The monster is out !!! (who can stand against the mighty NV20?)

[OneOfTheseDays]

Look, the Rampage may not be 500 bucks but lets get one thing straight. When both the NV20 and Rampage come out, they will cost significantly more than other card out there. They will cost a lot of money and i personally do not want to spend 300+ on a video card. I can get a 1ghz cpu for lower than that. If the price is above 300, which it probably will be for the Rampage and defn. for the nv20, im gonna wait till it goes down. Oh yeah and another thing, Nvidia doesn't need to sell cards as badly as 3dfx because they aren't the ones in deep financial trouble if their video card isn't an instant success.

 

[Pocatello]

Nvidia is not as stupid as Dave claims, as soon as there are comparable cards from ATI or 3dfx, the NV20 won't cost more than the competitions. Look at the price of the GTS and the V5, both offer similar performance and price, even though the V5 uses cheaper memory. Lets look at the cost of the Rampage, the Rampage will have SLI support. Even with my very limited knowledge, it would take 64MB or 128MB of 200MHz DDR RAM, 2 main processors, 2 T&L processors (according to rumors , the Rampage will have a separate chip to do T&L), while nvidia NV20 will have a single main processor with 32MB or 64MB of 250MHz DDR RAM. Truly, the only person who knows about about the cost of the Rampage is Dave, perhaps he will correct my errors. The only version of the Rampage that will cost less than the NV20 is the non-SLIed version like the V4 4500, and nvidia probably will make a cheap version of the NV20 to replace the current GF MX.

 

[PCResources]

<< As for NV20 being so wonderful, we can already say how fast it is going to be. Overclock a GF Ultra's memory by 20 MHz and add 30% more memory bandwidth into the factor and you have your performance gain. So figure it will be able to use about 1/2 its actual speced performance >>



Ok, so the story once again is that nVidia actually shows little performance improvements, but charges big time? I find myself wondering how a company that stupid can sell so much, all while a company as smart as 3Dfx cannot sell as much.

As for the increase in die size, if you were to use 75% of that die size, but instead use two chips to be able to keep up, wouldn't that be worse?

These posts of yours are always so very one sided Dave, you actually seem to know very little about the real performance of the NV20. In time benchmarks will tell if what you say is true or not, but then you will probably start to bash the benchmarkers.

Patrick Palm

Am speaking for PC Resources

 

[PCResources]

<< PCResources, your constantly on my case. >>



Constantly??? Oh well, i thought i was being very nice... I agree with you on most things, but what actions of nVidia don't you agree with?

Patrick Palm

Am speaking for PC Resources

 

[Blackhawk2]

<<Look, the Rampage may not be 500 bucks but lets get one thing straight. When both the NV20 and Rampage come out, they will cost significantly more than other card out there. They will cost a lot of money and i personally do not want to spend 300+ on a video card. I can get a 1ghz cpu for lower than that. If the price is above 300, which it probably will be for the Rampage and defn. for the nv20, im gonna wait till it goes down...>>

<<50 to 60 million transistors is a lot of real estate even at .15. add in the fact that nvidia is guinea pigging that process (neither intel or amd will be there for at least 6 months) plus some really high speed (read: expensive) ram and you get something that costs $$$>>

It also pumps out 15 watts of heat as well

Tile-based rendering to the rescue!

Now to answer the original question:
Q: who can stand against the mighty NV20?

A: Tile-based rendering can.

 

[OneOfTheseDays]

Ok tile-based rendering is probably going to be the used very much in the future, and there is no denying that. However, it isn't perfect. Look at the current tile-based renderers out there, kyro, they still have image quality problems and compatibility probs. Videologic has been working with tile-based renderers for years too, and they still have yet to perfect it. Unless these gigapixel guys really know their tile-rendering as well as everyone says they will, don't look at tile-based rendering with such enthusiasm.

 

[merlocka]

I'm just sitting and waiting for the next gen Trident chip to shred NV20 and Rampage... all for 39.95

Hey, at least I linkified.

 

[Blackhawk2]

Sudheer Anne, have you played with Sega's Dreamcast console? Notice any glitches in Dreamcast games or how about Sega Naomi2 arcade games?

No? They all use tile-based rendering
Its proven to work, and works extremely well. Transistor count is about 1/3 or less of any equivalent conventional renderer, it doesn't need expensive memory to gain bandwidth, it produces less heat (less transistor count) and it costs less to manufacture.

Sure it has two flaws, both of which have been overcome. 1st flaw is rendering transparent images, the extra fillrate saved by using tile-based rendering can be used to fix the overdraw incurred by transparent images. 2nd, supposedly it can't draw huge numbers of polygons, Naomi2 proved it can by using parallel rendering architecture ->10 million fully-textured sustained (this is key) in game polygons per second can be rendered with Noami2, this is more than a Geforce2 - about 2-3X more.

In case you don't beleive me on Naomi2's specs heres a link:

Naomi2 specs

 

[Blackhawk2]

There is some extra advantages to tile-based rendering that I forgot to mention:

8 layer multi-texturing
FSAA
Ability to do better texture filtering and/or special effects

Assuming you have spare fill-rate it can be used for anisotropic filtering instead of inferior bilinear or tri-linear filtering. Better special affects can also be acheived because you have more fill-rate to spare than a conventional renderer (which uses it for overdraw) does. This is all assuming you have a decent fill-rate to begin with, unlike Kyro which was underpowered in the fill-rate department and polygon processing power.

btw my brother is an ASIC Engineer and I've already bounced these ideas off him, and he agrees

 

[Pocatello]

Will the Rampage support Tile-Rendering? From what I've heard, it won't.

 

[Blackhawk2]

Pocatello, I agree I don't think Rampage will use tile-based rendering. It appears to be too far along in the design phase, it might even be finished the design phase, I don't know as I don't work at 3dfx maybe Dave could lend a hand. But 3dfx could prove us wrong.

 

[Pocatello]

It's a shame that Dave can keep a secret

 

[audreymi]

3D has been the life buoy of many graphics chip marketings managers:

One can get away with blurrier text than their competitor,
unstable driver releases that work in a few platforms,
current demands that exceed reasonable limits used in
last year's more popular motherboards, overheating, burning
hot heatsinks, $500 retail prices, rare memories,
seperately powered graphics cards AND

THE FPS COUNTER WILL STILL SELL PRODUCT
in spite of these system oversights.

I predict NV20 will continue to be a life buoy with
1/2 the floation value.

Do I smell cupcakes ?

 

[Blackhawk2]

audreymi, true Nvidia may get away with it currently (no high-end tilers yet) but when a tile-based renderer with high-end performance hits the market Nvidia will lose if their solution is not tile-based. As I've always said, a traditional renderer vs a tile-based renderer with equivalent specifications will always see the tile-based renderer win because of lower cost. 1/3 the transistors equals 1/3 the price BUT NOT 1/3 the performance

 

[Fozzie]

&quot;However, the past two months provided us with enough information to withhold or recommendation of the GeForce2 Pro. Sure, the card is powerful and priced within reasonable bounds, but the OEM only designs (with exception to the Guillemot card we examined here) and the fact that the NV20 is on the way means that most die hard gamers would be better off waiting for a more powerful card at the same price.&quot;

From the GTS Pro review.

&quot;Same Price&quot;? Ultra or NV20?


&quot;As I've always said, a traditional renderer vs a tile-based renderer with equivalent specifications will always see the tile-based renderer win because of lower cost. 1/3 the transistors equals 1/3 the price BUT NOT 1/3 the performance &quot;

Sorry Blackhawk, this is incorrect AFAIK. In fact a tiler of identical specifications(pixel pipelines, pipeline functionality &amp; more) will actually have more transistors because of the embedded Z-Buffer ram and hardware needed for the HSR compared with a traditional design with external local frame-buffers.

Current Tilers typically have lower gate count and power usage because they are not equivelent in specs compared with traditionals but only similiar/better/worse in performance.

 

[DaveB3D]

sigh..


NV20 will be insanely expensive...Pocatello, you are simple wrong. You are comparing this generation to next. You can't do that because they are completely different products. And they don't have an MX version coming (any time soon anyway) for the simple fact that there isn't enough they can chew off to reduce the price. So they can either cripple it or they can lose DX8 compatibility.

Patrick,

I'm sorry. I know you think you do, but you simply don't know what you are talking about (and yes, I'm saying this knowing what I know about you). Whatever your situation is or may have been you are simply wrong, and that is all I'm willing to say on the subject.


You guys can say I'm wrong all you want, but the simple fact is that I've spoken with NVIDIA a great deal on NV20. I obviously know what 3dfx has in store. I guess all I can say is that you'll see. This isn't the first time this year I've had this debate with someone about NVIDIA either. The last one was about when NV20 would be released and what date they'd get samples back from the fab (roughly.. the date.. keeping in mind I do know the exact tape-out date of NV20). I was right on that one, and I'm extremely confident that I'm right on this.

 

[glen]

My prediction:
The NVidia NV20 will come out at $400, then quickly drop to $350.
The 3dfx Rampage will come out at $369, then quickly drop to $329.

 

[ElFenix]

did you see that trident chip! holy nuutz! it supports FSAA, 32 bit colors, and even has a quote from The Boss himself! not to mention that its faster than a 3dfx banshee and &quot;designed for sub $800&quot; PCs.


now, i'm no engineer, but it seems to me that the yields of a device 75% the size of another could very well be double. you get 33% more devices out of a wafer to begin with. plus each larger device is 33% more likely to contain a flaw in the silicon, since they take up more space. combine that with a process that has pretty much all the kinks worked out, and i can see where you could get double the yields out of the larger process/half transistor device. combine that with savings realized by using a much more common form of RAM, and i can see where a 2 chip board could cost less to make than a single chip board.


i wonder if my power supply can stand up to the mighty NV20?

 

[Sunner]

Dave, couldnt you please be quiet regarding the NV20 vs Rampage debate?

Im saying this for two reasons.

1. As long as you cant provide ANYTHING to back your claims you have as little credibility as an nVidia engineer would have if he came here promoting the NV20 all the way to heaven without even trying to prove it.
Now I do nderstand that you're under loads of NDA's, but frankly, what does that mean to the rest of us, the end result is the same.

2. You do work for 3dfx, and regardless of what you say, In a debate which is about 3dfx vs nVidia, your credibility will be very much reduced, since you WILL be biased, purposely or not.
Something I understand perfectly, if I was in a debate about my company vs our arch enemy, I wouold certainly be biased, there's just no helping it.

This is not to say I dont appreciate you taking the time to post here, cause I do, and I think alot of other people do as well.

 

[Blackhawk2]

nevermind.

 

[PCResources]

<< I'm sorry. I know you think you do, but you simply don't know what you are talking about (and yes, I'm saying this knowing what I know about you). Whatever your situation is or may have been you are simply wrong, and that is all I'm willing to say on the subject. >>



Ok, Dave, we will see.

One thing though, &quot;specs mean nothing...&quot;.

Patrick Palm

Am speaking for PC Resources

 

[Blackhawk2]

Hey Fozzie if I'm so wrong take a look at the bold text and link.

I would say that is less than 1/3 and more like 1/5

<<...The graph clearly shows that the GP-1, using 4X FSAA and TrueColor, is a way faster than a Voodoo 5 5500 and also clearly faster than a GeForce2 GTS. Well, at the first sight, this may not be that surprising. But, according to the presentation, the GP-1 board is just equiped with low budget low speed 100 MHz SDR SDRAM, which offers just about 1,6 GB/s memory bandwith! This is less than 30% of the 5.8 GB/s the Voodoo 5 5500 and GeForce2 GTS bring into the race! Further, the GP-1 integrates just about 3 Million 3D transistors, in comparison to the 15 - 25 million transistors integrated in today's high end graphics chips...>>

And I love this part:

<<...What's the conclusion based on this results? High Performance is not necessary a result of expensive highspeed-memory and millions of transistors, but also and more important the result of efficent and intelligent chipdesign...>>

3dfx/Gigapixel GP-1

 

[Fozzie]

Blackhawk you misunderstood what I said and missed the point. The GP1 design is smaller and more bandwidth efficient compared with a brute force traditional like the V5 yes. I never said otherwise. What I said was that a tiler of like technical specifications will be more complex then a comparable traditional. If you create a tiler design comparable to the Geforce 1: T&amp;L engine, Pixel shaders, 4 pixel pipelines with equivilent functionality(free trilinear) the tiler will most likely be have more gates.

The reason the GP1 is smaller and claims to have equivilent performance is twofold, one the deffered rendering that allows it to remove opaque overdraw and the bandwidth gains from tile based rendering. Second Gigapixel uses a different method of AA that they claim allows 4x with no performance hit, that means if the GP1 is as fast as a GTS or a V5 with 4X AA it is 4 times slower without it.

&quot;The results shown by 3dfx may be a best-case szenario for the GP-1, the chip may be beaten by today's solutions by using standard 16bit- or 32bit-benchmarks without activating Full Scene Anti Aliasing.&quot;

All tilers in the consumer market and the GP1 have on paper inferior specifications, 1 or 2 pixel pipelines, limited pixel shaders, no T&amp;L, etc, etc. This does not mean they have inferior performance compared with Traditionals with external local frame-buffers. The main reason for this is simple because Videologic has only focused on the lower end of the market with their products. And the GP1 was Gigapixel's first operational design shown. They have also been more complex to design and more difficult to get to the market, this should change now that more R&amp;D has been acomplished.

We'll see how traditionals respond though with embedded frame-buffer and advanced occlusion detections.

 

[DaveB3D]

Actually GP-1 is from TNT/Banshee days. So even the raw spec was pretty much the same.

 

[Blackhawk2]

<<Blackhawk you misunderstood what I said and missed the point...>>

This is what I said:

<<....As I've always said, a traditional renderer vs a tile-based renderer with equivalent specifications will always see the tile-based renderer win because of lower cost. 1/3 the transistors equals 1/3 the price BUT NOT 1/3 the performance.>>

This is your response:

<<Sorry Blackhawk, this is incorrect AFAIK...>>

It seems Fozzie that you are the one who misunderstood.

<<The results shown by 3dfx may be a best-case szenario for the GP-1, the chip may be beaten by today's solutions by using standard 16bit- or 32bit-benchmarks without activating Full Scene Anti Aliasing>>

Yes I read this too, but GP-1 is gone. 3dfx/Gigapixel are already well beyond that, about 2-3 generations beyond with GP-3, GP-4

You might have forgot to read the following also:

<<...Further, they announced the &quot;GP-3&quot; core, which also should include a geometry engine (T&amp;L)...>>

Hmm the GP-1 being just paper specs, this ain't no Glaze3D:

GP-1 and Quake3

The writing is on the wall for Nvidia, if they can't read it thats their problem.