NV40 Test samples, pics inside

I think the 'its only got 4 caps' comment was coming from somebody who either knows a whole lot about the design(and knows that it should have more based on the design), or knows nothing much.

But, you can't simply count caps and say 'more is better'.

As for what capacitors are used for....storing energy.

When the circuit needs more instantaneous power than can be provided by the power supply, the voltage will sag, and you'll get brownouts (the chip starts working erratically if the voltage goes below the min operating voltage), or noise, or both.

Capacitors are used to smooth out the voltage by providing that instantaneous power when needed. They store up power, and provide it to cover the bursts that the circuit needs, then store up more. The bigger the cap, the more storage, the better the power isolation.

Thats the basic 'laymans' view. Not quite technically complete or correct, but enough to explain the idea.

EDIT: also note that capacitors come in other packages than those aluminum cans. They also come (in smaller capacitances) in surface mount, "orange squares". There may be more of them on the back of the PCB to spot handle problems (closer to the chip thats drawing the power), while the 4 large cans manage the overall power draw to the system.

Putting the capacitors closer to the power pins of the chip help to avoid voltage sag because the voltage drop is directly propertional to current * resistance. If you've got a short wire between the voltage 'source', the current can go instantaneously higher without dragging the voltage down.

EDIT#2: If you look at the picture closely, you'll see a lot of those "orange squares" (well, they look more brown in the picture), just to the left of the cans, just to the right of that strange black line (of which I have no idea what it is).

 

i´m gonna take a big guess using my little electronics ¿knowledge? and say that 4 caps seem too little for a card requiring 150W, since caps are used, for example, in circuitry that involves power regulation (as for example the caps you see on a motherboard near the mosfets), and 150"clean"W, mixed between 3.3V, 5V and 12V would require lots of power regulation.... BUT (hehe)... what if the need of two different power sources is providing clear enough power hence you only need to regulate and stabilize the signal a little bit, compared to what needed to be done before?

EDIT: russ beat me to it, anyway his explanation is more technically accurate as to what CAPS do

 

Thanks Joe, Russ, and MatiasZ for the explanations. I do understand capacitors, I learned about 'em the hard way when I was a young lad learning how to build breadboards with me bro and he asked me to lick the discharge terminals on a small one.

Two possibilities suggest themselves to me:

1. Simple answer, the capicitors are hidden somewhere underneath that cooler. (I doubt it, I think caps get hot and we don't want hot things in a cooler)

2. With two dedicated molex leads it has more power to draw on and doesn't need to build up as big of a reserve charge.

Mebbe?

 

Could be anything. However, I'd presume that whatever the solution is, it'll work OK. Nobody (from consumer to retailer, to OEM, to silicon vendor) likes returns.

 

I think the problem is current through the power connector wires. It looks like the wires are 20-gauge or thinner, and I don't think they could take more than 4 amps safely (even that's pushing it - I've seen anywhere from a conservative 1.5 to 5+). You have a 12V line and a 5V line, so that's 68W max per plug, assuming perfect power conversion to your desired voltages. You can only get so much power from AGP (the 9700 needs a connector and consumes 54 W), and from what we know NV40 needs gobs of power (120W? 150W?).

I guess that means two molex connectors coming from different power supply lines. Hope there's enough left for other system components

Does anyone think ATI was holding back on 16-pipes because of the extra power connector and unreasonable power supply requirements? Maybe they saw NVidia doing it and just needed a few modifications to the PCB for this "sudden redesign". Just a thought.

 

No, the caps shouldn't get hot. In fact, they dislike getting hot and express their dislike by spewing the electrolyte across the board. (Witness the recent debacle with el cheapo electrolyte caps at motherboards...)

No, doesn't matter. Doubling the power lines halves the resitance of the supply wires, but it's still WAY too high to rely on the ridiculously large caps at the supply. As Russ said, caps should be placed as close to the chip as possible.

Better regulation (that doesn't need as many filtering capacitors)? Better caps? Supply line noise tolerant chip? Trigger-happy bean counters? Take Your pick...

 

ROFL!

 

As far as I'm aware, ATI won't be using 2 power supplies.

 

This should be fun

Well I think NVIDIA would have been much better off with a two slot design than messing around and dissipating 120W into the case. Not with the new P4 chips putting out the heat they are now. Most Enthusiasts know not to rely on that PCI slot close to the AGP slot anyway. Mines filled with a VANTEC slot cooler which is a properly designed centrifugal fan.

Adding a molex connector might have been somthing they did to offset the costs of their PCI express bridge, capcitors especially good quality ones cost money and with two power leads they will be less reliance on capictors. You remember partners where allegedly bitching about the cost of the PCI express bridge. They also may be planning to for econmies of scale reasons staying with the same basic board design for the next generation which will need the draw of two power leads.

Personally if they produce a card that is only a few % faster than ATI that will have a cooler card and only one molex connector most people will opt for the simple choice. Unless of course they make a really good case on the PS 3.0 shader support issue. (That is if it performs well, just because they support doesnt mean it will work as good as it can.)

 

Yeah, yuck it up. Remind me to tell you sometime about the important lesson he taught me as a 10 year old about lighters and a pile of blackpowder we stole from my father's flintlock.

I think you'd like me bro Dave, he's "diggah" over at EB.

 

Capacitors don't like to be in a hot environment where they might break. Luckily they actually don't get particualrly hot.

Maybe, with a good power supply you don't need as much filtering. However, good filtering is traditionally needed in the analog output of the card (for analog monitors and tv-out). Less capacitors for the output filtering stages usually means lousy 2D image quality - but maybe nVidia has pulled an ace from their hats. Something to do with the small capacitors on the board, perhaps?

 

Re: This should be fun

I don't think you fully understand the problem.

 

Dont hold out on me.

 

For neither the 12-pipe nor 16-pipe versions? Impressive. I don't think you can safely get 100W+ through one molex and agp.

Is all that Inquirer stuff BS then? If ATI needs a 450W power supply (which, admittedly, is not likely IMO), they have to get that extra current to the GPU somehow.

 

Unless they somehow created a design that doesn't need that kind of power requirements. They may have created a more elegant solution to the problem than NVIDIA, or they may have been willing to sacrifice other things, like the ability for good overclocking..

 

Where did you pull the "100W+" figure from for the R420? I was under the impression that low-k made the power requirements of the card lower and that it was going to draw power much more in the double-digit range.

 

I kinda wish I hadn't read the boards the last few days, because it's completely ruined any enthusiasm I had for NV40. It just seems like whilst it will perform well enough (unlike NV30), the performance will come at a cost not worth paying over and above XT (or perhaps even Pro, eurgh).

Damn you all.

 

Actually, I always thought the small orange SMD capacitors were in the nF - 1uF range and are used for decoupling, signal-level voltage regulation, etc. at most...

 

Does Low K reduce the amount of heat by reducing the power needed? I had a good article on Low K somewhere, got to find it.

http://nepp.nasa.gov/index_nasa.cfm/934/

So less resistance as less current leak sounds like less current needed to me but I am just a layman, any experts out there.

P.S. Low K is dangerous isn't it????? LOL

 

Stryyder, digitalwanderer, that's my point. If there's only one molex, there must be much lower power consumption since 100W+ is probably impossible, and that's why it's impressive. Low power consumption (relatively) is the only possibility.