John Carmack doesn't believe in dedicated PPUs

John Carmack has always been vocally opposed to the concept of dedicated Physics processors for gaming PCs, but his latest statement on the topic really hammers the nail in. Talking to Boot Daily on the topic of Intel's QX6850 CPU, Carmack states that future multiple core CPUs will be able to take the strain of physics processing. Carmack does have a fair degree of interest in this area, with Ageia's PhysX card being supported on the Unreal Engine 3 platform, which happens to be developed by iD software's rival, Epic Games. If there's one thing we can be sure to garner out of this statement, it's that iD software's new gaming platform iD Tech 5 won't be one that favors dedicated physics processors.

[Via Digg]

[Darren Tilley @ Jul 22nd 2007 8:33PM]

This is the same guy who doubted the use of multi-core processors in games programming. In fact, he dismissed both consoles for exactly that reason. Now all of a sudden, he's an evangelist for the use of multi-core processors in physics intensive games.

Remind me to ignore what John Carmack has to say about everything.

[Josh @ Jul 22nd 2007 8:58PM]

Actually he has often spoke out about much he likes the 360 because of the multi cores in it.

[Alexander @ Jul 22nd 2007 9:29PM]

@josh

But the fact was that he was opposed to it, then [chip company] gave him bunches of the dualies, and he suddenly thought the idea was fantastic.

Basically, JC is just like everybody else--easily swayed when freebies are thrown in the mix.

[Stranger @ Jul 23rd 2007 10:29AM]

That's a silly comment, already back in 1997, years before SMP became common, did Carmack began working on a multi-threaded game engine, i.e. what later turned out as the Quake III-engine.

[I LOVE THE CAPS LOCK KEY @ Jul 23rd 2007 1:32PM]

Is it just me or doesn't John Carmack look like Janet Reno?

[XSportSeeker @ Jul 22nd 2007 8:51PM]

If this article is 100% correct, John is just feeding bullshit to the press.
You don't have to be a rocket scientist to know that physics processors are the future.
Physics involves very complicated calculations.
CPU's are being put to lots of strain to make very simple physics math in games nowadays.
Maybe multiple core cpus are enough to handle physics in games today, but tomorrow it'll be another story.

As gaming evolves, and gamers expect to see more and more realistic effects on games, physics will also have to evolve to a point where using just CPUs won't be enough.

Try, for instance, relying only on multi-core cpus to make real time calculations on a particle system. There's a very low limit as to how much you can do with that.

[kahlzarg @ Jul 22nd 2007 9:21PM]

I think physics cards will be reserved for the ultra top end of gaming, and I doubt they are the future. (though only time will tell)

I agree physics strain on a CPU is one of the heavier processes in a game, unified video architecture will allow that load to move onto the video card and I belive it to be more likely that a video card/physics card system will go the way of the videocards/daughter 3D system of the 3DFX days.

I think off the CPU physics is here to stay, but Ii find it alot more likely that it will be integrated into Hi-End 3D cards (and then filter down to a standard like AA, HT&L, HDR & shaders are today)

As we have sen with cards like the GeForce 7950 GX2, SLI on one card, and unified architecture of DX10, I think that we will find that 3D cards will eventually move into multicore solutions.

Even from a commercial POV, should dedicated Physics cards take off, you'll see them on NVidia's and ATI's boards by the next product cycle, on board, at a cost that is less expendive than a dedicated solution.

On a related note, I think that the next step in video game physics isnt more FX, but better replication, allowing your physics interations to appears across all clients on a server in the same way, as most of you know once an object is released to the physics system what is displayed differs based on the physics settings of each individual player.

[Victor @ Jul 23rd 2007 9:54AM]

But he is a rocket scientist...

[Hagrun @ Aug 29th 2007 3:40PM]

However because processors will also advance in tech (always ahead of software) then there is no telling that PPUs will actually ever become useful. Right now they aren't worth the money because nothing uses it, but even today most games aren't multithreaded. So why not use those other cores of the processor for physics instead of making your customers go out and buy more hardware to get the best possible gaming experience?

I'm not saying that you are incorrect, it's just something else to consider.

[Bob @ Jul 22nd 2007 9:02PM]

I really don't agree or disagree but I have to ask, is Carmack even still relevant? He seems to pop up every once and a while to talk trash about some new tech.

[icepop4who @ Jul 22nd 2007 9:20PM]

Future multi-core CPU will also overtake GPUs, causing the formation of InVidia, the child of intel and nvidia.

Because soon, there will more cores on a cpu than keys on a keyboard. QWERTY-core. Sweet.

[Andrew @ Jul 22nd 2007 9:25PM]

what a goober

[Josh @ Jul 22nd 2007 9:43PM]

If we are using the term 'PPU' to refer to a stand alone packaged chip, then I also doubt that they will become common.

What I expect we could see is an increased focus on providing SIMD instructions that can perform 5x5 matrix ops in very few clock cycles. The chip makers in the best position to do this are ATI and NVIDIA.

From what I understand, the heart of todays PPU's is the acceleration of operations on 5x5 matrices. Today's GPU's can already accelerate 4x4 matrix ops. I dont think NVIDIA, ATI, (or for that matter, Intel) will wish to see another chip maker walk into the market.

- Game software remains the driving force behind GPU innovation
- Physics simulation remains a driving need in game software
- Much of the infrastructure for performing PPU functions exists in todays GPU's.
- It is too easy for ATI and NVIDIA to gobble up PPU functionality into their existing product lines.
- It is too easy for Intel to gobble up the building blocks of PPU functionality into their generic cores as SIMD instructions.
- It will be too easy

Unles PPU's start becoming highly specialized beyond fast matrix ops and custom registers, I think they are destined to go the way of "math coprocessors" (remember those?)

[Whoa... @ Jul 22nd 2007 9:50PM]

...that dude looks like a chipmunk

[Mitch @ Jul 22nd 2007 9:51PM]

Kay I'm sorry, but this is news why? Slow sundays....

[Josh @ Jul 22nd 2007 9:56PM]

Oops - dont know why my post was indented as a reply XSportSeeker's post. And looks like I could do a better job at editing before posting. Sorry. :)

[Brad @ Jul 22nd 2007 10:42PM]

I won't comment on PPUs, but I will say something about John Carmack.

To all the haters in the posts above:

Maybe you aren't familiar with him, or his work.. let me enlighten you..

-Commander Keen
-Wolfenstein 3D
-Doom (And sequels)
-Quake (And sequels)

Besides being the lead programmer on those items, his game engines have been used for the following games..

- Half-Life
- Medal of Honor
- Call of Duty
- Soldier of Fortune II
- Return to Castle Wolfenstein
etc, etc, etc...

If you need any other proof that he is worth listening to, see his Wikipedia entry @ http://en.wikipedia.org/wiki/John_Carmack#Recognition

"If you aren't sure which way to do something, do it both ways and see which works better." - John Carmack

[sentient.exe @ Jul 22nd 2007 10:52PM]

Seriously, it's John Muthafukin Carmack peeps; show some respect.

[strider_mt2k @ Jul 23rd 2007 7:08AM]

Can we get him on "Enemy Territory: Quake Wars" then too?

That shit is taking FOREVER!!

[nick @ Jul 22nd 2007 10:47PM]

When you ask if John Carmack is still relevant, are you sure you aren't thinking of John Romero?

Carmack is responsible for Commander Keen, Wolfenstein 3D, Doom, and Quake. All games that defined and lead their genre at the time of release. You may not have heard much about him lately, because he's been working on a new engine for a new property.

[Mike @ Jul 22nd 2007 11:26PM]

If memory services Carmack didn't say that multi-core CPU's were "BAD" for games... just that it's much easier to develop for one fast core since it saves coders the hassle of dealing with the hell that is threading and everything that comes with it.

[Mike @ Jul 22nd 2007 11:26PM]

Here's what Carmack said at QuakeCon 2005 regarding multi-core CPUs and Physics...

[He] expressed skepticism about the returns of multi-core designs and multithreaded software, especially in the short term. [...]

Part of the problem with multithreading, argued Carmack, is knowing how to use the power of additional CPU cores to enhance the game experience. A.I., can be effective when very simple, as some of the first Doom logic was. It was less than a page of code, but players ascribed complex behaviors and motivations to the bad guys. However, more complex A.I. seems hard to improve to the point where it really changes the game. More physics detail, meanwhile, threatens to make games too fragile as interactions in the game world become more complex.

http://techreport.com/etc/2005q3/carmack-quakecon/index.x?pg=1

[mrnoface @ Jul 23rd 2007 12:11AM]

I think what he's saying is correct. Even if you put a physics card into your system, you still have to upgrade your video card (and even CPU too) to see some real effects without fps dropping because there will be more particles and post effects to be rendered, due to the PPU which creates more of those.

IMO, multi-core CPUs and multi-functional GPUs are way to go. Aren't there already some GPUs that support physic calculation?

[Taylor @ Jul 23rd 2007 12:16AM]

I personally think that using up a whole (precious) PCI-E slot to physics processing isn't worth it. Perhaps in the future we will see standard GFX cards (even consumer lines akin to the Geforce 7900GS or the 8600) with integrated physics circuits.

Besides, in all the FPSes I've played on my 7900GS (and on the 7300LEs at school), the physics have been fine - so long as I can shoot someone over a balcony, or see their body ragdoll when they get naded.

[Negativecool @ Jul 23rd 2007 12:55AM]

I respect John quite a bit given his sizable importance to the industry. But everything that has come out of John's mouth lately---since the new console generation has come---has lead me question if he is falling behind the industry he helped build. While Epic, Id's "rival", seems to be at the forfront at this point. The UE3 engine is dominating right now. It tramples the quake/doom engine and fully utilizes and embraces multicore processing. (although, yes, we did get a quick peek at his latest game engine, but the general feeling I got was that it looked good but does not go beyond anything already out there)

From a purly gaming prespective PPU's HAVE to be the future. The only problem is that the majority of the populace, using PC's only for chat/e-mail/youtube/iTunes/office software programs/or what have you, will have NO need for it.

John represents the gaming industry, so why he would speak out against PPU's is beyond me. He talks as if he is representing the average consumer who would have no need for advanced physics processing.

[Semi-Evolved @ Jul 24th 2007 12:20AM]

Why do they have to be the future? Why not GPU or CPU integrated matrix circuitry, and even more pervasively multicore CPUs?

Carmack speaks out about them because he can see the other trends in the industry. Like the trend towards heavily multi-cored CPUs. They completely obliviate the need for dedicated cards of the sort that Ageia peddle.

As for the UE3 vs D3/Q4 engine, of course the UE3's dominating right now. Doom 3 was released in 2004 for crying out loud; the first UE3 game was released over two years after. There's only so much one can do with any given engine before serious rewrites are needed to keep up.

[Mr. Picklesworth @ Jul 23rd 2007 1:21AM]

Lucky for him, he is a rocket scientist :P
www.armadilloaerospace.com

[Mr. Picklesworth @ Jul 23rd 2007 1:22AM]

Damnit, this replying stuff is never working for me!
That was in response to XSportSeeker, who said:
"You don't have to be a rocket scientist to know that physics processors are the future.
Physics involves very complicated calculations."

[Mr. Picklesworth @ Jul 23rd 2007 1:21AM]

I'm siding with Carmack on this one. The whole point of multi-core processors is to not have to distribute these jobs to specific hardware, but to instead keep everything centralized under one type of hardware, with the specializing left up to software.
Hardware support is hell as it is. Adding yet another component to the mix would be a pain for everyone but the hardware sellers (who would be happily rolling in their money generated from false need).

On the other hand, keeping it to a single type of processing, the system is easier to manage. Expandable and adaptable processors (such as, dare I suggest, Cell) are the future here, not yet more specialized equipment. The video card and sound card exist because they output to those types of external hardware, but a PPU does not output to anything other than the processor. Therefore, what the hell is the point?

To add to it, I wouldn't mind PPUs as a choice (for those people who just go with what exists instead of waiting for Hell to freeze over) if they were not sold in such a way that, without them, you get arbitrarily locked out of features. There isn't a huge amount in the PPU scene, so the software development is very closely linked to the idea of selling hardware.

[bill @ Jul 23rd 2007 1:22AM]

I cant believe my eyes. Are there actually people out there that don't take the word of Carmack seriously. Thats like god speaking to you and your replying "THAT DUDE LOOKS LIKE A CHIPMUNK!" Are you kidding me, have some respect for your damn deity.

Seriously though, this man has done so much for the gaming community that its inconceivable to not at least acknowledge his opinions and claims as valid. This is the same guy that singly handedly forced the PC gaming world into 3d acceleration. The same guy that introduced per-pixel lighting (and at the same time the need for a flashlight) into realtime engines. The same guy whos company literally birthed the FPS genera. How the hell does your opinion even begin to sway the scales against a man as accomplished as John Carmack, chipmunk hater?

As to why games like UT are using PPUs. Well thats a simple answer. There are no other options at present. Nvidia has already put its chips in the pool and begun teh ground work for physics calculation on their GPUs and if you read any of the intel and AMD roadmaps (but you didnt did you?)you would know that there will be more then enough cores on their processors to handle to workload.

Carmack writes code not speeches. He has often been called stiff or the like. If I were to attempt a guess at what he means by this one. I would say that he feels that PPUs are filling a gap that is needed at present but will be made irrelevant in the near future.


Ohhhh yea just for your info in his spare time between developing cutting edge grafx engines, he is on one of the teams developing the next lunar lander. Take that you chipmunk hating sons a bitches.

[iceman2cool_92 @ Jul 23rd 2007 2:15AM]

Well said !

[dspasic @ Jul 23rd 2007 3:06AM]

bill, r u John Carmack or his wife.

[rudebo @ Jul 23rd 2007 5:19AM]

What is this politics ?
So the guy had a view that didnt turn out to be true now he's not credible ? Bollocks.

I really appreciate his policies towards this. ID games have always been supported on multiple platforms and he has always strided to keep things flexible - so you dont have to buy an exact product for it to work. I think he's honest in his intrest to further the field while keeping things multi-purpose and usefull and not turn everything into products that just run one thing.

my 2c

[John from Buffalo @ Jul 23rd 2007 6:45AM]

John C. is no longer relevant. He is SOOOO 1999.

[Peter @ Jul 23rd 2007 8:16AM]

Geez, I though Carmack was merely stating the obvious here. PPU are pretty much a complete non starter. I laughed when I saw dedicated physics cards. They have been tested as basically useless, adding a few particle to explosions. I have a hard time believing there a people who think this is going to take off. Just because Some engines support this doesn't mean they have strong belief either, more likely they are using a 3rd party physics library, that supports it.

As far as Carmack relevance. Well, gamers are a fickle lot. While Carmacks games are no longer #1, the engines are still top notch, but more importantly has been Carmacks pioneering work in developing the 3d gaming as a whole, basically inventing/popularizing most of the techniques and NOT PATENTING them. Things like Binary Space partitioning, Raycasting, Surface Caching, or the Carmack reverse.

He doesn't patent anything and after a few years usually gives away his game source code.

Just because his games are no longer #1 doesn't mean he got stupid as games are half art/half science and Carmack is the science side. Among programmers Carmack still gets and deserves great respect. We are not such a fickle lot and we recognize rare talent when we see it in our field.

[Magallanes @ Jul 23rd 2007 8:44AM]

Today PPU give a "boost" of a mere 5-10% and even in some cases CPU-only based is more fast that a specific PPU.

To create a game and program for a videocard is really a pain-ass work, mainly because some incompatiblities into different kind of videocard. And if you add PPU.. then you can expected more incompatiblities/troubles.

[Xzavier @ Jul 23rd 2007 9:42AM]

Bear with me here on this one. I have to search inside the attic!

This sort of reminds me of the old 8086, 286, maybe 386 with the FPU housed inside another processor, the co-processor. What happen to the 2, they got married! Of course before they got married their was a big debate whether or not a person will ever use the FPU processor. Most people agreed "at that time" that there is almost no need for a separate FPU processor. The PC “at that time” had more power than the average person needs. However people had the OPTION to buy 1 or both.

Then I think around the early/mid 90's or so, “I really can’t remember”. I started to read about processor with dual cores and multiple cores. Another big debate broke out... The popular OS's at that time, and most applications are only written for single/processor. Not for multiple processors or a single processor with multiple cores. Never the less, I thought those things would NEVER arrive, however finally here they are!

I am not going to go up against John Carmack and I think EVERYBODY should listen to him, however if he states something, he's not just blowing wind out of his arse in any way! He is writing the code/engine/game and we are just simply playing it. Remember that!

To me this looks like the normal debate of the "evolution of technology" if you can call it that!

Only time will tell where those PPU processor will go!

[bob @ Jul 23rd 2007 10:19AM]

Once PPUs adopt Steorn Orbo technology, I think they'll really begin to take off.

[MosquitoControl @ Jul 23rd 2007 1:15PM]

Dedicated physics chips will never sell.

Right now they're expensive. For what?
When 3D accelerators hit the scene people knew what they were getting. They'd open the new issue of CGW or PCG and see screenshots comparing Quake to vQuake to glQuake. They'd see the remarkable difference. They'd see transparent water.

Physics can't be seen in a screenshot. Hell, it's hardly noticeable in a video, particularly grainy youtube ones.

Beyond that, 3D chips didn't really take off until they started incorporating 2D. Suddenly it made sense to have one, as it had a dual purpose, one that you needed, anyway. Physics chips are exclusive for gaming.

And they cost as much as several games.



Never going to catch on. Not unless they come into double digit prices.

[Kahlzarg @ Jul 27th 2007 6:35AM]

Here is a good article that those subscribed to this thread might find interesting.

http://www.anandtech.com/video/showdoc.aspx?i=3048

Covers some good points made by both sides of the discussion.