Center Foundation

A community blog for members of Center Foundation and friends.

Wednesday, January 26, 2005

Skin Graft at Seventy-Five Miles Per Hour

It is I, the magnificent Tuan! I have been called out by Maddog to join this thing and I figured that if Maddoggy were behind it that it just had to be a good thing. I suppose that I could post the usual summary of my life outside of college but instead I will highlight the most interesting events since graduation. Since last May I have survived a tornado, waded through flood waters, broken down in New Jersey, shoveled a public road to escape a snow drift, and switched from nVidia to ATI. True story.

(I am also the proud owner of a sweet ninja sword and a throwing star)

posted by Hedrick @ 1:47 AM  

11 Comments:

  • At 12:03 PM, Blogger rkstormer said…

    I'm with the Dog. ATI! Boooo! What posessed you?

    But wait here's the important question: what ATI card is it? In the lower price range the ATI cards might be better as the lower priced nvidia cards in the 5000 series (52xx, 56xx, 57xx, 58xx) are disappointing.

     

  • At 12:04 PM, Blogger rkstormer said…

    Where did this post's title " Skin Graft at Seventy-Five Miles Per Hour" come from?

     

  • At 12:53 PM, Blogger jaybox said…

    Hey, Hedrick. Good to hear from you. Where are you working?

    -joel

     

  • At 1:55 PM, Blogger Hedrick said…

    It's an ATI Radeon 9800 Pro. I looked at the similar product by nVidia and the price was $50 to $70 more. Also, it had amazing reviews by customers and professionals. I have become quite upset with nVidia. My brother brought his computer home at thanksgiving because his nVidia card burned out. The fan quit working and the heat sync got fused to the CPU. Then about two months later my nVidia just crapped out for no apparent reason.

     

  • At 2:11 PM, Blogger rkstormer said…

    The 9800 Pro is pretty good. Probably better than the comparable nVidia FX card if that is what you were comparing it to. I also have had issues with nVidia card crapping out, but then I haven't had an ATI card in some years so I don't have anything to compare it to.

    I've had a TNT2 that overheated and crashed the system unless I underclocked it (I believe this was only because it was a model that shipped with a higher clock speed than nVidia recommended), A GeForce 4400 Ti that stopped working after about 2 1/2 years, and a Geforce 1 that crapped (but it was really old and the fan had gotten clogged with junk and so I assume it overheated). Videocards have by far been the system componant that fails most often.

    Anyway, I still went with a geforce 6800 GT for my new card. I got a case this time with 4 fans (two in, two out) in addition to the normal videocard, cpu, PSU, etc. fans. Also, my motherboard and videocard have temp monitors now that will automatically underclock the sucker and warn you if the temp gets too high. You can also view the temps yourself in real-time. I'm hoping this stuff will help.

     

  • At 5:01 PM, Blogger Hedrick said…

    Maddog, you are amazing.

     

  • At 5:28 PM, Blogger rkstormer said…

    Correct. The latest ATI cards still use SM2. The 6000 geforce cards use SM3. ATI is trying to convince people that they don't need SM3. In theory you should be able to do just about anything you'd want in SM2 but long shaders would take multiple passes (they could be done in one pass in SM3). Also, from a programming perspective SM3 in a lot friendly (no limit on program length, allows all regular loop and branch statements, etc). The compilers have to pull some crazy stunts to get some stuff to run on SM2. For the time being I don't think anything really takes advantage of SM3 though. And I don't think it's clear when/if this will happen.

    My card has a two year warranty, and my last card burnt out after 2 and 1/2 year, hmm...

    If you are buying a case from a decent brand it really doesn't cost much more to get some fans (think about it, a fan is really cheap to make) that's why I decided to try some. The card on the other hand is perhaps more than I should have spent.

     

  • At 5:36 PM, Blogger rkstormer said…

    I forgot to add. Maddog's also right about the FX cards. The FX series was disapointing (though the really high end ones like maddog has were alright). The 6000 series on the other hand is quite good. ATI tends to fair a little better on HL2 benchmarks, but nVidia walks all over ATI in Doom3 benchmarks.

    Maddog, what do you think of my assessment of SM2 vs SM3? I'm not sure SM3 is even going to ever take over necessarily.

     

  • At 11:41 AM, Blogger rkstormer said…

    Easy there Maddoggy!

    I aggree that we are going to want 32 bits. The unreal team aggrees as well: http://www.beyond3d.com/interviews/sweeneyue3/
    I would hope that 32 bits is enough for a while. SM3 does provide 64-bit frame-buffer blending (which should help with the rounf-off error). In this interview the Unreal3 guy says they only use SM3 for performance increase.

    But remember nVidia was already using 32 bit precision with SM2, right? They don't have to support SM3 to get 32 bits.

    Concerning the AI, physics, etc on the GPU. Have you seen any data that shows that the GPU can actually do these other things faster than the CPU? If so then our CPU's must be horribly designed. The GPU is a special purpose processor designed to do graphics and shouldn't be capable of doing general purpose computation faster than our CPU's.

    I'm all in favor of SM3, but I'm just not sure that it is offering significant enough of improvements, and as long as ATI is ditching it, it may not take off too much.

    Here's some more down and dirty data about SM pixel precision in case you feel like browsing: http://www.digit-life.com/articles2/ps-precision/

     

  • At 3:26 PM, Blogger rkstormer said…

    I don't think I've ever seen or heard of a game that uses fluid flow simulation (I'm seen one or two where blood appeared to flow over a surface a little but I don't think the way they were simulating it could be called fluid flow). I agree that it would be pretty cool and useful in future games. And I agree that better physics and animation in general need to happen.

    I believe the reason it works for the fluid/gas simulation is because this is very parallelisable(in fact a grid of cells/pixels is ideal). I'm not convinced that general physics simulations would fair as well. If, for example, you were simulating rigid body movement/collision. Soft bodies might work I guess. I'd have to research this a bit (or maybe you can just tell me?).

    If GPU's do in fact speed up something (like physics) that would be broadly useful, I think we are going about this the wrong way. I've been thinking about this recently. If CPU's can't get much faster for the time being, then we need to introduce special purpose co-processors (remember the good old days when we worried about our floating-point co-processors?). Video cards are not yet fast enough that we can use significant GPU time doing other junk. You see all these tech demos of real-time fur, reflections, refractions, translucency, etc. But they are virtually no where to be found in any games. Even in HL2 and Doom3 we see virtually none of these things. Sure we have some nice water refractions but those are special purpose shaders (to render a plane of water) and I doubt they could be put on a general model and look right. We see reflection but they are either on a plane or just using an enviroment map from what I can tell.

    I don't think pixel shaders are fast enough yet, or if they are then just barely and only the latest generation of cards. The Unreal3 guy was talking about this a bit (in the link I gave before) and I read an interview with a guy working on CounterStrike: Source levels and he was talking about having to limit how many surfaces were normal mapped because it would run too slow (don't know what there target vcard was). I think there is much more that can be done in shaders than we have seen (certainly ore than I have seen in games), so it's not just a question of how many polys they can push.

    Of course GPU's will get faster and maybe have cycles to spare, but we want the most we can get which means a separate *PU for handling non graphics stuff.

    What do you think?

     

  • At 11:12 AM, Blogger rkstormer said…

    The *PU would would be different from the CPU. Likely it would work on parallelisable problems like the GPU but would be more general purpose rather than designed for graphics specifically.

    PCI-E should help the problem of reading stuff back from the card (otherwise what's it good for?). They are going to have to fix some things like this so that it can handle collisions and such since that's the main thing physics is used for in games.

    I think it's about time for a phone call...

     

Post a Comment

<< Home