It's amazing what a PIII upgrade will do...

[Rags]

SSE and 3DNow! have much more capabilities other than video T&L excercises, they can help with compiling, Photo Editors, Data processing, MP3 encoding/decoding, other encoding/decoding. I think that Video game coders would be wise to put to use both the Video card AND the CPU to good use, to gain all the processing power the system can pump. The CPU is not obsolete, yet .

Rags



------------------
P3 450@560, BH6, 128MB PC100, Delta CD ROM, Mill.G200/G400, 8.4G WD, 8.4G seagate, SB16 (temp.), Yamaha Burner, Zip, Sportster 56K int., Storm Scanner. System SPECS subject to change without prior notification to end reader or my spouse. ™

[Mark F]

Hey Zepher,
What do you mean :"If you can get a motherboard that divides right (no real good ones yet :/ but some new abits coming real soon might" ? Not sure what you mean by divides right.

Mark F.

------------------
OH NO, my retractable cup holder swallowed a CD

[Rags]

He is referring to the BX chipset's AGP clock divider being on the chipset itself. No software in the world can change the divider from either 1/1 or 2/3. The camino chipset is due out in a couple of weeks, ABit has a board called the CX6 coming that will use this chipset, it will allow a divider of 1/2.

Rags



------------------
P3 450@560, BH6, 128MB PC100, Delta CD ROM, Mill.G200/G400, 8.4G WD, 8.4G seagate, SB16 (temp.), Yamaha Burner, Zip, Sportster 56K int., Storm Scanner. System SPECS subject to change without prior notification to end reader or my spouse. ™

[Himself]

Rags,

All you say is true and I agree that should be the case, DVD decoding won't go away for instance. However, I don't think any of those things are what sell cpu SIMD, in my opinion that is all about faster games. A 3D scene can only be transformed and lit once, you can do some sort of load balancing but from the sounds of the T&L cards coming out, why would they need the much slower cpu to share the load? In the case of NVIDIA, from how long they dragged their heels on 3D-Now! I don't expect much more driver development with 3D-Now! emphasized at all. I pretty much expect all the 3D-Now!/SSE code in the drivers to be tossed in favour of sending the data to the video card instead. Well, there might be some left in the culling section or something but the T&L bit was where you saw the real speed improvements.

A PIII upgrade might still do some minor speed improvements (accelerated internet browsing? ), but I don't think amazing will be what you call it once T&L cards come along. I am thinking "MMX".

[Rags]

Himself,

You are correct, the 3DNow!, MMX, SSE are all marketing products. 3dfx, is coming out with a card that has an unreal amount of fill rate, that will have NO T&L, SSE can help here. But who wants to monkey around with optimizations, right?? VooDoo has, and you can see great benefits as evidenced by their cards and how well they perform in the K62/3 systems, we are talking about 65% improvements in frame rate in some instances. That is pretty tall. The problem I have with T&L is that you will have to have a sufficiently higher processing speed on the "GPU" to achieve the same results on a CPU, think for a moment, there will have to be cpu cycles wasted to move the data from CPU to video card (DX7). Also you have to account for a nice fillrate, which the GSPot does not currently have on paper (remember this is on paper), so all the T&L in the world is not going to help Higher res. frame rates, only fillrate can help that.

If anyone is purchasing a P3 for sse alone, they are wasting their money, because right now, there just isn't a whole lot that can benefit from it, but the titles are increasing. However, the P3 is your only choice right now for a higher clocked cpu, unless you want a celery. And the celery works fine for what most people do. For what I do, it didn't cut it. Multitasking took a big hit, moving from one window to the next, having a compiler running in one window, while d/l mp3s, posting on the forum, and playing mp3's the compiler would start crawling. With my P3, it is nearly 150% better. I guess the extra l2 cache (even though it is slower) helps, as does the higher clock speed, and my compiler is SSE optimized.

Rags



------------------
P3 450@560, BH6, 128MB PC100, Delta CD ROM, Mill.G200/G400, 8.4G WD, 8.4G seagate, SB16 (temp.), Yamaha Burner, Zip, Sportster 56K int., Storm Scanner. System SPECS subject to change without prior notification to end reader or my spouse. ™

[Himself]

You are correct, the 3DNow!, MMX, SSE are all marketing products. 3dfx, is coming out with a card that has an unreal amount of fill rate, that will have NO T&L, SSE can help here.

That's certainly true, but you are thinking short term, I don't doubt that T&L cards will rule the earth by the end of 2000.

But who wants to monkey around with optimizations, right?? VooDoo has, and you can see great benefits as evidenced by their cards and how well they perform in the K62/3 systems, we are talking about 65% improvements in frame rate in some instances. That is pretty tall.

I am no disputing that OGL games are where SSE/3D-Now! show a lot of benefit, that's just my point, the increases will get replaced by the video card when T&L is standardized. And I am not talking just about the G256, there is the Savage 2000 to consider, unlike some, I don't dismiss them as a joke of a company. This is still Sept of '99 remember, I don't doubt that in a year there will be a lot more T&L cards around.

The problem I have with T&L is that you will have to have a sufficiently higher processing speed on the "GPU" to achieve the same results on a CPU, think for a moment, there will have to be cpu cycles wasted to move the data from CPU to video card (DX7).

I think that will be an issue, but not as dramatic as you might think, only when games are looking like Toy Story or something will that really be an issue. What you lose by extra bus traffic you gain in not performing fpu operations or context switches from FPU to SIMD (where applicable) and you can use FPU writes for more speed depending on the architecture.

Also you have to account for a nice fillrate, which the GSPot does not currently have on paper (remember this is on paper), so all the T&L in the world is not going to help Higher res. frame rates, only fillrate can help that.

I agree the fillrate of the G256 won't be enough once games really use T&L, but they don't exist anyway so it's an interesting geek feature on the card, just like EMBM etc. It's a neat way for NVIDIA to avoiding continued work on 3D-Now! support from where I sit.

[Marmita]

Himself u r a little wrong abour SSE and 3DNow going away. First there is not enough spec. about the GeForce but I'm sure it won't take %100 of the T&L calculation on itself coz that's a stupid thing to do when u got some powerful instruction sets in the main cpu to help u. As for the Savage2000 (that I really think will be the next big thing) it has more then %150 the fillrate of the GeForce and it will have a T&L unit too.. but S3 announced a new technology which will share the T&L calculation between the cpu and the graphic card. a company that will not use the SSE/3DNow will simply lose performance... if 3Dfx will use a P3-750 SSE instructions with a huge FillRate I'm sure they'll have better performance then a GeForce not using it and with a poor fillrate...

There are 3 important things u must remember:
1. lighting is done through a lightmap which is the same if u play 640x480 or 1600x1200, hense the T&L calculations arn't mor complicated for a high res game they are exacly THE SAME !

2. ALL the games we have are FILLRATE LIMITED !!! T&L wont brake any speed record without a big fillrate backing it up...

3. The next Matrox card will kick nVidia's, 3dfx's and S3's ass !

[Marmita]

to make a long story short, Why should any company waste time on a feature that we already have in our CPU and it's getting faster as CPU's are getting stronger. It's a much better idea to create a super fillrate card that will work with the CPU's instruction... if anyone read the reviews on the GeForce, Its only 120Mhz and it runs VERY HOT. it burned the reviewer's finger when it touched it... It's virtually impossible to o/c it (only +10Mhz is safe)

Let intel do there job... they already have a great implementation that is getting faster with each new cpu that comes out. I expect from my next Graphic card a huge Fillrate and some nice features like EBMP

btw whats the G400 and the Max's FillRate ?

[DarkSyd]

The graphic chip companies are at least trying to pick up the slack left by Intel/MS. If they WERE doing their jobs, you'd be posting to the forums from your Dick Tracy watch while playing Quake III on your holoshades. Remember what an 8mhz Amiga could do 14 years ago? If you don't, just ignore the previous ramblings

------------------
Cel 266@448 on Asus P2b (1009), 128MB, OEM G400/16, Quantum Obsidian S12, Fujitsu 5.25gb, MX300, Dlink 530, Grey Cat, Orange Cat (still MIA)

[JeremyGray]

"The problem I have with T&L is that you will have to have a sufficiently higher processing speed on the "GPU" to achieve the same results on a CPU, think for a moment, there will have to be cpu cycles wasted to move the data from CPU to video card (DX7)."

I keep reading postings, previews, etc. all over the web, and I can't help but think that people keep missing the point of hardware accelerated T&L. I'll provide a highly simplified example:

To render a 3D scene, the vertices of poly(s) must each be run through several matrix operations before they can be displayed on the screen. A bunch of culling and clipping takes place as well, but we'll ignore that for now. Vertices must be translated from local space to world space, then view space (the projection viewing volume), then screen space. Most of this is currently done in a combination of application software and driver software without hardware T&L.

Without hardware T&L: for a 1000 vertex mesh, do thousands of matrix multiplications in a general CPU, and then send 1000 vertices to the hardware for rasterization.

With hardware T&L: send 1000 vertices plus a few 16 float matrices to dedicated hardware.

Hmm. While the CPU is very fast, it is a generalized processor, and can be easily beaten by dedicated hardware even at a much lower clock speed. Does anyone know the PSX clock speed? :-) If you do, you'll understand the power of dedicated hardware. With hardware T&L, you send just slightly more data across the bus, but place the processing of that data in the SIGNIFICANTLY more efficient hardware.

And to those many people who wrote or read previews, etc. on the web claiming that hardware T&L takes control and optimization options away from developers, this is completely false. Keep in mind that there are many steps taken to transform a vertex from local space to screen space, and that the developer can easily draw the line between what (s)he will perform in hand-tuned software in the CPU and what will be handed off to the hardware. It's simply a choice of doing more or less of the work before handing the data off to the hardware (to the driver, actually - and this is something currently done by developers even without hardware T&L in order to optimize their own transformations instead of leaving it up to the driver software).

[Marmita]

what the hell r u talking about ??? what did the 8mhz amiga do ? (I owned one) it did nothing ! U could have taken a better example and say: "Look what the PlayStation can do with much less horse power"

but then again u must remember that these machine (like the amiga) where tuned for games... the G400 can render a quake3 scene faster the a P3-500 coz its tuned to do so... a PC is an all around performer it must play games and in the same time be able to run complicated programs and simple office programs like word. The amiga was more of a console machine then a computer.

I'm not saying that these two companies r doing there best, they are doing there best to make money... Microsoft's products are mybe a little buggy, But this company is one of a kind there product r the BEST around and if it wasn't so we would all buy Word Perfect or Corel's Office suite or mybe Lutos 123...
If u don't like intel and microsoft I suggest U take an pricy but better performer Athlon and pair it up with a Red hat linux or mybe even BeOS which are much better OS's then Windows... But then again u'll have the fuctionality of a 8Mhz amiga in a 650Mhz monster...

[DarkSyd]

Amiga was tuned for gaming? Hardly, it was tuned for multimedia and just happened to play games as well. It was capable of turning out quality stuff (Babylon 5 for example) at a time when PC's were good for nothing more than word procesing and playing text based adventure games.

Sorry to start this here..I shall post in the 'obsolete but cool hardware' forum next time

And Jeremy, my money is on the PSX being 33mhz.

[Wombat]

Two words:

Video Toaster.

[JeremyGray]

DarkSyd, 33MHz is the magic number. It's amazing what you can do with dedicated hardware at even that kind of speed. Any Bleem! user can tell you how much PC-based horsepower it takes to emulate a 33Mhz PSX (emulation issues aside). :-)

Anyway, I just wish people would fairly evaluate the promise of hardware T&L. I've read a lot of uninformed previews out there during the weeks since the GeForce and S2000 press releases. A lot of authors seem to simply not understand how this stuff works, and they are passing on their lack of understanding to a lot of people.

[This message has been edited by JeremyGray (edited 09-14-1999).]

[Himself]

Marmaduke or something like that wrote,

Himself u r a little wrong abour SSE and 3DNow going away. First there is not enough spec. about the GeForce but I'm sure it won't take %100 of the T&L calculation on itself coz that's a stupid thing to do when u got some powerful instruction sets in the main cpu to help u.

Let me put it this way, PIIs are not exactly slow cpus without SIMD, SIMD takes extra time and energy to code for, most of it in assembly language. According to specs the G256 is around 3-4 times faster than the fastest PIII using SIMD, whatever is left over for the cpu to do wrt transformations is minor stuff, you won't need to go the extra mile and bring out the SIMD guns. As we've seen, getting video card companies to do a proper job of 3D-Now! support is like pulling teeth, with an excuse like T&L forget it.

As for the Savage2000 (that I really think will be the next big thing) it has more then %150 the fillrate of the GeForce


Well, no point in playing the numbers game, we'll see what the benchmarks show. I agree, on paper the S2000 looks better and I agree that the G256 needs more fillrate independent of any comparisons with other cards.

to make a long story short, Why should any company waste time on a feature that we already have in our CPU and it's getting faster as CPU's are getting stronger.

Why should they waste time on extra coding in assembly language to support SIMD stuff when the video card can do it faster anyway?

You should be clear on who it is who will be doing the coding, it is primarily the video card companies doing everything in the drivers, game developers won't see any of the T&L hardware directly.

btw whats the G400 and the Max's FillRate ?

On par with the rest of the lot, TNT 2's etc.
Depends on the clockrate, around 300 or so average.

Remember what an 8mhz Amiga could do 14 years ago?

Yeah, coprocessing with a good OS. T&L won't fix Windows, that's for sure.

Jeremy,

You seem to know your stuff, how exactly does T&L work, say from within OGL? Do current games use OGL's T&L for all their model animations, if so how would that work wrt T&L on the video card? Collisions? Getting data from the card again? Is there an OGL command to tell the drivers to send the data to the video card to be processed? Not a clue here.