Scalable graphics engine - in 15 months so much has changed!

[g__day]

* * *Ressurection selection time from 15 months ago on my favourite crusage - apologies in advance!

Having owned PCs for over 20 years now I know abit about strategic upgrades. I used to price point devices for months at a time to figure when they would reach a capability and affordablity for me and then budget accordingly. I have felt that for the past 2 years the major bottleneck on a rig destined to play 3D games is primarily the video card. So I upgraded video cards fairly regularily. Having 6 PCs at the moment I tend to buy a big card for my main PC then take its card out and put it into my second fastest PC and so on down the line until I retire the video card from my oldest PC each year.

To the point of this thread. If video cards are following Moores law cubed as NVidia like to hype, then a delayed upgrade sure buys you alot more.

For example I have 1 Geforce 3 Ti500 card, 1 Geforce 2 GTS card, 3 Geforce MX 400 cards and a Geforce 1 DDR card, so I have decided to skip a generation and not get Geforce 4. There is no compelling reason for me to update now versus waiting 6 - 12 months now and getting far better bang for my buck, at either the market leading product or a mid range price point.

If video cards were really doubling in speed ever 6 months (closer to 50% faster every 6 months I'd say) then waiting another 6 months beyond an upgrade means you get more than double the bang for your buck.

Imagine if someone in NVidia marketing produced a chart like this saying that even every 6 months video cards gained 50% in speed:

Rel. Date / Chipset / Cost USD / 3d Marks
Q1 2001 --- NV15 ---- $400 ------ 5,000
Q3 2001 --- NV20 ---- $400 ------ 7,000
Q1 2002 --- NV25 ---- $400 ----- 10,000
Q3 2002 --- NV30 ---- $400 ----- 15,000
Q1 2003 --- NV35 ---- $400 ----- 22,000
Q3 2003 --- NV40 ---- $400 ----- 33,000
Q1 2004 --- NV45 ---- $400 ----- 50,000
Q3 2004 --- NV50 ---- $400 ----- 75,000
Q1 2005 --- NV55 ---- $400 ---- 113,000
Q3 2005 --- NV60 ---- $400 ---- 170,000


It'd make you think about your buying habits for the next 3 years wouldn't it? Imagine saying in 3 years we will have a card that is more than ten times faster than NV30 will be delivered at!!!

And furthermore imagine if they put in an extra column in that table to say what new 3D games that hardware would allow you to run if you bought it immediately - at best there might be only 1 - 2 games a year in that square

Sobering - isn't it?

 

[DemoCoder]

It is no differnent with CPUs, memory, or harddrives.

I paid far more for 1 gigabyte years ago than I pay for 120gb today. My last computer had 128b of slow memory and cost far more than my current one with 1gb of DDR.

Let's not get into how much cheaper my Athlon 1.2ghz is compared to the Pentium3 I had earlier.

Whether Moore's law is to double every 18 months (in harddrives, it's actually less) or 6 months, it makes no different. The present value of your hardware today is worth almost zero within 2 years. Depreciation is incredibly quick.


But why complain? No one forces you to buy the newest and greatest thing. Just because something better comes out every month doesn't mean you have to have it, unless you have an ego problem and simply must own "Da Best"

I count myself lucky to live during a time in human history where I can see technological evolution happening in real time.

 

[g__day]

It is hugely different in two key ways!!!

First. Expontential rate of advancement == Expontential rate of obsolence

NVidia actually say 100% performance improvement every six months - no other piece of hardware is getting that sort of lift. CPUs, HDDs and memory don't double in both size and performance every 180 days!!!

Too they are not generally the primary bottleneck on any modern PC suited to playing 3D games - nearly always the video card is.

At least with other h/w components you don't wait over a year until s/w appears that can use it. How would you feel if you bought your amazing 120GB hard drive - but could only see 2GB of it until 2 years after it was released as a s/w patch becomes available pal?

Second. Software to fully use S.O.T.A. video h/w is always 12 - 18 months behind said h/w introduction

3D game engines graphic processing code is not very scalable. People don't say in advance here's what hardware we should have in 6, 12, 18, and 24 months times - these are its specs, we'll write software now that anticipates this so when folk have these amazing things it'll be ready to go from day one.

Oh no - a $20 billion a year industry couldn't be that clever could it?

Instead they say write games that don't scale for the 80% of the market that has the oldest slowest hardware out there.

A scalable graphics engine - too hard - pass.

Instead NVidia and ATi sell their amazing leading edge hardware and tell folks to wait 12 - 18 months before any software is written for it. Nobody minds waiting until 2-3 generations of faster hardware is out before software to utilise it fully starts to appear do they?

Cg and rendermokey may help this somewhat - but this problem of an out of sync multi billion dollar hardware and software industry is simply ludicrous.

 

[DemoCoder]

It is hugely different from two reasons!!!

First. Expontential rate of advancement == Expontential rate of obsolence

CPUs, HDDs, and Memory have been following Moore's law consistently. Moore's law is a geometric growth rate. The only difference between the equation govering Moore's law and the equation governing GPUs is the exponent. Both exceed linear/polynomial rates.

NVidia actually say 100% performance improvement every six months - no other piece of hardware is getting that sort of lift. CPUs, HDDs and memory don't double in both size and performance every 180 days!!!

No, they double every 365-540 days. So what.

Too they are not generally the primary bottleneck on any modern PC suited to playing 3D games - nearly always the video card is.

False. Fast GPUs need fast CPUs to feed them. Many 3D games are CPU limited. Try Medal of Honor or Jedi Knight II. Or try any 3D Sim for that matter.

Second. Software to fully use S.O.T.A. h/w is 12 - 18 months behind said h.w introduction

3D game engines graphic processing code is not very scalable. People don't say in advance here's what hardware we should have in 6, 12, 18, and 24 months times - these are its specs, we'll write software now that anticipates this so when folk have these amazing things it'll be ready to go from day one.

So? How is this any different than the chicken and egg problem in ANY industry. pick one: CDs, DVDs, HDTV, Fax machines, Beta/VHS, etc. There have to be early adopters who buy into a platform before large amounts of content are delivered for that platform. The first buyers of CD, DVD, LaserDisc, etc got screwed paying enormous prices, with no content available. Later, they became incredibly cheap and everyone benefitted.

DX9 is a new platform. It needs games. But before games are written for it, consumers need to own it.

Oh no - a $20 billion a year industry couldn't be that clever could it?

Instead they say write games that don't scale for the 80% of the market that has the oldest slowest hardware out there.

What are they supposed to do? Force everyone to upgrade? That's a recipe for going out of business in a free market.

A scalable graphics engine - too hard - pass.

It's not always possible for every type of game.

Instead NVidia and ATi sell their amazing leading edge hardware and tell folks to wait 12 - 18 months before any software is written for it. Nobody minds waiting until 2-3 generations of faster hardware is out before software to utilise it fully starts to appear do they?

Early adopters don't mind. They pay a privilege to have the stuff before you do. Are you jealous that you don't have money to burn? You were probably one of those people who used software rendering and didn't buy the Voodoo1 when it came out because NO GAMES EXISTED FOR IT. glQuake and glQuakeWorld came out way after its introduction, and that was a patch. Quake2 was the first shrinkwrapped FPS to support it.

At least with other h/w compoents you don't wait over a year until s/w appears that can use it. How would you feel if you bought your amazing 120GB hard drive - but could only see 2GB of it until 2 years after it was released as a s/w patch becomes available pal?

Actually, it's a 240gb raid 0, and most of it is empty. Unless you are doing DV, few consumer apps can fill up that much space.

And if you didn't know, the first 160gb drives wouldn't work on anyone's systems fully until the ATA/133 motherboards came out later, and then people had to upgrade.

if you attach a 160gb IDE drive to an ATA/100 motherboard ,you can only see 128gb.

 

[DemoCoder]

This argument is like those communist criticisms in Russia and China of Western supermarkets. "You have 23 different types of soap, toothpaste, and toilet paper, and new ones coming out all the time! What a waste, a misallocation of resources!"


How anyone could complain of things getting better, FASTER, is beyond me.

 

[g__day]

Some good examples but you have missed the thrust of my two key points.

So what if two games out off all in existence today are CPU bound rather bandwidth of the graphics memory limited. I said nearly always - not in every case. NVidia seem to be wanting to load far more of the load for 3D graphics off the CPU onto the GPU if their paper on ray tracing is anthing to go by.

I am happy to be an early adopter at times - but the fit for use argument applies. Your analogies whilst poignant completely miss the thrust!!!

Lets say a new car comes out every 6 months that is the same cost but twice as good as the old one - but you can't buy any fuel for it that allows this performance for 18 months - you still happy? Imagine selling better guns, planes and boats to your defence forces with the smae limitation - still happy?

Fine you have 23 varieties of everything - vastly improved - but you cant unlock the core potential of your product for 18 months - by which time it has been superceded 3 times over - because a dependent enabling product will not be produced - still happy?

You'd have to be truly crazy to accept that reply!!!

You are telling me that you'd be happy to have say 120GB of HDD with 2GHz of CPU and a lovely 1GB of RAM and only be allowed to use 10% of its potential for 18 months until the Operating System software is ready. Well isn't that clever - aren't you lucky!!!

* * * * * * * *

No I don't want anyone to upgrade at all - I want scalable graphics engines for most - not all games. At present very few games have anything near the scalability I'd like. I want better coordination between the game developers/ Microsoft and SGi / and NVida and ATi as where the industry is going.

Imagine you controlled a $20 billion dollar gaming industry (as the 20 largest game companies do) and say your sister ran the only two leading suppliers (NVidia and ATi) of graphics hardware and your brother ran the only two major APIs (openGL and Directx 9). Would you like to co-ordinate things better than has been done in the past?

Try reading my post again carefully to understand my key points - you got the wrong end of the stick my friend. Your replies are often out of context to my points.

 

[DemoCoder]

Everyone buys stuff they don't need. If every consumer only bought exactly what they truly needed or could use, the economy could collapse.

How many useless home exercise equipments are in closets?

 

[g__day]

Lol - your arguments are wandering even further a field, it may be time for you to rest and try again in your morning

 

[Geeforcer]

In the graphics industry, hardware comes before software. Using your car/fuel example, the new fuel will only be developed once the car is on the market. No one is going to write games for DX# without any hardware to support it. If Nvidia/ATi were to wait for software to appear before making cards to take advantage of it, they would go out of business. There would be no new cars, ever.

Another thing that you forget to mention in your analogy it that the new cars can run anywhere from 20 to 200 percent faster then old ones while offering better handling and smoother ride with EXISTING fuel.

 

[g__day]

That is a better argument - except you said "hardware comes before software", instead of more precisely saying "software comes 3 generations after hardware".

Why should this be acceptable to a $20 billion dollar industry?

Do you feel that say the 20 leading graphic architects (think John Carmack's of this world) and the major 2-4 video hardware manufacturers could sit down a few times each year and do a three year plan?

In any other industry that would be taken almost forgranted!!!

You ask these 20 guys/gals what in priority order they most want - you work out roughly when it can be delivered and if it is cost effective - you agree an API and how to test for its existance/readiness and away you go.

This industry seems to be very, very poorly optimised to me.

 

[DemoCoder]

g__day,
it may be a $20 billion industry, but the fact of the matter is, game development is a HARSH business. You spend 18 months developing, pouring your heart and soul into something that has 40 hours of play, and if you're one of the lucky ones, you get a hit that lasts for a few months. If you're not, you sell a few thousand units and then you go out of business.

Games are perishable goods. They don't have a long shelf life, but they take a lot of effort to development. Now you're telling these guys, who are the equivalent of subsistence farmers, to put alot of extra work into supporting a TINY TINY market (people with top end accelerators)

Probably 90% of that $20 billion is collected by a minority of publishers.

 

[g__day]

DemoCoder

That logic seems questionable to me - are you saying idsoftware don't make much money or that quake 2 or 3 aren't licenced and re-used extremely heavily?

Have not the quake, valve and unreal engines been collectively re-licensed 20 times over? Isn't that the way clever, leading companies develop game engines?

Personally I would expect the top 20 games development companies to reap most of the value being at the top of the food chain. Sure once in a while a surprise comes out - Croteam with Serious Sam spring to mind. Are you sure most of the revenue goes to a myriad of small companies that regularily go out of existence?

And no I am not asking "subsistence farmers, to put alot of extra work into supporting a TINY TINY market (people with top end accelerators)". I wouldn't expect any start-up to do this, so why table such a mis-directing, absurd question when I am talking about the biggest, richest companies - not the smallest, poorest?

Once again - read what I said carefully - I am asking the biggest, richest, most successful, talent rich, top 20 companies to develop scalable graphics engines that support current, emerging and future 3D cards. Not the poorest to only support the richest - but the richest to support all.

At present the richest seem to only be capable of surving by satisfying those with the oldest cards alone - versus supporting a full spectrum of video hardware. Is this problem simply unsolveable?

How are you continually missing what I am saying - how do I make myself and my desires clearer?

 

[no_way]

To accelerate time-to-market for 3D apps:
its been suggested a hundred times before, graphics companies themselves could hire software teams for pure graphics engine development, so that a kick-ass engine, taking advantage of all the new features of hardware, can be ready when hardware is ready.
After all, who else knows better how to write most optimal rendering routines for their HW than vendors themselves.
In this day and age, source art shouldnt be a problem anymore, developers can allow their artists to work with high-res data, downscaling so that it meets the requirements of lowest targeted hardware is mostly trivial.
At the moment, we have most well-known 3D coder JC working on an engine that is built around capabilities that were made available by original GF... thats at least 2 true generations old ! ( HW T&L -> GF Radeon , programmable shaders -> NV20 r200 , and now 2nd gen shaders with HLSL etc -> r300, nv30 ). And presumably only now is he beginning to consider the possibilities of current and upcoming HW. He may write shader code paths for NV30 for DOOMIII for example, but thats only a special optimization, the engine isnt built around the true capabilities of HW. DOOMIII will prolly not use HW accelerated HOS much for example, because its not something you can just patch in quickly. To really take advantage of the featureset, it has to be known before the design of rendering architecture.
And DOOMIII isnt even ready yet.. how long before first non-Id game will use its engine ?

Im actually surprised that NV for example isnt licencing any engine already, they could hire some of their long-time partners with respectable portfolio to do the job for them, if all of their own developers are tied up ( Crytek , Fun Labs , Vulpine ?? )

 

[Oompa Loompa]

How are you continually missing what I am saying - how do I make myself and my desires clearer?

Clarity != Realism.

You're asking for a level of scalability which is impractical. An engine which spans the pre-hardware T&L era, the DX7 fixed pipeline T&L era, the DX8 programmable pipeline era, and the DX9 floating point enhanced-programmability pipeline era just isn't in the cards.

Your request is akin to retrospectively asking for automobile engine parts to have been standardized between 1900-1940. We're still in the exponential growth period of hardware accelerated graphics.

 

[Randell]

the reality for the majority consumers buying this hardware is that the level of DX support that a peice of hardware supports is only really of interest to developers/programmers.

If I buy an R300/NV30 am I concerned that DX9 games wont run on it for 18 months? No, I will be buying it for its performance in current games and over say 12 months (I'm more concenred about getting AA & AF performance & quality up in my kit, I cant influence the speed of DX development). In fact if all a cards capabilities were exposed straight away you would probably find the performance level too low.

I doubt that I will run many DX9 games, if at all, on my first DX9 compliant card. In the same way I dont think I own a game with PS effects currently ( I dont own Morrowind or Comanche 4) on my DX8.1 card.

Not everyone buys into the product cycle, most dont. The cycle is there to push development forward and most people do sit out a generation or 2. And not everyone is in sync in the buying cycle so they all need to be catered for.

I upgrade a piece of kit, put my redundant kit in my sons's machine and sell his redundant piece to my local Computer Exchange, but try to keep the pace to when they still offer a reasonable amount of money Works for me

 

[g__day]

Ommpa -

you presume - where did I ever say pre hardware T&L? You invented that not me!!! Read more carefully please, even by inference my table started with the NV15!!!

Personally I would do the same as John Carmack with Doom III and presume Geforce 2 minimum.

We are also still in the period of exponential speed improvements of CPUs - its just we don't have to wait 18 months before more than 10% of their capability can be unlocked!

Its more like I want engine and transmissions and turbo chargers to be synchronised between the years of 1999 - 2002 and beyond. Its not retrospective - the industry should have thought of this for years, just about every other industry does. Imagine if air-defence was like this - planes are 50% faster each 6 months - but only on the runway - not in the air - until 18 months after when the proper control panel is installed.

Randell -

Fair enough but are you happy about the way it is today - do you ever want to see it get alot better? - don't you wish you could unlock that power sooner? Don't you feel the industry just isn't optimised well amongst the major powerbrokers?

 

[Evildeus]

I really don't see how you can think like that: stopping the evolution for what?

The point is you can delay your purchase as long as you think that:
- The 6 month of waiting worth the purchase today
- You don't have the money, you think you will have it, and you prefer to have a 50+% boost in 6 month than a VC today

But that's for someone who actually do NOT need a VC today.

I think it works like that in all market. The customer buy a product in function of its needs, its budget, and its expectation.

 

[Neeyik]

G-day, you might find that people take your points a little more seriously if you were less prone to the immature use of exclamation marks!!!! You don't need several on the end of every sentence, you know!!!! It looks like you're writing as if you're still in school!!!! The same is true for the smug attitude displayed in your replies.

Your table of data used in your first post is flawed as you are clearly using the same version of 3DMark with all the supposed chips. Half the tests in 3DMark2001 are CPU-bound and, like Q3A, beyond a certain level of graphics hardware, the score improvement is pretty small. I should imagine that by the time you hit the NV40/NV45, they will all be producing similar scores - why do you think nobody has produced a 20k+ score in 3DMark2000? The best genuine one is a little over 18k and that's using a 3GHz P4 and a GF4 running at 335/745MHz core/memory. The producer of that score himself has pointed out that he can get over 17k in 3DMark2000 without overclocking the GF4 at all.

It's only now that games have started to up the ante as far as graphical requirements are concerned, e.g. Morrowind or Neverwinter Nights. The former in particular has minimum requirements well beyond any other game - 32MB video card with DX8.1 compliant drivers. By comparison, Dungeon Siege will run on an 8MB card.

So to the point of this reply, to the point of your original post. "A delayed upgrade sure buys you a lot more". Well gee, as if nobody here is that smart that they couldn't figure that out. Of course it will - it's pretty much common knowledge that £300 in 3 years time will buy you a substancially better product than what £300 would get you today. It's called "technological progress", and we should be grateful that this is the case in PCs. Look at something like a car and the prices go up each year; of course you don't upgrade your car every year because there is no need to do so.

The need to upgrade in computer gaming isn't as big as hardware manufacturers would like you to believe, granted, but each year there are new people wanting to upgrade and if you (a company) want to get the most sales, you've got to have the best products out there and that means getting new gear onto the market frequently.

 

[g__day]

Neeyik

Congratulations - you seemed to miss the points I raised equally well.

3DMark was to give you an example you could understand easily. NVidia are clearly wishing to invade the CPU relevance in 3D graphics by all their MD says and their ray tracing paper. So in the future whatever benchmarks or games you desire - bet your bottom dollar NVidia will spend big time to shift as much as possible of the processing from the CPU to the GPU where transistor count and parallelism is king.

Smug - sigh - despairing if anyone will get the point - is hopefully closer to the mark.

!!! - big deal - does it really matter to you - is the grammar so unclear that you blatantly miss the 2 key points I raise? Why else did you find it so hard to get to the crux of the logic? Immature - are you attacking the person or their logic - cause you clearly haven't touched the logic I have used so far. Why shoot the messenger if you don't like the message?

In simple english - hardware advances are far, far outstripping software advances and have been for years in a multi billion dollar industry. Until very recently no one seemed to mind or try to improve things.

Evildeus -

I have no idea what you mean when you say "I really don't see how you can think like that: stopping the evolution for what? "

I see no one here talking about stopping anything - do you?

I am talking about synchronising development of gaming software with top end gaming hardware?

* * * * * * * * *

Can anyone think of an industry of such size where for years products come out so regularly with so much incredible performance potential - and it is hardly tapped because another critical component is not delivered until 18 months afterwards - an no one complains?

Frankly this sedate "its like that live with it attitude" astounds me. Is this the best optimisation of potential a multi billion industry can show?

I am surprised more people aren't crying blue murder such capability is left to become obselete before its potential is unlocked!

 

[Neeyik]

NVIDIA can rightly talk about keeping up and even breaking Moore's Law because Moore's "Law" was nothing more than an observation. He recorded that transistor growth had been expontential over the past 6 years and then wrote, and I quote:

"Thus there is a minimum cost at any given time in the evolution of the technology. At present, it is reached when 50 components are used per circuit. But the minimum is rising rapidly while the entire cost curve is falling (see graph below). If we look ahead five years, a plot of costs suggests that the minimum cost per component might be expected in circuits with about 1,000 components per circuit (providing such circuit functions can be produced in moderate quantities.) In 1970, the manufacturing cost per component can be expected to be only a tenth of the present cost.

The complexity for minimum component costs has increased at a rate of roughly a factor of two per year (see graph on next page). Certainly over the short term this rate can be expected to continue, if not to increase. Over the longer term, the rate of increase is a bit more uncertain, although there is no reason to believe it will not remain nearly constant for at least 10 years. That means by 1975, the number of components per integrated circuit for minimum cost will be 65,000."

I find it rather funny that so many people make a big deal out something he said in 1965 and that Moore himself said that the rate of increase is "uncertain". Marketing is such a wonderful thing!

Anyway, hardware advantages in graphics chip will always outstrip games, purely on the basis that the biggest selling games always aim for the lowest common ground. The US top selling games of last year (according to the NPD group) were:

1. The Sims
2. Roller Coaster Tycoon
3. Harry Potter & the Sorcerer's Stone
4. Diablo 2 Expansion Set: Lord of Destruction
5. The Sims: House Party Expansion Pack
6. The Sims: Livin Large Expansion Pack
7. The Sims: Hot Date Expansion Pack
8. Diablo 2
9. Sim Theme Park
10. MS Age Of Empires 2

No Quake3-based game there at all...

I know what you're suggesting - that hardware manufacturers should slow down their rate of product release in order to get game manufacturers up to speed and operate using the latest hardware. That simply not going to happen; all that "old" hardware isn't going to vanish overnight and the games will still need to incorportate that support if they want to sell lots of games. There is a much greater proportion of older hardware than newer in the user base and it would be mad for every game maker to spend money covering the entire base.

The key is to get people to upgrade and to do that you need to pump the market with fresh products. Using the car analogy from before, most people don't need to upgrade their cars and new car sales isn't all that hot either, but the manufacturers will still shove new models out twice a year to retain market share and interest.

NVIDIA et al must do the same; look what has happened to Matrox. A long-ish break from fighting at the top of the graphics card market (the G400 was a good product at the time) but a few years out of the battle and look what they return with. Great on paper, poor in performance and at the price they are asking for it, I can't see Matrox reclaiming much of their R&D costs.

A few games have begun the trend on cutting out the lower market (I forgot to include MotoGP in that 2 player list from before) but I'd bet money on none of them being in the list of best selling games for 2002.