Review Index:
Feedback

John Carmack Interview: GPU Race, Intel Graphics, Ray Tracing, Voxels and more!

Carmack Speaks

Last week we were in Dallas, Texas covering Quakecon 2011 as well as hosting our very own PC Perspective Hardware Workshop.  While we had over 1100 attendees at the event and had a blast judging the case mod contest, one of the highlights of the event is always getting to sit down with John Carmack and pick his brain about topics of interest.  We got about 30 minutes of John's time over the weekend and pestered him with questions about the GPU hardware race, how Intel's intergrated graphics (and AMD Fusion) fit in the future of PCs, the continuing debate about ray tracing, rasterization, voxels and infinite detail engines, key technologies for PC gamers like multi-display engines and a lot more!

One of our most read articles of all time was our previous interview with Carmack that focused a lot more on the ray tracing and rasterization debate.  If you never read that, much of it is still very relevant today and is worth reading over. 

View Full Size

This year though John has come full circle on several things including ray tracing, GPGPU workloads and even the advantages that console hardware has over PC gaming hardware.

Continue reading to see the full video interview and our highlights from it!!

Note: Due to popular demand, we are going to have a transcript ready (likely tomorrow) as the third and 4th page of this article.

First up on our list of topics was the importance of mathematics in the world of game and game engine design. Carmack stated that though his math background was a bit overstated in some cases he was able to build the engines we all know and love with a basic, applicable knowledge of high school topics such as geometry and calculus. The key is knowing how to apply these things to a problem that you haven’t seen before as opposed to being able to answer a question on a test. Of course, as engines have developed into more and more complicated pieces of software the need for higher level mathematics are required for physical simulations though you can still solve much of the world’s problems (at least in a software sense) by time-slicing and iterating.

The topic of the GPU hardware race came up early in our talk and the response Carmack gave us was pretty interesting. Stating “I don’t worry about the GPU hardware at all, I worry about the drivers” seemed to be a reiterated point. This became very apparent to id Software while developing RAGE where even though the PC had truly an order of magnitude more horsepower than the consoles, it struggled to keep up with the “minimum latency”, get feedback here, update data there, etc and do it all to maintain a 60 Hertz frame rate. DirectX 11 and multi-threaded drivers might have helped things but he still claims that they are far from the solution he envisions: direct surfacing of the memory system. The process of updating a textures on the PC is on the order of “tens of thousands of times slower” than on the Xbox 360 and PS3. AMD did implement a “multi-texture” update specifically for id Tech 5 which should help, but from the interview you can tell that Carmack really does want more done on this topic.

View Full Size

One interesting side effect of this talk – Intel’s integrated graphics actually has impressed Carmack quite a bit and the shared memory address space could potentially fix much of this issue. AMD’s Fusion architecture, seen in the Llano APU and upcoming Trinity design, would also fit into the same mold here. He calls it “almost a forgone conclusion” that eventually this type of architecture is going to be the dominant force. You might remember our discussion of this topic with Josh’s analysis of AMD’s Fusion System Architecture – it would appear that AMD has a potential ally on its side if they are paying attention.

Carmack still thinks the Intel integrated graphics is on feature parity with other integrated options and that people are going to be surprised in the not-too-distant future when the “free” graphics you get with your Intel CPU is good enough to play pretty much any game you want. He admits of course that the software and driver implementations from Intel need a lot of work and he has “high hopes” that with the shared memory potential there Intel will push forward with this “closer to the metal” mentality. There is also the outlet for console developers to more directly develop for integrated graphics than for discrete graphics (as it would be more similar to the console architectures) and games might run faster on integrated than low cost discrete solutions.

View Full Size

Larrabee was discussed as well – and though a couple of years ago it was thought this might be the “sweep” architecture across all the coming generation of consoles, it clearly didn’t meet the performance requirements to be successful yet. Instead, it seems obvious that there are again going to be multiple architectures on the pending console designs and there could be “strong contenders” based around the ARM architecture. But any next-generation console NEEDS to be a 64-bit architecture and with ARM just now integrating 64-bit designs for the first time, it leads us to believe we are a couple of steps away from seeing ARM in your next gaming console.

Carmack does hope that Sony avoids the Cell architecture all together due to the difficulty in development.

August 13, 2011 | 09:27 AM - Posted by Anonymous (not verified)

I thought the interviewer saying "OK" "OK" "Right" "Right" constantly was really annoying.

August 13, 2011 | 09:48 AM - Posted by Anonymous (not verified)

OK Right

August 13, 2011 | 10:16 AM - Posted by Ryan Shrout

Mm hmm...

August 13, 2011 | 09:52 AM - Posted by Anonymous (not verified)

Thanks heaps for the transcript! I really appreciate your effort.

August 13, 2011 | 07:10 PM - Posted by Tim Verry

You're welcome, I hope it helped!

August 13, 2011 | 11:56 AM - Posted by pdjblum

Ryan, not intending to mean or insulting, but so I might not feel any more of an idiot that I already am, did you really instantaneously get what he was saying? If you did, you are incredibly knowledgeable in this area of expertise. Wish I got half of it. I will listen again, but doubt that will improve my comprehension. Certainly doesn't take much to get John talking, and I guess in consideration of time constraints, you were regularly saying "OK" and "right" to move him along and get to the next question. Personally, though I really liked the range of questions, I would have been fine listening to him answer one question in the time allowed. Anyway, good job.

August 13, 2011 | 07:53 PM - Posted by Ryan Shrout

You are correct - part of the reason I was saying those things was attempting to find a break point to interject and get some more questions in. Obviously I failed.

Also, as for how much of it I understood, I think I got about 75% as we went - I don't think I am near the level of many other technical writers out there but having dealt with this for 11 years I picked up quite a bit.

August 13, 2011 | 08:49 PM - Posted by Tim Verry

lol, now you know how I feel trying to get a word in edgewise with Katy! :P

August 13, 2011 | 09:32 PM - Posted by pdjblum

I assure you, understanding 75% is quite impressive. As far as knowing as much as other technical writers, you bring a lot more to the table that is more important to what you do.

August 13, 2011 | 11:58 AM - Posted by Anonymous (not verified)

I really, REALLY want Carmack's q2 shirt.

August 13, 2011 | 03:14 PM - Posted by WilhelmR (not verified)

Transcript = awesomeness.

Thanks guys :)

August 13, 2011 | 04:41 PM - Posted by Rahul (not verified)

Good interview and good diverse (and interesting) questions!
Listening to Carmack is always interesting. Thanks for doing this!

August 13, 2011 | 04:49 PM - Posted by Marc Roussel (not verified)

http://www.youtube.com/watch?v=oB1eMC9Jdsw&feature=player_embedded#at=77

:)

August 13, 2011 | 05:07 PM - Posted by Anonymous (not verified)

This interview is the exact opposite of an interview with say Robert Dinero. With him you could bearly get him to say anything other than yes and no to your questions so the time was filled with the interviewer trying to fill the space by talking. Carmack is an interviewers dream. You ask one question and sit back for 20 minutes a let him go. That being said, he is obviously brilliant. Notice for instance he rarely says "um".... or "right"... or any other things that allow you time to think stuff out, it just flowes out perfectly.

August 13, 2011 | 07:13 PM - Posted by Tim Verry

Yeah, he is a talker, that's for sure! He always has a lot of interesting things to say.

August 17, 2011 | 11:39 AM - Posted by Brad Hollister (not verified)

Not to belittle Mr. Carmack in anyway, but most of his interview answers were really at a high-level / project manager style. This is nothing a smooth talking software salesman couldn't pull off. And for the lack of "um..." or "right...", all that suggests is that he has confidence and has talked on these topics many times before (probably everyday at id). In fact, give a listen to some very brilliant people like Dr. Robert Shiller (Yale) or Dr. Summers (former president of Harvard) and you'll find many "um..." and pauses, which appear to be signs of measured speaking and reflection. Mr. Carmack definitely has been a very successful software engineer which is in part due to proper timing (right code for the right market time).

More importantly, we should all be discussing the content of what he said, not the style or the judgement of the man. I found the hybrid ray-tracer (ray casting for only specular reflections in the context of a rasterized pipeline) quite interesting but superficial in presentation. However, the unnecessary ratio analysis of developer time / rasterized based shaders versus developer time / x86 cores (pertaining to the offline raytracing done for textures) not to be completely necessary - at least in the way he presented it (again nothing more than a stock analyst could pull off).

August 14, 2011 | 06:11 PM - Posted by Anonymous (not verified)

Have anyone seen the CentiLeo GPU ray tracer? They show virtual memory access working very well on consumer GPU, it was presented on Siggraph very impressively. Here are some first news about them http://gpgpu.org/2011/08/04/centileo-out-of-core-ray-tracer

August 14, 2011 | 06:23 PM - Posted by Cyco-Dude (not verified)

well, it's about as i figured: as these games get more and more complex, it gets harder and harder for the modding community. with quake 3, even i could make a map. now, you'll need a very talented team to do that. it's a real shame, and a good reason why i'll bet this game disappears 1.5 years after release.

i'm still waiting for a true successor to quake 3...i wonder if those days are over?

August 14, 2011 | 06:43 PM - Posted by FlowSim_geek (not verified)

I am pretty sure the transcript should read "[...] people start looking at Navier-Stokes equations [...]" instead of "[...] people start to look for your Stokes equations [...]" in Carmacks first answer. Those equations are actually used to simulate (turbulent) fluid flow, see: http://en.wikipedia.org/wiki/Navier%E2%80%93Stokes_equations

August 14, 2011 | 09:41 PM - Posted by Tim Verry

yeah, you're right. I listened to that at least twice and both times I heard your, but listening to it this time I heard Navier. I'll see about updating it, thanks for the heads up :)

August 14, 2011 | 08:21 PM - Posted by Anonymous (not verified)

'doing a render “man-like thing” going down to micro-polygon levels'

should actually be

'doing a RenderMan-like thing going down to micro-polygon levels'

August 14, 2011 | 09:43 PM - Posted by Tim Verry

Ah, so it's a proper noun... I hadn't heard of that before so took his words to mean rendering a human like form/character hehe. I'll see about updating it.

August 15, 2011 | 11:06 AM - Posted by pdjblum

Listened for a second time and was able to understand a lot more than I did the first time. With this listening, was able to recognize when John was going off on a tangent. Ryan did his best to reel him in. Great job and great interview. Thanks.

August 16, 2011 | 01:14 PM - Posted by Ryan Shrout

Thanks, appreciate the comment!

August 15, 2011 | 12:54 PM - Posted by BJ Blazkowicz (not verified)

“well we’ve got a Bézier spline, or a nerve, or something like that,”
should probably be
“well we’ve got a Bézier spline, or a NURB, or something like that,”

NURB is short for Non-Uniform Rational B-spline. Makes a bit more sense in context.

August 16, 2011 | 04:16 AM - Posted by Tim Verry

hehe, nice catch. John needs to subtitle himself when he starts talking super technical :)

PS Awesome username

August 19, 2011 | 03:56 AM - Posted by Anonymous (not verified)

Great guest and great Interview Ryan. Thank You, and thanks for 166 great hardware podcasts you've given us.

You guys could talk for 350 minutes and I'd not stop listening! Enjoyed watching two smart people didcussing intelligently about interesting topics, but you said Ok, OK too many times. Although certainly not your intention, it came accross as if the guest was boring and you needed to terminate each thread. Also, now that you've got this guy, why sit him in front of the elevators and in such a noisy environment with crew members pushing carts around in the background? Heck, with a guest like that, I would've rented a presidential suite room, sat on the couch and had a relaxed long conversation.

March 26, 2012 | 04:57 PM - Posted by Angel G (not verified)

I`ve played DOOM-I with great excitement, because it was the first game with such technology.
But have to say that their games need to become "inspired content", take for example SW KOTOR-II; Witcher-2; etc.
I mean that the game should make you feel, take decisions and witness the consequence, not just point & shoot everything, because it`s boring to kill thousands of opponents (at least for me).
It`s boring also to search labyrinths without any clue where to go....

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.