New "facts" about the NV20 :)

[Czar]

http://www.3dhardware.net/articles/comdexfall2000_part1/



<< Same amount of pipelines but more texture processing power per pipe
Bumped up clock frequency
50% faster than the GeForce2 Ultra (seems a bit low. )
15 Watt heat dissipation - that is a lot

Yes, so I know this isn't much at all and that these specifications are just what we have heard and again we won't guarantee them being true. While a 50% in performance increase is not bad but I think the NV20 is so overhyped that everyone expects more. However this could mean a 50% increase in performance in terms of clock frequency and that would be a lot.

We were told that Guillemot is working on a new video card that will dissipate 15 Watts of heat which is a lot when it comes to video cards. It is so much that they have approached a cooling company for them to develop a special cooling system that is supposed to draw ambient air from outside of the PC case and suck it inside. This is because the ambient air inside the case is too warm to be efficient. I find this a little odd considering that the CPU is being cooled by the air inside the case.. This new cooling would thus occupy a slot on the back - bummer and hope this is not true. Although we do not know what chip Guillemot is supposedly using for this new card we doubt it is a Geforce2 Ultra, leaving the other reasonable choice being a NV20.

>>

 

[NFS4]

I surprised that no one else is interested:Q

 

[Remedy]

The only reason no one is interested is because you didn't post it first brandon!

 

[NFS4]

Hehe...I was getting ready to post this, but something told me to do a search first. Sure enough, it was already posted

 

[ElFenix]

they're scared by 15 watts? the geforce 256 was 16 watts!

more texture processing per pipe, guess that means one more texture unit, bringing it up to three. same as the radeon. 3/2 is 1.5, so theres 50% more theoretical gigatexels or pixels or whatnot.

 

[Blackhawk2]

Wooooaah some people are concerned about the heat problems with the current Athlon systems...just wait till they try to stick one of those NV20's in, nuclear meltdown

 

[NFS4]

At least it's not as bad as the 6000 would have been

 

[Blackhawk2]

NSF4, true. 15 watts is a lot of power to be dissipating and the original Geforce needed special instructions to slow it down when it was getting too hot. If the information is true I wouldn't buy it with that much heat dissapation, my computer room is small and with that much heat it might become an oven

 

[LordSandMan]

Also, in my opinion, the Geforce 256 did NOT have enough cooling, mine would lockup all the time, even without being overclocked. Then I put a big pentium heatsink on it. No more problems.

 

[NFS4]

<< Also, in my opinion, the Geforce 256 did NOT have enough cooling, mine would lockup all the time, even without being overclocked. Then I put a big pentium heatsink on it. No more problems. >>


Never had a problem with my SDR GeForce. In fact, it is o/c 150/200 with no problems with the retail heatsink/fan.

 

[IBMer]

more texture processing per pipe, guess that means one more texture unit, bringing it up to three. same as the radeon. 3/2 is 1.5, so theres 50% more theoretical gigatexels or pixels or whatnot.

Just remember that the GTS and the GTS Ultra have awesome Pixel and Texels claims, yet do not hold them up in tests.

As a matter a fact that GTS Ultra doesn't not even hold the regular GTS's Texel and Pixel fillrate even in 16-bit color. So then again they can claim 5billion pixels per second, but with out 50GB/s memory bandwidth they won't be close.

 

[Leo V]

&quot;...they have approached a cooling company for them to develop a special cooling system that is supposed to draw ambient air from outside of the PC case and suck it inside...&quot;

Is that company called Rainbow Vacuum Cleaners? :frown:

It sounds [pun] like I'll be keeping my MX for a long, long time. The last thing I need is a vacuum cleaner running 24/7 right next to me. Jet airliner takeoff noise was bad enough, but this'll be like flying a Sopwith Camel in the heat [pun!] of a World War I battle...wheee

 

[RagingGuardian]

I guess it was too hyped up in the first place for real. Anything short of phenomemal would be a disappointment to most of us. Needless to say that I am disapponted.

On the upside I can use a Tbird/NV20 combo to keep my room warm now it's winter.

 

[Syborg1211]

LordSandMan- That is exactly what I did. Were you one of the people I told to do that? I can't remember quite well. I put a Pentium one cooler on my geforce 256 to get it to 165/365 to score 6500 in 3dmark 2k.

Anyways, as long as sufficient cooling will be dealt I am going to get one of those suckers. There has to be some price to pay for such a great performance boost and I gues heat was one of them AGAIN.

 

[LordSandMan]

No, you didn't tell me to do that, but it works great. I figured it out because I couldn't touch the back side of my GeForce board without burning my fingers. It runs much cooler now. I also put an 80mm fan on it.

 

[Mr.IncrediblyBored]

You guys seem surprised. Even on a .15um process, 55 million is a lot of transistors and Its going to dissipate some major wattage.

Edit: I doubt nVidia will add more texturing units per pipeline. Adding register combiners for per-pixel operations will be/is more important.

 

[DaveB3D]

Na, I've heard 6/2 with a single loopback per texture unit.. so they can do 4 textures in a single pass with 2 clocks. With NV25, I've heard 8/2 with a single loopback.

 

[NFS4]

You keep spilling the beans on the NV20...I say for every spec you leak about the NV20, you should give us one for the Rampage

 

[RagingGuardian]

Why do you doubt they'll add more texture units per pipeline? Even it is to match the fact that the Radeon features 3 texture units per pipeline I'm sure they'll do it. Besides I'm sure there are games on the horizon that will take advantage of the extra texture unit(s).

 

[DaveB3D]

3 texture units is stupid (IMO) and I've yet to be able to figure out for the life of me why ATI ever even considered doing that.

In my thoughts the my best guess as to why they'd do it is for EMBM. That is really all you are going to be using 3 texture passes for.

However, when DX8 comes into the picture 3 texture units is really dumb. Consider:

DX8 is based around the idea of 4 texture passes.

So in order for ATI to do the typical DX8 stuff (that it is able to do), they must do their first pass and process 3 of the textures and then do another pass and do the last texture. So in doing that second pass you waste two of your texture units. It just becomes a waste of silicon really because you are doing two passes either way. They'd have been better off dropping one of the pipes and adding a loopback.

So I think it is pretty clear why that is so strange to do that. It just doesn't seem very smart to me. No?

 

[Mr.IncrediblyBored]

Ya learn something new every day. How does adding loopback differ from current multipass rendering schemes?

When the VSA does multipass rendering, does it write the pixel to the framebuffer? And then re-read the current pixel/texel value and send it through the pipeline again (costing memory bandwith)?

Loopback allows the chip to not need to write the pixel to the framebuffer before the second pass, correct?

Is this right, or don't I have any idea what I'm talking about (I have a feeling it's the latter)

Edit: Dave, where'd ya go?

 

[JayPatel]

i had my original geforce 1 clocked at 172/225 stably for its entire life in my system.

 

[DRGrim]

Can you put a new heatsink/fan on a video card without messing with the waranty? You should, because it just makes it less likly the card will overheat. I don't want to mess with mine, because I have very bad luck with computer parts, they are always breaking. Thanks

 

[DaveB3D]

Sorry. Went to the grocery store. A guy has got to have food ya know.

First, on multi-texturing in general. That is how it really works. You write a pixel to the frame-buffer, read it back and write the second texture pass if you don't have enoufg texture units. Assuming you have enough texture units, you just write and then you're done. This isn't VSA-100 exclusive though. That is one of the advantages of deferred rendering. You write everything directly to the tile buffer.

Typically (I think on every current architecture) requires a resend of geometry in order for multi-pass rendering outside of the texture unit limit (basically reading back from the FB). So if you can do 2 textures in a single pass, in order to do 4 passes you send the geometry through once and do your passes and then send it through again and do the two additional passes.

Now that said, a loopback is exactly how it sounds. You send through, write your texture and instead of writing to memory you flip around in a second clock cycle you write the next texture pass. Now I think this requires a bit more FIFO memory, but I'm not 100% sure (I forget and don't feel like checking right now ).

 

[Mr.IncrediblyBored]

Ok, that's what I thought. Thanks.