John Carmack: "I am not a believer in dedicated PPUs."

[The Continental]

CustomPC (http://www.custompc.co.uk/news/601119/john-carmack-reckons-physx-is-useless.html) by way of Boot Daily (http://www.bootdaily.com/index.php?option=com_content&task=view&id=649&Itemid=56&limit=1&limitstart=1) snagged a choice quote from the champion of all things rendering tech about the often hyped, seldom delivering dedicated physics processors that have shown up in the last year or two. What the QX6850 really does is help galvanize Intel as the preeminent leader in quad-core technology and should help bolster developer support as we see the industry move towards more CPU cores. id Software's leading guy, John Carmack, told us this regarding his feelings of CPU cores and physics:

“I am not a believer in dedicated PPUs. Multiple CPU cores will be much more useful in general, but when GPUs finally get reasonably fine grained context switching and scheduling, some tasks will work well there.”He's a heart breaker, dream maker...

[Deadend]

So, the PPUs are pointless as they have yet to do anything that can affect game play? Or who knows. But if Carmack does not like it, odds are good it sucks.

[alienchild]

Agreed, now bring us some AI cards instead.

[Scramble]

All these extra dedicated cards are bullshit. We now have multi-core CPUs. Those extra cores can be used to process physics, AI, extra graphics etc.
Since I've had my dual core, I haven't noticed say, Counter Strike Source running any better than it did on my 2ghz single core,
but they could use the extra core to help render more realistic physics and particles. Developers need to work with this technology that's very quickly becoming standard.

[Hatchetforce]

My friend John Sonedecker built some demos for Aegia. I have one of those cards and have used it in a couple of games. Too bad those titles (GRAW PC) were less than eye popping. But the facts are until someone has a stroke of brilliance and comes up with the software, the cards offer an unparalleled game experience as regards physics. People can knock them if they want and yes they are another expense to toss on the burning money pile, but the damn things work and they work well. There is a vast difference when they are running in a game and when they are not. I like the Carmack but he isn't ALWAYS right, just most of the time.

[DarkDaY]

My friend John Sonedecker built some demos for Aegia. I have one of those cards and have used it in a couple of games. Too bad those titles (GRAW PC) were less than eye popping. But the facts are until someone has a stroke of brilliance and comes up with the software, the cards offer an unparalleled game experience as regards physics. People can knock them if they want and yes they are another expense to toss on the burning money pile, but the damn things work and they work well. There is a vast difference when they are running in a game and when they are not. I like the Carmack but he isn't ALWAYS right, just most of the time.


he might not be always right, but im p r o b a b l y going to take his opinion over a n y o n e elses, especially someone in a forum, no personal offence, but he has proven time and again no matter what anyone says, espeacially about the doom 3 engine.

far as Im concerned, he is still the end all be all when it comes to the truth.
or the best source of it that we trolls can get.

[Qoz]

#Scramble

And they do.. but CS:Source is really old and run extremely well even on a singlecore so why update? Multicore-coding is hard to do well. The code produced now-a-days is hardly 100% effecient use of the cores and it is often hardcoded as an afterthought. This means the usage of cores will not scale if more are present and the exra FPS gain is minimal.

Consoles and PC's will all have more cores in the futute. Dual-core is a standard now when purchasing a CPU. Quad will be a standard purchase in ½-1 year. So developers are learning/coding right now.

Games take many years to create, and multicore-support in the engine is a fundamental design-choice to be taken early. We have therefore not seen many games using our dual/quad cores, but they will start to come real soon I'm sure. Alan Wake is optimized for multiple core for example. I think Crysis is too. The Unreal3 engine should be, and that is used in many fortcoming games.

Regarding the thread then he is right. AGEIA is in REALLY tough spot right now with all the Quad/Duo cores being so cheap. Many developers have made the choice to use an extra core for physics, because they have this extra power they do not use, and many people (more than AGEIA owners) have these extra cores. They are sitting with their game in development (engine on 1 core), so they use some ressources to make physics run on a seperate core instead of supporting AGEIA. Or do both.

I think the gamers who would be willing to purchase AGEIA, will buy a Quad-core instead. AGEIA chose the wrong time to introduce a physics-addon card. If they want to survive they need capital and kick-ass 2nd generation card with some must-have effects (real simulated water/bending etc.) and a killer-title to use it.

[MrWonderstuff]

I havent a problem with a PPU, just the cost. Additionally the uptake from developers has been slow with only a handful of games adopting it. Now if NVidia bolted on a PPU onto their next generation of cards...well hmm.

[JCtheMC]

Did we need Carmack to tell us this? Isn't this a widely known fact?

Dedicated physics cards were a stupid idea to begin with. Good luck Ageia.

[claws]

The idea of the cards wouldn't be so bad if the performance wasn't so awful. They're a good order of magnitude slower than they need to be, and as such they'll just get murdered.

[Esquilax1138]

+5 for the Pat Benatar reference!

As for the physics cards, their time was in the 90's when multi core CPU's were a dream for desktops. Now with multi core CPU's and video cards being even more powerful than the CPU's, put that all togther and you have no need for another add in card.

[Ancalagon]

Carmack knows lots and lots of stuff about graphics engines, opengl, games, and 3d graphics engines using opengl in games.

not much else unfortunately. physics in any ID game so far have been a joke.

that being said, he is kinda right, in that many more people have more than 1 core, not so many have ppus. and an extra core can be used for more than just physics.

for ageia to win, I think they need better board deals, lower prices, and most importantly more games. like lots more games. like 50% of new titles need to be physx compatible.

they should even consider integrating the ppu onto high end motherboards, or onto high end soundcards. a £100 sound card that boasts physics performance and sound performance? in short, can take a whole load off the cpu.

the other problem is, having not released a new card in a while, how powerful are these cards? if they dont release new cards often, the extra cpu cores will be faster, cheaper and more widely used (which is already true I think).

[JCtheMC]

any ID game so far [has] been a joke.
Edited the quote for truth and win.

Ps. I don't like Carmack. He hasn't made a decent game in over a decade, yet people still treat him like he walks on water.

[apethedog]

John Carmack *invented* many of the techniques used in 3D games today. How to use BSP trees to make your levels show up properly, how to light areas. He came up with all of these techniques, and many more. Quake was the first 3D engine that belongs in the category of modern games. Every single other 3D game that came after it is based on the work he did.

You could fault Edison, and say that the lamps he made aren't all that useful when compared to the LED's we use today, but that doesn't mean he doesn't deserve respect for inventing it in the first place.

(Ps: Edison did not *actually* invent the light bulb - this is a myth, but, for the sake of arguing, let's say he did, as it's a good example)

[Schnoogs]

(Ps: Edison did not *actually* invent the light bulb - this is a myth, but, for the sake of arguing, let's say he did, as it's a good example)

It was design that actually worked...hence the credit.

Talk about a sad attempt at looking smart.

[apethedog]

I didn't know he didn't invent the light bulb until, 3 days ago, it was mentioned in a documentary. Hence it still being fresh in my mind. I don't care enough about trivial facts like that to do fact-checking.

I do not attempt to look smart - I'm actually pretty confident in my own intelligence. I genuinely don't feel a need to prove anything. To do so, or feel that need, would betray a level of uncertainty.

I just felt that John Carmack was being unfairly judged on criteria which aren't relevant.

[Mojopin]

Talk about a sad attempt at looking smart.
At least you achieved in sounding like an asshole... ya?

[kraemer]

What the f#ck ever...
If you look back, its history repeating itself. The Carmack totally "poo-poo'd" multicore CPU's. (look for a story and thread here on EA)
Multicore sure hasn't hurt, looking at whats come out on 360 and PS3.

Like I said before, he isn't Jesus. He just hasn't gotten the memo apparently.

[kraemer]

Furthermore
GPU's are NOT GPU's any more anyways. They are large arrays of stream processors. Go look at an analysis of the latest Nvidia Geforce 8 series. Its not a "3d chip" at all. The same with a PPU. Its simply a chip with math functions that are specific to physics computations. If you look at whats going on with AMD/ATI their new architecture lends itself to having "other cores" with specific specialties, like say stream processors or physics. So it wouldn't be unreasonable to assume that physics function cores will end up in the CPU or GPU despite what Carmack thinks.

[Schnoogs]

At least you achieved in sounding like an asshole... ya?

Go play in the street please.

[51|RandoM]

What the f#ck ever...
If you look back, its history repeating itself. The Carmack totally "poo-poo'd" multicore CPU's. (look for a story and thread here on EA)
Multicore sure hasn't hurt, looking at whats come out on 360 and PS3.


Multicore CPUs deserve to be poo-poo'd. The only reason they're in favor is because it isn't possible to continue to make increasingly faster single core chips at the rate the industry has previously.

People have this misguided notion that multicore is an instant win. SMP systems have been around for decades, yet good multi-threaded code is a RARITY. Creating and debugging it is hard work, very hard work, of an often tedious nature.

The PS3 and 360 are invalid comparisons. Console devs don't have a choice there, it is hard to make a competitive next-gen game without using the multiple processor architecture present within each.

PC software, on the other hand, you don't have to write for multiple cores to be competitive. Matter of fact, considering the large install base of single core machines out there, you're kind of shooting yourself in the foot if you write code that only runs well on a multicore machine.

John Carmack is responsible for one of the two titles that compelled people to purchase a GPU.

Somebody should do the same for the PPU, if possible. Until then if you have one in your computer it is just an electronic appendix.

Shame we lost this thread in the server move. We'd already hashed all of this out and determined the only people vigorously defending the PPU were those who'd already purchased one.

[Schnoogs]

People have this misguided notion that multicore is an instant win..


All those servers with thousands of processors are totally wasting their time ;)

Random you may not know this but Windows, Linux and MacOS are multithreaded themselves. When you launch an application each application is itself a process that can be swapped on and off of the processor. So in other words who gives a shit if games and multithreaded since every modern user has a dozen or so apps running at any given time and on single core they have to be swapped on and off by the OS.

People need to understand that multicores will immediately benefit anyone with a modern OS.

[Sir Kodiak]

SMP systems have been around for decades, yet good multi-threaded code is a RARITY. Creating and debugging it is hard work, very hard work, of an often tedious nature.

Most applications on a computer don't have to be multi-threaded for multiple cores to be useful, as Schnoogs points out. You only need multi-threading in applications with high performance requirements, such as the operating system, graphics applications, and games. Everything else can be satisfied by the performance increase that comes from less competition for resources.

Writing good multi-threaded code is hard, I admit, but language support and developer tools have improved, so the situation isn't as bad as it was decades ago. Additionally, writing for SMP systems is much harder than writing for multi-core systems as the communication costs between processors is significantly higher than between cores.

PC software, on the other hand, you don't have to write for multiple cores to be competitive. Matter of fact, considering the large install base of single core machines out there, you're kind of shooting yourself in the foot if you write code that only runs well on a multicore machine.

Good multi-threading should have a minimal speed cost when running on a single-core system and may actually still result in a speed increase due to less problems with blocking. The question is more whether it's worth the extra program complexity when only some users see a significant speed boost. Still, with the need for middleware to support the 360 and PS3, you're going to see a lot of PC games that support multiple cores on the PC. And with Intel and AMD both pushing multi-core CPUs, more and more people are going to see the benefits.

[apethedog]

Hmm, strange. I can't imagine why he would have said multicore CPU's wouldn't be useful.

The thing is, I suppose once physics cards become mainstream, games are guaranteed to start using them. But as long as nobody owns one, what those physics cards do can still perfectly well be done by the main CPU.

Once a killer game comes out that requires their use, people are going to buy those cards - and all other games are going to need to compete with said killer game. By offering the same features. And that will be the birth of them.

An idea...

What if Microsoft made a "gaming for windows" card, which included, pretty much, an X-box 360 (without the CPU on it, obviously). They would probably not even cost marginally more than 3D cards do, today, and you'd be guaranteed/capable of playing every Xbox 360 game being made.

I would probably buy one of those over an actual Xbox, just for the comfort of being able to play on my desktop, and getting to use my nice PC screen.