?i=2751
Short summary: Great for synthetic benchmarks, probably not real-world ready.
Slashdotted instantly. Perhaps PPUs will be good for rendering simulations of the Slashdot Effect.
... honestly
Someone needs to come up with a chip to offload slashdot traffic!
Its a neat idea, but at what they're charging, will many people add yet another card to their motherboard?(Heck, my PCI-e slots are already jammed full)
Without question, one of the hottest topics throughout the industry this year has been the advent of the discrete physics processor or "PPU" (Physics Processing Unit). Developed by a new startup company called Ageia, this new physics processor gives game developers the opportunity to create entirely new game-play characteristics that were not considered possible using standard hardware. Since its original inception, both CPU and GPU vendors have come to the spotlight to showcase the ability to process physics on their respective hardware. However, the Ageia PhysX PPU is the only viable solution which is readily available to consumers.
For the foreseeable future, the only vendors which will be manufacturing and selling physics processors based on the Ageia PhysX PPU are ASUS and BFG. With ASUS primarily focusing on the OEM market, BFG will enjoy a monopoly of sorts within the retail channel, as they will comprise the vast majority of all available cards on store shelves. Today, we will be running a retail sample of BFG's first ever Physics processor through its paces. Judging from the packaging alone, you can tell that this box contains something out of the ordinary. Housed in an unusual triangular box with a flip-down front panel, consumers can glimpse the card's heatsink assembly through a clear plastic window.
BFG Tech PhysX
Card And Bundle
Flipping the box, consumers are presented with a quick listing of features complete with summaries and a small screen-shot. Most importantly, the package also lists the small handful of games which actually support the PPU hardware. This short list consists of City of Villains, Ghost Recon Advanced Warfighter, and Bet on Soldier: Blood Sport.
Upon opening the packaging, we are presented with a standard fare of accessories. Beyond the actual card itself, we find a power cable splitter, a driver CD, a demo CD, and a quick install guide. Somewhat surprisingly, we also find a neon flyer warning of a driver issue with Ghost Recon Advanced Warfighter that instructs users to download the latest driver from Ageia to avoid the problem. This is a bit disheartening as there are only three games which currently support this hardware. With this in mind, it is hard to not feel as though the hardware is being rushed to market a bit sooner than it probably should have.
Directing our attention to the card itself, we find a rather unassuming blue PCB with a somewhat standard aluminum active heatsink assembly. Amidst the collection of power circuitry, we also find a 4-pin molex power connector to feed the card as a standard PCI slot does not provide adequate power source for the processor. At first glance, the card looks remarkably similar to a mainstream graphics card. It's not until you see the bare back-plate with no connectivity options that you realize this is not a GeForce 6600 or similar product.
Thankfully, the BFG PhysX card does not incorporate yet another massive dual-slot heatsink assembly as so many new pieces of high-end hardware do these days. Rather, we find a small single-slot active heatsink that manages to effectively cool the PPU while keeping noise at a minimum. Removing the heatsink, we were pleased to find that BFG has done an excellent job of applying the proper amount of thermal paste and that the base of the heatsink was flat with no dead spots. After powering the system, we see that BFG has dressed the card up with three blue LED's to appease those with case windows.
With the heatsink removed, we have our first opportunity to glimpse the Ageia PhysX PPU in all its glory. Manufactured on a 0.13u process at TSMC, the die is comprised of 125 million transistors. Overall, the size of the die is slightly larger than the memory modules which surround it. Looking closely at the board, we see that the 128MB of memory consists of Samsung K4J55323QF-GC20 GDDR3 SDRAM which are rated for a maximum frequency of 500MHz. Unfortunately, neither BFG nor Ageia have disclosed what frequency the PPU memory and core operate at, so we are unsure
From what I was able to read of the article before it got slashdotted, it sounds like games that can take advantage of it require installation of the Ageia drivers whether you have the card or not. This leads me to believe that without the card installed, those games will use a software physics engine written by Ageia, which is likely to be unoptimized in an attempt to encourage users to buy the accelerator card.
Also, it's likely to use a proprietary API (remember Glide? EAX?) that will make it difficult for competitors to create a wider market for this type of product. I really can't see myself investing in something that has limited support and is likely to be replaced by something designed around a non-proprietary API in the case that it does catch on.
Arguing about vi versus Emacs is like arguing whether it's better to make fire by rubbing sticks or banging rocks.
Reposted link for those who missed the AC post:1
http://www.anandtech.com/video/showdoc.aspx?i=275
There's an excellent review up at AT.
I think that while this card can do some amazing physics stuff, we aren't ready to make use of that capability for anything more than a little eye candy. Not in networked games, at least. Trying to keep everyone's world in sync is hard enough as it is, without adding even more objects that need to appear in the same place for everyone.
http://www.pcper.com/article.php?aid=245
http://www.pcper.com/article.php?aid=244
The one time I go to actually read the artical and it's been Slashdotted.
Here is an alternative artical for anyone interested:
AGEIA PhysX tested with GRAW and Cell Factor
Go ahead and call me unreliable; reliable is just a synonym for predictable.
Well coral cache has page 2 at leastc le.aspx?page=2&articleid=816&cid=2
http://www.hothardware.com.nyud.net:8080/viewarti
I'm not sure what you've heard quantum computers promise to do. I agree with you that optical computers, or rather, some form of classical computer will be the predominant type.
As far as we know, quantum computers can do (a) few things better than a classical computer:
Factoring numbers (this is why they'll kill public key encryption as it stands now)
Computing discrete logarithms (another function of cryptographic interest)
Search unstructured data (a quantum computer can do this in O(sqrt N) time as opposed to O(N) for a classical computer to run through every element)
Simulate quantum systems (fairly obviously)
Other than that, there's no asymptotic advantage to using quantum computers. For doing your taxes or even hosting the Transgalactic Wikipedia, classical computers are here to stay.
compact_support
Anandtech posted these video sequences to show what you see with and without the card.
The Anandtech article states that the physics hardware slows down the framerates which Aegis can't possibly be happy about.
short for "abated"
It isn't worth much until games actually start using it.
http://www.anandtech.com/video/showdoc.aspx?i=2751
"The added realism and immersion of playing Ghost Recon Advanced Warfighter with hardware physics is a huge success in this gamer's opinion. Granted, the improved visuals aren't the holy grail of game physics, but this is an excellent first step. In a fast fire fight with bullets streaming by, helicopters raining destruction from the heavens, and grenades tearing up the streets, the experience is just that much more hair raising with a PPU plugged in."
http://www.TheGamerNation.com/Forums
Ive been following them for a long time- their software demos blew my mind a few years ago (the one with the towers made of bricks that you could destroy oh so fun). We should wait for real games to make use of the physics. Ghost recon uses it as a gimmick. The tech demo game as listed in one of the articles is a real showing of what the card is capable of. When the game engines catch up and use it as an intrical part rather than a gimmick it will usher in a new era of gaming. It really will, look at what happened with hardware 3d.
Wow, I love how everyone sounds so intellectually sophisticated and interesting when they use such pithy phrases.
Is there any competion for Aegis? Reviews are all fine and dandy but product comparisons is where the decisions should be made. It should be based on which PPU can perform a given task faster/better. Competition would also drive each competitor to better their own product to beat the other. However, they shouldn't be mutually exclusive (ie. If you use Product A, then you can't use a program with only Product B support).
I wonder how long it will be before there is a mainstream demand for a separtate physics unit (probably as soon as games require them). It sounds like a great idea to take some of the load off the CPU. Does this mean that now game performance will be more directly linked to the speed and power of the GPU and PPU and that the CPU will be more of an I/O director and less of a number cruncher?
I've seen numerous posts of people saying that they do not have any available PCI slots. Will the introduction of a new type of card lead to larger motherboards with more slots or might it lead to a small graphics card that does not monopolize the PCI space? Also, there is the concern of adding another heat source to the mix.
"Get you facts first - then you can distort them as you please." -Mark Twain
Why are these cards not PCI Express? Most anyone who would buy one of these would have a motherboard with PCI Express slots.
If that was the case what you'd get was asynchronus logic. This has been done before, There have been many asynchronus digital computers. There was even an ARM compatable asynchronus CPU designed. Look up AMULET Project.
s ync.html
What I really think you are talking about is a system that on a macroscopic level is asynchronus, but on a microscopic level is synchronus (e.g. Each chip has it's own internal clock, or each part of the chip has its own clock; However there is no more "System" clock).
See http://www.cs.man.ac.uk/async/background/return_a
I wonder what this will mean to the people who are currently using GPUs for data parallel-ish computation? This is sort of a specialization of the GPGPU idea... I wonder if it would work well for doing "real" physics computation?
I'd rather be flying
Not necessarily true. While dedicated cards for physics haven't existed, dedicated cards for other operations have, and much of the physics calculations themselves are still being done in games, just with an extra load on the CPU in software rather than a dedicated unit. As physics becomes a bigger focus in the realism of 3d games, perhaps it is in fact a foreseeable evolutionary step that specific devices would exist to process this.
I really don't see a custom "Physics Processor" being a long-lived add-on for the PC platform. It's essentially just another floating point SIMD processor with specialized drivers for game engine physics. With multicore+hyperthreaded CPUs coming out very soon, the physics engines can be offloaded to extra processing units in your system rather than having to fork out money for a card that can only be used for a special purpose.
In addition, there's already a hideously powerful SIMD engine in most gaming systems loosely called "the video card". With the advent of DirectX 10 hardware which lets the card GPU write it's intermediate calculations back to main memory rather than forcing it all out to the frame buffer, a whole bunch of physics processing can suddenly be done through the GPU.
Lastly, the API to talk to these cards is single-vendor and proprietary. That's never been a long term solution for longevity (unless you're Microsoft), so it won't really take off until DirectX 11 or later integrates a DirectPhysics layer to allow multiple hardware vendors to compete without game devs having to write radically different code.
So, between multicore/hyperthreaded CPUs and DirectX10 or better GPUs with a proprietary API to the card... cute hardware but not a long term solution.
Fear: When you see B8 00 4C CD 21 and know what it means
And that's not even mentioning a lack of DRM. Doesn't Hollywood own gravity these days? I'm sure a patent was filed somewhere - or was it a copyright?
"It's the height of ridiculousness to say for those 9 lines you get hundreds of millions."
Nobody seems to care about good old fashioned fashion anymore. Everything has to be technically superior these days. The article was slashdotted, but for those who missed it , it wasn't worth reading, the best thing I can say is "Oh".
I like graphic cards that can make people say "we don't need them yet", because this is the attitude that gets their job outsourced to "offshore". Short sighted vision. Which is ironic when it comes to graphic cards.
much like that cat that ate cheese then crouched in front of the mouse hole...
I've abandoned my search for truth; now I'm just looking for some useful delusions.
Should we petition HotHardware to change their name to Not-So-HotHardware or maybe Lukewarm Hardware at best?
Looks like their servers exploded from the slashdoting.....
"We'll need 2000 crickets, 4 cans of Easy Cheese, and the fluid from 18 glowsticks for this plan to work...." - ph0n1c
it sounded something like *SWOOOOOOOOOOOOOOSH* ...
This card isn't exactly a DSP, it's not so much designed to process streaming signals. It's more like a vector processor optimized for physics calculations such as vector scaling, angular velocity/acceleration calculations, you know general calculations you would do in your low level college physics or mechanical engineering classes.
The processor can do very fast calculations on large vectors better(faster, more throughput I don't know, I didn't design it) than a CPU which is a GENERAL purpose processor. It can do nearly anything at the cost of efficiency. That is why graphics cards exist. The CPU can do the DirectX calculations, but a video card does them much faster by having a specialized architecture (vector shaders, pixel shaders, fixed function shaders, etc).
What Aegia did is made a software physics engine that can run on the CPU, but can do the same thing much faster and on a greater scale on their specialized card. So instead of calculating 20 fragments from a grenade it can calculate 100 (not exact numbers, just an example). In theory it can make games much more realistic. Imagine a rock slide with hundreds of rocks instead of just 3 small boulders falling at you or an water fall instead of a trickle of a few drops. You get the idea.
Doesn't look like a very good performance improvement for the money. In fact, CPU's new "dual-core" marketing push may just eat up the dollars for something like this. If you simply move your physics engine to hardware, it only solves 1 part of a larger, and very delicate puzzle.
http://www.amd.com/us-en/assets/content_type/Downl oadableAssets/So32v64-56k.wmv
Nice comparison concerning current 32-bit applications/limitations over 64-bit. If this video is TRUE, then I won't bother with a PPU - my Athlon 64 3000+ may already to be able to handle those extra physics calculations while any WELL-PROGRAMMED game will use any extra resources I have available for extra object/texture/physics rendering.
Sorry, IMHO, PPU is at a loss. Mod down at will.
Still waiting on Serviscope_minor to wake up to fucking reality and realize that Jessica Price isn't going to fuck him.
Your history of glide is a little backwards. OpenGL predates glide and was a clean rewrite of SGI's original graphics API with input from other graphics vendors at the time. The current graphics pipeline was a solved problem by the early '90s. The primary problem that glide solved was how to make a $30000 workstation into a $300 graphics card. The answer was to throw out most of the pipeline and make a passthrough card that didn't even do video. Glide itself wasn't anything particularly fancy and mostly consisted of functions to send untransformed polygons to the hardware. It took a few weeks for there to be a glide->OpenGL compatability layer.
3DFX simply failed to keep up with NVIDIA. They did an incredible job integrating both video cards and graphics engines together in the RIVA128 chipset as well as adding a basic lighting and transform pipeline to the hardware. They also did a much better job supporting the standard software APIS of OpenGL and DirectX. They still do a much better job with drivers than ATI.
Michael
Some day Slashdot will allow people to edit their posts for grammar and spelling, or perhaps there will be a Slashdot editor who knows grammar and spelling.
"so many waited with baited breath" should read "so many waited with bad breath"
maybe our fellows geeks at FermiLab can find some use to such physics engine to simulate things instead of runing all those expensives experiment ;^).
Imagine a Beowolf cluster of this things
maybe running Linux
???
profit
Now how can I put this to use on my homework!
Just as the FPU was once an addon separate from your CPU, I wonder how much longer this so-called PPU will remain an addon separate from your GPU. I imagine that soon we'll have graphics cards with built-in PPUs, or even GPUs with PPUs built into the die as we have FPUs built into CPUs now.
If you like what I've said here, and want to read more, go to http://www.krillrblog.com
Maybe in a year or two they'll develop the technology to an interesting level, but right now it seems a bit pathetic.
>> Dedicated cards? Probably. Dedicated computers? Definitely,
Either say "Dedicated cards? Probably not. Dedicated computers? Definitely" or "Dedicated cards? Probably. Dedicated computers? Definitely not".
---- "XML is like violence. If it doesn't fix the problem, you aren't using enough."
shifting functions away from the main CPU is not new, and adding a card or device to handle non-core functions is not new, so it's the specific functions that the card handles that are new(ish) Prior Art would include FPU chip, co/extra-processor cards (like PC cards for a Mac, Sidecar for Amiga), MPEG-2 cards, etc.
There was an unknown error in the submission.
I want to see how they will implement this in X11 or Xgl-type desktops. When My icons collide into each other, I want it done realistically! When I kill Firefox because it's frozen, I want to see it shatter into a million pieces! And then have those pieces push around the rest of my desktop.
This isn't serious, of course, but the reason I say this is I wonder if there are applications for things other than video games.
You can lead a horse to water, but you can't make it dissolve.
Site was slashdotted(haven't seen that in a while, hothardware must run on some pretty wimpy servers/net connections)...
In any event the Broussard interview comments sums up my thought on this PPU to a tee:
"And I still don't see people spending $200 for a PPU like they do for a GPU."
He's completely correct. To me the PPU has the same value as a sound card, say around $60 max, but preferably c. $25-$40. $259 if more than I have ever spent(or plan to) on ANY component in my systems other than, maybe, the CPU, and it had better be a DAMNED fast CPU for that price. (OK, I guess that I'd fork over >$250 for a GOOD CRT too... not an LCD though.) Geez, I even balk at spending over $100 for a mid-range GPU(my prior system ran for over 3y with a $25 3D GPU, which I "upgraded" before upgrading the entire system (for twice the original cost, $50) got some improvements, but the newer system... graphics look a little nicer but... I'd feel ripped if I paid $300(or more) for something that would only be slightly better for anything other than pointless bragging...). I just don't see enough value in GPUs to spend 100s of $s on them, so I DEFINITELY don't see the value in spending $250 on the PPU, especially when as the article continuously mentions there is very LITTLE hard info about exactly what their processor is, how it operates, it's speed, etc. And in their initial review they were unable to force running of the high quality physics in software alone. Personally, I'd really like to see that too, or with Havoks(?) engine using CPU + GPU v. the $250 card. (Of course another problem is that most games still only use one CPU, and AFAIK many have problems with multiple CPUs and/or hyperthreading, yes I know there are workarounds, but you end up witht he app running on ONE cpu.)
Still, I would have expected a bigger improvement in performance on existing stuff. There may be too much of a bottleneck getting in and out of the physics processor, which is the usual problem with coprocessors. I'd expect more improvement in fluids, particles, hair and cloth physics, which usually don't feed back into the gameplay engine and thus can be done concurrently with the main engine work. If you're banging boxes around, the main game engine probably has to wait for the physics engine to get the new box positions, so there's no big win there. Even if you have feedback to the game engine from cloth, you can probably delay it a cycle, so that when the cape gets caught in the door, it doesn't yank on the character until one cycle later.
What the point here with this processor? A dual-core (or even quad-core) with a dual-sli GPU isn't enough!! We need a Physics processing unit now. Man, do they ever invent ways to siphon more money from a fools wallet. $300, good lord that is a lot of money. Developer can't utilized the processors that are in the computer already for the extra physics. With questionable support in the near term and total lack of a competing product, I am not shelling out a thing for this.
You don't have to be smart to use a Mac, you just have to be smart enough to buy one
You guys do know what happened to the idea of separate "486" and "486sx" lines, right?
No, I don't see anyone trying to sell us "x86 math coprocessors" any more either.
-Styopa
How is framerate a measurement of the performance of a physics card?
Also the current limitations imposed by dx9 probably cause bottlenecks given the number of objects being returned by the PPU, i think vista will remove this bottleneck to a lrage extent.
A dual SLI GPU (not to mention quad) is a total waste of money...a huge CPU is a total waste of money too...people still buy them. Why not save some money on the 4 GPUs and huge CPU and invest it in a PPU?
Physics isnt eye candy. Theres alot of physics that CPUs are struggling with right now...and im not talking fluids or ragdolls...im talking collision detection and movement logic...thats all physics and a dedicated card can do it faster and more precisely. One think people dont realise is that physics has to be calculated at a MUCH higer refresh rate than the display frequency for good accuracy.
No, it also has an additional power connector - this is mentioned early in the article.
Build it, Drive it, Improve it! Hybridz.org
AMD has licensed the technology behind what is supposed to be a VERY fast FPU device. It can supposedly do the math faster than a quad core AMD CPU. Intel also attempted to get the rights but AMD apparently got them and with their Hypertransport is supposed to be in a better position to actually use it. Obviously not released yet but I believe there was an article posted here a month or so talking about it. So yeah, seperate CPUs CAN be an advantage for specialized tasks...
Also, DRC has come out with a programmable coprocessor with the ability to tie into the Hypertransport on AMD CPUs. It's not so much different from some of the other add-in cards that do the same thing except that this new one can get direct access to memory and the CPU through Hypertransport. Specialized CPUs are apparently far from dead!
Hrm, and adding this to a graphics card makes sense to me. Chicken and egg problem - no one wants it because there's no games and there's no games because there's no market yet. These cards have been sold in OEM computers by some manufacturers already thoguh so hopefully that will jumpstart things. Personally I want to see what UT2K7 will do with it - if it's used to good effect I *will* buy one...
Build it, Drive it, Improve it! Hybridz.org
That's an obvious application. Maybe Illusion will support Ageia in the successor to Battle Raper II. Illusion already has basic tearable cloth, but it's not physically simulated.
We've come so far since the 8087 Math Co-Processor. Who would have ever thought there'd be a PHYSICS Co-Processor, back then?
Saying this PPU is just a 'Math(s) Processor' is like saying a graphics card is just a rasterisor...its not true.
At its most basic it is just a FP calculator in the same way a GPU is just a chip...but in both cases its the surrounding stuff that makes them fast. Things like automatic optimisation schemes and specific enhancements to the architecture.
Take for example mipmapping or antialiasing or automatic shadow culling schemes...all designed around image theory and human perception...arent applicable to anything else other than image generation.
A PPU will have similar things that are based specifically on physics and will make it more than just a 'Math Co-Processor'.