Outstanding performance of the NV40 = old school 3dfx mojo?

[Scali]

Tile based Renderer =/ TBDR.

IMHO only Gigapixel and PowerVR were/are TBDR

Firstly, deferred part wasn't mentioned before, why stress it now?
Secondly, this doesn't invalidate any of the statements I made.
So this still leaves me to conclude that it is no more than 'mystic powers'.

 

[Arun]

If the texture caching mechanism of the NV40 is his puppy, I find it kind of ironic that a man who contributed to the original invention of this mechanism will now bring the industry its nicest innovations in a decade. We love you too, 3DFX!

Uttar

 

[KimB]

Not just NV. S3 were ahead too for a while (Savage3D, Savage4, and the early days of Savage2000).

The Savage3D had pretty poor drivers, and the only claim to fame that S3 really has is S3TC (in 3D graphics, anyway). The Savage4 was too little too late, and the Savage2000 was DOA due to the GeForce 256.

From what I seem to recall, S3 really had some excellent products before the 3D revolution. Then came the Virge.....

 

[KimB]

It was essentially tile based rendering, similar to that used by Power VR. Why do more work with more transistors when it isn't required! Basically the Gigapixel stuff gave similar framerates to the competition with less transistors and lower power consumption (if my memory serves me correctly).

It was meant to, at least. I don't think Gigapixel ever actually made a product. They certainly never sold one.

 

[Ailuros]

Tile based Renderer =/ TBDR.

IMHO only Gigapixel and PowerVR were/are TBDR

Firstly, deferred part wasn't mentioned before, why stress it now?
Secondly, this doesn't invalidate any of the statements I made.
So this still leaves me to conclude that it is no more than 'mystic powers'.

Oh yes it does there's a fundamental difference between an immediate mode renderer and a deferred (or display list) renderer.

We have tiled (or chunck) based memory optimisations since the early Voodoo days.

For the record Trident never built anything out of the patent they bought from Microsoft for the Talisman architecture AFAIK, yet I'd personally would count Talisman to the batch and somewhat exclude Intel from the list.

TBDR is an alternative philosophy; obviously the two major vendors and their research teams have found that they gain better results while deferring only parts of the rendering process on their IMR based architectures. One example would be "application driven deferred rendering" (an unshaded pass followed immediately by a shaded pass).

Whether NVIDIA is internally still doing some research on TBDR based on Gigapixel's philosophy I can't know, but I have severe doubts they are.

To be completely honest there wasn't that much known about Gigapixel, from what was known they must have had quite a few interesting ideas/patents that must have been put to good use. Apart from that the few details known about GP1/2/3 never seemed to introduce anything special compared to what PowerVR had under development at the time.

Here's an old newsblurb about GP3; no idea if the specs are accurate.
ram, it's author is an active member here, so he could maybe clarify a few things about it:

http://www.3dconcept.ch/cgi-bin/showarchiv.cgi?show=763

 

[Ailuros]

It was essentially tile based rendering, similar to that used by Power VR. Why do more work with more transistors when it isn't required! Basically the Gigapixel stuff gave similar framerates to the competition with less transistors and lower power consumption (if my memory serves me correctly).

It was meant to, at least. I don't think Gigapixel ever actually made a product. They certainly never sold one.

That's true from what I recall. Someone correct me if I'm wrong but 3dfx gave GP1 and GP2 free for licensing after the acquisition and the reason why Microsoft didn't choose them for the XBox was their failure to present working silicon on time.

 

[Scali]

It was meant to, at least.

If we have to be absolutely realistic, then we have to conclude that no tile renderer has ever delivered on this promise.
I like the idea of tile rendering, and I think if it is executed properly, tile rendering will indeed be more efficient (mostly speedwise, and lowering memory costs that is... the chip itself will actually require more transistors I suppose, because of the required internal buffers).
But I have never seen it done.
The best I've seen is a Kyro II, which I have had for a while. It could compete fairly well with my gf2gts speedwise, and the rendering quality was excellent, especially the 16-bit mode was nice. But there were faster cards available by then.
I hope that the new PowerVR architecture will be licensed and put into production, it may be the first to actually deliver.

Other than that, I think ATi and NV will try to avoid (deferred) tile-rendering for as long as possible, because they will basically have to start over completely when they make the switch. This will take time, and put them at a disadvantage with the competitors. So I think it's a stalemate for now. On the short term, it's probably better to get the most from the current technology.

 

[jvd]

The best I've seen is a Kyro II, which I have had for a while. It could compete fairly well with my gf2gts speedwise, and the rendering quality was excellent, especially the 16-bit mode was nice. But there were faster cards available by then.
I hope that the new PowerVR architecture will be licensed and put into production, it may be the first to actually deliver.

You mean your kyro 2 with a pipline set up and clock speeds matching that of a tnt 2. Which in some cases was faster than the geforce 2 gts ultra ?


How about the dreamcast . Clocked at a 100mhz with a 1x1 pipline design putting out beautifull graphics that in its 2 years of dev time looked as good as games with 2 years of dev time on the newer ps2 ?

T power vr architecture that matchs the r300 spec for spec would blow the r300 based cards out of the water .

 

[JoshMST]

Will this mark the introduction of "The Texturing Computer" rev. 2.0? Yes, yes Rampage was the greatest vid card that never was, but we cannot overlook the simply amazing architecture that it would have brought to the scene at the time. From my understanding, the greatest strength of Rampage was its flexibility in allocating resources dynamically, rather than the relatively fixed functions that the GeForce 3 had. Not to mention that 3dfx was able to do more with less transistors. Fear was supposed to be a killer product, and it was based on the same texture computer design, built upon the Gigapixel technology. To have the brains behind this design take the knowledge of what has come before, and apply it with the latest developments in design and manufacturing, I think that the NV40 will be a very impressive chip.

It is just unfortunate that NVIDIA tried to do so much with the NV30, and failing on so many different levels as compared to what ATI had done with a new, and very focused design. This mistake by NVIDIA cost them a great deal of mindshare, as well as marketshare. Whatever anyone has to say about NVIDIA, they are definitely a company that in the past has shown that they learn from mistakes. I don't think their internal culture has changed so much that they are significantly different there. It is a well run company, with a lot of talent, and a very driven CEO that is in contact with his company every day. After talking to ex-3dfx guys that work at NVIDIA now, the main thing they have to mention is that their lan gaming skills have gone downhill drastically because they don't play anymore. It is a very disciplined working environment. However, the NV30 project tried to do too much, ate up too many resources, and was not nearly focused enough to compete with the great design that ATI brought to the table. When leaks about the R300 started to show, NVIDIA went into panic mode and tried to get the design to compete, but it was too far gone by that point. I think initial ideas of clockspeeds for the NV30 was around 350 MHz to 400 MHz. Now we see why the dustbuster was put in place to keep it running at 500 MHz! Looking back, one cannot help but admire what ATI did with the R300 (small board design, quiet, relatively cool, and really damn powerful).

Bring on the vid card wars! April is going to be great!

 

[Tim Murray]

"And lo, the PowerVR looked down upon his flock from Heaven and sent his only Series 5 to them..."

Okay, so it's not the Second (or Fifth) Coming. But meh, I'm still looking forward to it.

 

[Dio]

The Savage4 was too little too late.

I wonder if I'm allowed to say why nowadays?

 

[KimB]

I like the idea of tile rendering, and I think if it is executed properly, tile rendering will indeed be more efficient (mostly speedwise, and lowering memory costs that is... the chip itself will actually require more transistors I suppose, because of the required internal buffers).

Not necessarily. Tile-based rendering requires caching of the entire scene, and for large geometry densities, that may end up being worse than using an external z-buffer.

Then you have to consider that as shaders get longer and longer, the memory savings of deferred rendering will mean less, as most of the time will be spent doing math ops instead of accessing memory.

 

[Scali]

You mean your kyro 2 with a pipline set up and clock speeds matching that of a tnt 2. Which in some cases was faster than the geforce 2 gts ultra ?

gts or ultra? Make up your mind.
And as I said, it was a good match for my gf2gts. Sometimes faster, sometimes slower, generally about average.
However, the gf2gts was not the fastest and most advanced card out there, when the Kyro II was launched. The ultra was much faster, and the gf3 was either launched already, or followed shortly after the Kyro II. It was both faster and more advanced than the Kyro II.

How about the dreamcast . Clocked at a 100mhz with a 1x1 pipline design putting out beautifull graphics that in its 2 years of dev time looked as good as games with 2 years of dev time on the newer ps2 ?

The DreamCast has a giant advantage over PowerVR-chips in PCs, and that is the API.
PCs need to work with OpenGL and/or Direct3D. These aren't the ideal APIs for the PowerVR chips. The DreamCast allowed you to push the data in an optimal way for the PowerVR chip (opaque polys first, then translucent, alphatested, etc).

T power vr architecture that matchs the r300 spec for spec would blow the r300 based cards out of the water .

That's my point. Nobody has ever made such a device, so we don't know how well it will perform.
Perhaps the extra transistor cost for the on-chip tile memory etc will mean that clockspeeds are inherently lower for tile-based devices.
Perhaps tile-based rendering and programmable shaders don't mix very well.
Or whatever other reasons you may think of.
We won't know if tile rendering can actually beat conventional devices until we actually see a device that does.
Nobody has ever made a tile renderer that is at the cutting edge of both features and production methods, like the conventional devices are.

 

[Scali]

Then you have to consider that as shaders get longer and longer, the memory savings of deferred rendering will mean less, as most of the time will be spent doing math ops instead of accessing memory.

On the other hand, deferred rendering can mean that these long shaders are run for much less pixels, since overdraw can be eliminated (which ofcourse is possible with conventional hardware aswell, using a z-only pass first).
But that is my point exactly, we won't know how good deferred tile renderers are until someone builds an up-to-date one.

 

[Joe DeFuria]

That's my point. Nobody has ever made such a device, so we don't know how well it will perform.
Perhaps the extra transistor cost for the on-chip tile memory etc will mean that clockspeeds are inherently lower for tile-based devices.
Perhaps tile-based rendering and programmable shaders don't mix very well.
Or whatever other reasons you may think of.
We won't know if tile rendering can actually beat conventional devices until we actually see a device that does.
Nobody has ever made a tile renderer that is at the cutting edge of both features and production methods, like the conventional devices are.

Lol...I've been saying the same thing for well over 5 years now.

Usually you get a response along the lines of "well, PowerVR doesn't decide what parts to make, so it's not their fault!". (Which of course is irrelevant and besides the point....but is often a springboard to criticizing the PowerVR business model for the PC space....but I digress...)

 

[nobie]

I think some of the advantages of a TBDR are now somewhat moot because ATI and NV have co-opted alot of the capabilities, with features like HSR and Z-buffer compression.

 

[jvd]

gts or ultra? Make up your mind.
And as I said, it was a good match for my gf2gts. Sometimes faster, sometimes slower, generally about average.

I'm talking about the ultra. It was awhile ago so i don't remember if it was just the geforce 2 ultra or gts was still in the name .

However, the gf2gts was not the fastest and most advanced card out there, when the Kyro II was launched. The ultra was much faster, and the gf3 was either launched already, or followed shortly after the Kyro II. It was both faster and more advanced than the Kyro II.

Yes and the kyro 2 cost less than the geforce 2 ultra much less. Why would u expect it to compare against a top of the line gpu ? Which at the time was being beaten by the geforce 2 ultra ? So hey i guess at the start there may have been a benchmark the much cheaper kyro 2 beat the geforce 3 in .

Not bad for a card you could get for sub 150$ at the launch of the geforce 3 .

The DreamCast has a giant advantage over PowerVR-chips in PCs, and that is the API.
PCs need to work with OpenGL and/or Direct3D. These aren't the ideal APIs for the PowerVR chips. The DreamCast allowed you to push the data in an optimal way for the PowerVR chip (opaque polys first, then translucent, alphatested, etc).

Well lets see the powervr version of the neon 250 had to work with windows ce. Which had a version of dx on it. So that is thrown out right htere. Then it had to work with segas os .

Not only that but it was also a lower clocked version of the pc part.

The reason why that pc part did not do well is because the dc took most of their time and effort.

That's my point. Nobody has ever made such a device, so we don't know how well it will perform.
Perhaps the extra transistor cost for the on-chip tile memory etc will mean that clockspeeds are inherently lower for tile-based devices.
Perhaps tile-based rendering and programmable shaders don't mix very well.
Or whatever other reasons you may think of.
We won't know if tile rendering can actually beat conventional devices until we actually see a device that does.
Nobody has ever made a tile renderer that is at the cutting edge of both features and production methods, like the conventional devices are.

Well in the past we have seen it (kyro) , In the console market we have seen it (Dreamcast) in the arcades we have seen it (naimo (sp?) namio 2 ) and in the moble market we are seeing it right now (mbx ) Or and not only is the mbx more powerfull than the ati and nvidia offerings but it has a much smaller transister count , offers free 2x fsaa and suses much less power .

SO i think we've seen it . We also know that it doesn't need cutting edge production methods to stay in the game against brute force parts .


I can agree that we really need to see a dx 9 lvl card from them.

But because of their lack of 3d add in cards doesn't mean they haven't proved to have great tech and leading edge tech in other fields

 

[Hellbinder]

Re: Outstanding performance of the NV40 = old school 3dfx mo

So would it be fair to say the role ex-3dfx staff played on the NV 30 architecture was minor, and with the NV 40 ex-3dfx workers got their hands really dirty?

I have no idea, and I think it's quite a stretch to insinuate anything like the above. nVidia just got bit on the ass finally by using the latest process with NV30, something that had historically always seemed to work in their favor.

The Problems with the Nv30 are a hell of a lot bigger than the silicon process they used.

Put that sucker on .15 micron and it would be even SLOWER with a lower core speed than it was when released.

Nv30's problems are architectural.

 

[Hellbinder]

I like the idea of tile rendering, and I think if it is executed properly, tile rendering will indeed be more efficient (mostly speedwise, and lowering memory costs that is... the chip itself will actually require more transistors I suppose, because of the required internal buffers).

Not necessarily. Tile-based rendering requires caching of the entire scene, and for large geometry densities, that may end up being worse than using an external z-buffer.

Then you have to consider that as shaders get longer and longer, the memory savings of deferred rendering will mean less, as most of the time will be spent doing math ops instead of accessing memory.

Wow, Now where have i heard comments like this before....

Oh yeah these are the same comments certain people allways chuck out there every time this subject is approached.

The problem is Its not founded on reality.

Perhaps you should read up on Deffered Shading?? Or some of the other information out there that clearly demonstrates that large Geometry may not be the damper its said to be?

Heck its been addressed in these forums 100 times at least.

 

[JoshMST]

The more I have discovered about process technology, the more I am finding out that NVIDIA's problem really wasn't TSMC's 130 nm process. So for once, I agree with Hellbinder! From all that I have heard, TSMC's 130 nm process was clean by Spring of 2002. Again, it appears that the changes to NV30 were a kneejerk reaction to what ATI was going to release (or it could be that the entire project was flawed in the beginning). Still, from my perspective, making any chip, especially one with over 100 million transistors, work at any speed is pure magic to me. The complexity of what each company does is mind boggling, and the fact that ATI pulled it off so well is even more impressive!