What other hardware/Technology is on the horizon?

On the topic of Visual Studio .NET...

We decided about 4 weeks ago to use C# for our current project, and man...Between the IDE and the language, it's pretty darn awesome. I've had no formal training on C# at all, and it has been a piece of cake to use (though I've been doing Java development straight-up for the past 18 months).

Without any doubt, Microsoft knows how to make a very good development IDE. IMHO, nobody comes close. How does that saying go? Imitation is the ultimate form of flattery?

Take a good look at the devel. IDE's in Linux to get a sense as to what Linux developers think of the MS devel. tools...You might as well call a lot of these things Linux Visual Studio...I mean, they intentionally try to make them as close to Visual Studio as possible...and there's a good reason why. VS _is_ good.

I would also put MSDN up against any Help application out there. Having some 3GB worth of information @ your disposal is no joke.

 

And, quite obviously, if someone else can't, you don't. The problem with Microsoft is either that there is no "someone else" (anymore) or Microsoft doesn't need to compete on quality to eliminate them...we do agree in general on what a monopoly is, right? Perhaps this quality issue is why people complain. If you don't think quality is a valid reason, we can skip this discussion...

Do you work for Microsoft? J/k, j/k. In any case, what about Open Source programmers who ARE end users?

While couched in terms suitable to a Microsoft executive "interview", you do point out valid problems with the Unix environment in general. Too bad as a result of long past history and recent judicial failure it seems there won't be any competitors in the (Wintel) PC space who do have the funding to run usability experiments. You know, for that competition stuff you mentioned earlier but haven't quite seemed to connect to Microsoft or monopolies as far as I can tell.

Which categories and what criteria for "best"? I'm suspecting the criteria are going to be related to features Microsoft has direct control over, like inter-operability with other Microsoft products, and that quality is going to tend to place a ways down the list. But maybe they've had lack of competition in the categories you have in mind long enough to have established quality leadership, or bought it from someone else.

Than what? I thought they had the OS thing all sewn up, and somehow looking at history I thought this gave them an edge in planning and allocating their resources in other areas around what they knew they would do with their OS (and in fact what they did with their OS was favorable to maintaining this edge). But if there is another OS that can compete directly on quality and successfully contends with the simple fact that "everything" is written for Wintel machine and codebase, please mention it.

Ah, competing entrenchments. So you're saying all those other companies are profitable and healthy and retaining market share against Microsoft? I'll take your word for it for now, and so I should blindly (IMO) hope Microsoft doesn't change focus on what they offer for "free" in the future?

*blink* Aren't those tools where Microsoft dictates the target? Don't they have a bit of an advantage there? Also, am I mistaken, or does Windows XP not ship with OpenGL driver code at all? Is that competing on quality? Or perhaps there was some other reason for that selective omission?

Microsoft had done things that have shown benefits. The problem is other things would have shown more benefit, like competition. Which is really the actual point about capitalism you were making initially, so it is still confusing to me you think it justifies Microsoft's monopoly or makes complaining about it "whining" (nice of you to maintain the high ground, btw).

I do understand your point, and problem is that as the list increases, the quality of the items in the list stops being determined by competition. It is possible to both include the functionality and allow competition to continue (even though Microsoft spent a lot of time in the Antitrust case trying to say this was not the case), it is merely that it was in Microsoft's interest to do only one, and the consumer's interest to do both. The two interests aren't the same thing, and as Microsoft's interests are served more and more you are surprised that consumers complain?

Mighty blithe in your pronouncement on what should and should not be. Hey, if I'm buying my Server OS, shouldn't database functionality be included? Where is the line drawn, and why is it drawn there today and will be tomorrow?

Since my DVD player software came with my video card and DVD drive, I didn't think to comment on that. Now that you bring it up, it seems a good thing Microsoft is there to allow them not to bother including such, it's not like Microsoft would charge the video card manufacturer and makers of the DVD drive for some sort of "Media Center" driver certification or something similar instead of WinDVD's or PowerDVD's makers prices for packaging...

I am really curious as to where and why you draw the line. It's strange that we ever had DVD and CD players. It's like at some point there wasn't OS functionality encompassing this. *scratch* Pondering that, and looking to the future, I'd like to know which market segements are immune to this phenomenon....I think games may be as long as Microsoft is limited to the number of "sub-studios" they can attract. What else?

And, again, the problem is that what can be "commodified" is a longer and longer list as time progresses, and that quality is not the deciding factor on whether something is commodified (wouldn't DVD player software makers have to pay Microsoft money to properly utilize some of the planned media features as exposed on Microsoft's OS in any case?). If you don't have a problem with it, that's up to you, but your attempt to even remotely link this to the principle of competition you began this post with seems a failure to me.

 

I definitely prefer Borland's IDE, though it is true that their software isn't the best for writing 3D apps.

 

Actually AFAIR there has been public comments from IMGTEC themselves that Series 5 will be released next year one way or another (as in if they find a partner or not).

Joe, since your in this thread I'll take this opportunity to ask you something

Is there something wrong with your Beyond3d account or something? Because I've PM'd you about 3 times over the last month and you still haven't replied since your last PM to me over a month ago.

Anyway PM me sometime, we still have a wager to sort out don't we?

 

I seriously hope PowerVR Series 5 comes out next year, and that it's hot shit. I mean that it's competitive or even better than R350 and NV35.
maybe that's asking too much but, PowerVR used to kick ass. If Series2
had been out for the PC market in 1998 like it was supposed to (all the manfacturing resources went into producing the PowerVR2DC variant for Dreamcast in late 1998-fall 1999) it would have wiped the floor with Voodoo2/SLI and TNT.


I was also really disheartened that all Videologic/ImgTech has had over the past few years is the T&L-less, twin pipline, Series3 based KYRO/KYROII. I wanted badly to have Series 4 with its 4 pipe config+T&L unit (was it a DX8 part???) - I am sure that Series 4 would have provided all the graphics rendering power of NAOMI2 and then some. Actually, Series4 would have probably far surpassed NAOMI2's twin PowerVR2DC rasterizers + ELAN T&L unit. I really wanted Series 4. It should have been out spring 2001 at the latest... sigh

 

Right now, there's only 1 3D card company selling and shipping DX9/Ogl2.0-supporting hardware--and that's ATI. Nobody else is. But honestly, given the realities of this market I will be very happy if just two very strong, very competitive companies emerge. I find that infinitely preferrable to a single company emerging as victor. In other words, I want to do whatever I can to prevent nVidia's slogan of, "Lighting every pixel in every monitor" from coming true, not because I dislike nVidia but because I don't want to see any one company win totally--that will kill advances as I'm sure you know and guarantee we'll all pay more for less. I guess I'm saying that we don't have a crisis point until a single company stands as winner of the endgame, so even two strong competitors is not to be complained about (could be a lot worse.)

My prediction: not going to happen. Matrox was tottering when it released Parhelia; current release by ATI of R300 is too far ahead for Matrox to keep up with--company doesn't have the R&D funds to make the kind of commitment that, for instance, ATI made when it purchased ArtX for $400,000,000 US--and that's what it'll take, IMHO. (Not only that, but I don't think companies like ArtX who can execute as well as they design grow on trees, exactly.)

I am very interested in whether the power VR will be marketed from now on as a contender for the performance crown, or as strictly a value proposition for the low end of the 3D OEM market. The technology has always intrigued me but I fear I am right in thinking that economies of scale relating to the price of high-performance ram for 3D cards has already rendered PowerVR impotent. I would like to for once see a very powerful Tiler get produced--but I've wanted that for a long time now and have no reason to expect that '03 will be any kind of a different year.

Not going to happen. BTW, I think Xabre is definitely low end 3D. The purpose behind Xabre was to produce a very cheap 3D graphics chip with some marketable qualities (like AGP x8 compatibility) so as to sell at a price maximizing profit. I think the Xabre line is already doing what its designers wanted in that respect.

Big disappointment to me. I thought that Creative Labs would take P10 and compete with it. Apparently, CL doesn't see the product as terribly competitive and so scrapped any plans it may have had to enter the 3D- card fray. Basically it seemed to me that CL simply wasn't interested in competing in the 3D card market--and I'm not sure whether that says anything much about the potential of P10 or not. It could just be that CL doesn't want to enter this market with a compeittive product which they have to support themselves. Rather, I think they'd rather license a reference design by someone else. It would have been nice to know something more definitve about P10 than the basic theories behind it.

Does anybody really care (what time it is?--*nah*) about low end 3D?

 

ImgTec has the major downside that it is an IP selling company, meaning that there are more than one factors that they can't control themselves, once they've sold a license. Recent example was ST Micro abandoning the PC graphics market, thus Series4 was cancelled.

I'm a bit confused about the ram part in your comment: if you want a high end Tiler with next year's measures, then it's inevitable that it must carry high frequency DDR ram, should it be aimed to compete with the top dogs. High end means high end in pricing too and with the level of complexity/transistor count advanced programmability requires, it shouldn't turn out that much cheaper than competing products. Or do you expect to see past dx8.1 level hardware with just SDRAM?

For the record the canned Series4 carried already DDR ram (synchronous to clock speed - most likely 250MHz and if there would have been a refresh later on 300MHz).

 

obviously you werent running 3.5 or previous!!!
and let's get this straight- i am NOT a linux newbie. in fact, i rather dislike Linux myself. I'm a UNIX guy and i see linux as being (origianlly) an immitation which has taken on a life of its own. Capitolism?! HAH! Ive got news for you- IT DOESNT WORK! When a company gets to a certain point, they no longer have to compete with quality as they can take loosing a LOT of money if thats all they have to do to kill the competetion. Remember how much an Intel cost back when AMD was putting out crap knock-offs? I have a better example- Bell Telephone. Do your history on telecommunications and you'll see how Ma Bell took advantage of everyone because they didnt allow competition. No good 2D/3D work can be done on UNIX? Then why do ALL of the CAD/CAM guys at my uncles 100-billion-dollar engineering firm use UNIX? For that matter, why does almost EVERYONE in his firm use UNIX? These are no computer newbies either, several of them have been involved in projects such as the first 10MB hard drives. Now, I'm not going to say that UNIX the holy grail of OS's, I'd think that should be capable of producing a much better os. now as for MacOS X- i think its a GREAT evolution of UNIX. If Apple starts using AMD's, or puts out MacOS X for the PC i'll be one of the first to grab it. I also love BeOS. Ever get BeOS to crash? I hanvt ONCE been able to get BeOS to crash, and it was able to get higher connection speeds with my 32k winmodem than windows could- that was back before linux had any support at all for winmodems. what happened to Be? Well, MS managed to out-PR them. If you've tried tried BeOS 5 then you would know that its far superior to anything MS had when it came out. only problem was poor PR. without anyone using it, developers didnt care towork with it, and so it died.

 

Humm, doesn't you example illustrate exactly the opposite (that the free marked does indeed work)? Back in K5/Pentium era, when Intel's CPUs where superior to the ones offered by the rest, Intel was able to charge huge premiums and still expect good sales. Attracted by the high margins on high-performance CPUs AMD developed its own high-end line, almost immediately increasing its market share and profits, while driving the prices down. Faced with increased competition and eroding market share, Intel responded with accelerated development and price cuts. Capitalism at work.

 

And who cares if he wasnt?
NT 3.5 came out in Sept. 1994!
Thats over EIGHT YEARS AGO!
who cares if that wasnt stable? Their OSes are stable now...
Good lord. How long do you plan on holding an unstable OS against them?
You know, pentiums SUCK (all of em!!! - even now!) because the first ones had problems in huge spreadsheets (loss of precision)

You know, judge a company by current products please.

 

Exactly. Any product that comes from the company that made products that had the FDIV and F00F bugs just has to be crap....

 

Well, my remarks about the ram simply go back to the original idea behind a tiler--that it could make much better use of less local bandwidth than a comparable "brute-forcer", which would have to use much more expensive (at the time) ram. It wasn't so much that the Tiler architecture might be faster than a brute force approach, but that a Tiler could provide comparable performance results for less money because it was much less dependent on local bandwidth than the BF architectures. What proved out, though, was that the necessary architectural complexity of a tiler gpu would exceed the complexity of a BF gpu to such an extent that it was hard to demonstrate any real performance advantages between them. Indeed, in all cases the BF architectures still proved faster, and because of economies of scale 3D card ram prices greatly diminished and thereby nullified this advnatage of the tiler concept. This did not however close the door on someone making a better, more efficient tiler architecture which might enable the approach to reach its theoretical performance potential.

The implementation of Tilers to date, IMO, sort of loosely parallels AGP texturing technology. AGP was dreamed up by Intel at a time when 3D-card videoram cost $50-$75 per megabyte. AGP texturing was seen as a way to use system ram to double for onboard 3D card ram so as to save money (considerable amounts of money at those ram prices.) However, before AGP bus technology could develop to a point of being useful, ram prices plummeted to <1/50th their former prices, and in a blink of an eye the concept of AGP texturing was rendered an obsolete, inferior technology (compared to local bus 3D cards texturing from their own onboard ram over their own local buses.) So, tiling seems to me to have already outgrown its value as an approach, which in the beginning was chiefly economical (IMO) and provided a way to utilize more local 3D card ram for less money (since the ram wouldn't have to be as fast as that required by a brute-force architecture to max out on performance.)

I still like the concept, though, and am always hoping that someone will be able to come up with a tiler architecture which will surprise me and all of us... But to be honest, I don't expect it. But I'm always open to pleasant surprises..!

 

WaltC,

I disagree with your argumentation. You say that bandwidth/economics was the key idea behind tilers, and while bandwidth is one issue the real key idea behind a tile based deferred renderer is efficiency which includes bandwidth, memory usage, processing power and a lot of other things (including features, PowerVR is still the only one to offer true per-pixel hardware translucency sorting on Dreamcast).

I am not sure what you mean with architectural complexity ? Are we talking about gatecounts ? I think KYRO proves that a tiler is not more complex than a traditional architecture since it has a gate count at the same level or below similarly spec-ed IMRs. KYRO has also clearly shown the architectural advantage in benchmarks like Villagemark but also in real games where at the time independent reviews showed that under acceptable conditions (as in games where opaque overdraw was sufficient) a KYRO could match up to a much higher spec-ed GF card.

Tiling is still about bandwidth and efficiency and all the HyperZ and other Marketing claims you hear are about the solving the same problem. Efficiency remains the root problem of 3D Graphics and DDR2 and 256bits buses are not solving it (they parially solve the bandwidth problem). You have to remember that things will only get worse, we get more and more data flow due to increasing polygon counts, we now use full floating point buffers and textures and with MRTs your looking at 512bits per pixel being written out (really kewl if you end up overwriting that pixel several times !). Do we really want to process a pixel with a 100+ instruction shader when its hidden ? And MSAA remains stuck with a memory footprint problem no matter the addition of compression, so forget about 16x MSAA on an IMR unless they want to introduce 256MB of ram.

But in the end we'll just have to prove it

 

Architectural complexity is not very tightly connected to gate-count; for example, a large register file adds very little to the complexity of an architecture, but still consumes lots of gates - similarly, control logic tends to add a lot to the complexity even though it rarely consumes very many gates at all. It's more appropriate to measure complexity in e.g. lines of HDL code or the number of man-years (or possibly just years) needed to design/verify the architecture.

And memory bandwidth? I can see a rather hard wall around the 1 Terabyte/sec mark (where you start hitting the physical limitations of FR4 and MCM packages), but nothing closer than that ...

 

OK, granted there was considerably more prior-art for IMR rendering and its more obvious to implement. Still not entirely sure how that lines up with WaltCs comment about performance.

 

well then you must have more luck with your computers than me or any of my three friends sitting next to me. and ill tell you why it matters- because NT4, 2000, and XP are all evolutions of NT3.5. the PPro and PIV are not evolutions of the original Pentium, they are totally seperate architectures that are all based on the x86 instruction set. You see how fast hardware companies change their products? PI>PPro>PIV / Itanium, K6>K7>K8, G3 > G4 > (K8? / 64-bit PPC?), GF1>GF3>NV30? SIMM > DIMM > DDR DIMM / RDRAM > QDR. Granted, some companies are somewhat stagnating (Sound Blaster line...) but overall, the industry is moving at a much more rapid pace and with their enormous resources I would expect Microsoft to be leading, not dragging us behind with backwards compatibily concerns. The Itanium is backwards compatible (albeit slowly) and the K8 is able to provide backwards compatibility while gaining the full benefits of 64-bit computing. The NV25 and NV30 are fully capable of doing traditional TnL without the need for specific dedicated hardware.

as for capitolism.... it only encourages corrupment. i'll say the same for communism and "anarchy," so dont think im arguing fo those either.

 

WaltC,

If deferred rendering/tiling wouldn't show any advantages at all, then today's IMR's wouldn't result to hierarchical Z-buffering, fast Z clear, Z- compression, tiled memory approaches and what not. I've already seen notions that all the pre-mentioned techniques might get combined with application driven deferred rendering (unshaded pass followed by a shaded pass) with PS f.e. on IMR's.

The only point where I'm dead curious about is how a high end Tiler with today's standards and with what efficiency overcomes the vertex bandwidth limitations of TBR's of the past. If then a TBR manages to handle vertex bandwidth as any optimized Z-buffer accelerator, then the overall efficiency should be (at least in theory) impressive.

Care to elaborate what then the tile HSR logic on NV30 stands for? There must be a way to attempt to achieve the efficiency a competitor has with bandwidth saving techniques and 25% more raw memory bandwidth.

 

LOL. Yeah, totally new architecture. Right. You dont think they share anything other than instruction set?
I suggest you read a book on computer architecture.
As for you and your friends - luck? No, but maybe skill. I work with dozens of servers in a REAL actual work environment. I work with thousands of client machines. And you know what? I know what OSes are stable and which are not. Call me when you have actual experience - and stop reading slashdot and taking it as gospel.

 

(1)Is there any compatiable problem when some campany use the techology like tiler?
(2) Can we make the Z-compression ratio up to 1: 10 instead now 1:4 and result in huge performance increase?

 

No to (1). I just don't think that any form of tiling/deferred rendering on any platform has outpaced it's use. I see it being the exact opposite. The issue is not where it's used, it's the final results that count to me.