NVidia Ada Speculation, Rumours and Discussion

[Albuquerque]

TDP is pretty much meaningless with out knowing the amount of chip area the heat is spread over.

Depends on what we're talking about.

At the end of the day, if chip consumes 800W and is either one square millimeter or 100 square millimeters, it's still 800W of heat that needs to be relocated to the surrounding air. The density of that heat matters for proper heatsink development and materials science (ie make sure the chip itself doesn't burn up from that power), yet it's all still just heat. The heatsink could be a 12" forged aluminum pan with four tablespoons of butter sizzling in it, ripe for a 16-ounce bone-in well marbled 2" thick dry-aged ribeye to stop in for ~90 seconds on each side to get to a fancy crispy outer crust and a warm medium rare on the inside.

You could obviously construe an argument where the chip is the size of a picnic table and thus 800W becomes somewhat pointless. That's also a logical fallacy as the chip sizes we're talking about are constrained to the reasonable size of an ATX-compatible PCB mounting scheme

 

[Bondrewd]

TDP is pretty much meaningless with out knowing the amount of chip area the heat is spread over.

You're still getting 600mm^2-ish die tops since it's the fucking N5.
Expensive as shit!

 

[Kyyla]

500W is enough to warm one room in a well insulated Scandinavian home when it's very cold (below 0F). Not many people want an extra space heater inside their study.

 

[Granath]

500W is enough to warm one room in a well insulated Scandinavian home when it's very cold (below 0F). Not many people want an extra space heater inside their study.

Simply open a window.

 

[DegustatoR]

Or buy a 4070.

 

[Kyyla]

Simply open a window.

Yeah great logic. Stupid consumers not wanting to live in a sauna. Sure you can build undesirable products, don't be surprised when they don't sell.

 

[DegustatoR]

Yeah great logic. Stupid consumers not wanting to live in a sauna. Sure you can build undesirable products, don't be surprised when they don't sell.

Like 3080s and 3090s which don't sell right now?

 

[no-X]

Kyyla is talking about consumers, not miners.

 

[DegustatoR]

Kyyla is talking about consumers, not miners.

Same consumers who are camping out over nights in front of stores in hopes of getting some 3080s and 3090s before miners?
Power consumption is mostly irrelevant. What matters is price/performance and consumer facing qualities like noise under load.

 

[trinibwoy]

500W is enough to warm one room in a well insulated Scandinavian home when it's very cold (below 0F). Not many people want an extra space heater inside their study.

In that case wouldn’t they just buy a card with the power consumption that meets their needs? Nobody is forced to buy a 500w graphics card to play games.

 

[Kyyla]

Yes not many people will.

 

[Rootax]

It will not solve the situation, but nVidia and AMD need to be more precise with voltages too. Now it seems that they put it higher that needed so a lot of chip pass the validation process. I get it, but in consequence we have chips having more voltage than needed and consuming more power for nothing.

With small to moderate undervolting, you can still have stability, and saving 10-15% watts in most situations. Even more if you willing to spent a lot of time finding the perfect spot.

Idk how they can fine tune that in factories...

 

[trinibwoy]

It will not solve the situation, but nVidia and AMD need to be more precise with voltages too. Now it seems that they put it higher that needed so a lot of chip pass the validation process. I get it, but in consequence we have chips having more voltage than needed and consuming more power for nothing.

With small to moderate undervolting, you can still have stability, and saving 10-15% watts in most situations. Even more if you willing to spent a lot of time finding the perfect spot.

Idk how they can fine tune that in factories...

Even if it was easy to assign the perfect voltage for each individual card at the factory it would create chaos as people would try to play card lottery to find a “good chip” and it would be RMA city. Much easier to just slap a standard voltage curve on all cards and call it a day.

Having said that you reminded me that I need to try under-volting too. Most of the annoying noise in my PC comes from the CPU cooler but reducing GPU fan noise can’t hurt.

 

[Scott_Arm]

It will not solve the situation, but nVidia and AMD need to be more precise with voltages too. Now it seems that they put it higher that needed so a lot of chip pass the validation process. I get it, but in consequence we have chips having more voltage than needed and consuming more power for nothing.

With small to moderate undervolting, you can still have stability, and saving 10-15% watts in most situations. Even more if you willing to spent a lot of time finding the perfect spot.

Idk how they can fine tune that in factories...

100% agree especially with ampere. The could have stock voltage lower. I run my card at 800mV at around 1860MHz but there’s a wide range of safe voltage and frequencies between that and stock with marginal performance decreases.

 

[Albuquerque]

Yeah, my 3080Ti loves a bit of healthy undervolting. It can't get quite as far down as Scott (not sure which Ampere you've got) however 850mv is good for 1800MHz stable in CyberPunk 2077 -- which turned out to be the most strenuous GPU stability tester I own.

 

[techuse]

https://wccftech.com/nvidia-geforce-rtx-40-ada-lovelace-gpus-september-launch-upto-850w-tdp-rumor/

Is this even believable?

 

[Scott_Arm]

Yeah, my 3080Ti loves a bit of healthy undervolting. It can't get quite as far down as Scott (not sure which Ampere you've got) however 850mv is good for 1800MHz stable in CyberPunk 2077 -- which turned out to be the most strenuous GPU stability tester I own.

I actually ended up changing mine to something like 1960 @ I think 900 mV. It made a bigger difference then I expected in particular games. Halo Infinite is particularly hard on gpus, and the extra 100 MHz made a big difference. I was probably on the edge of a bottleneck that I alleviated to some extent. In general I prefer the 800 mV profile I have in most games.

 

[DegustatoR]

https://wccftech.com/nvidia-geforce-rtx-40-ada-lovelace-gpus-september-launch-upto-850w-tdp-rumor/

Is this even believable?

We have 1500W PSUs so if the question is if it would be possible then I don't see why not.

The real question though is why would this be necessary.

 

[Scott_Arm]

Get the electrician to install a dedicated breaker for the pc. Trending in a dumb direction.

 

[Silent_Buddha]

Get the electrician to install a dedicated breaker for the pc. Trending in a dumb direction.

Yup, if the released products actually use 850 watts (probably more when stressed, like Ampere) then that'll be more than the first Microwave that I owned (800 watts). And it'd be running for hours on end unlike my old microwave. Then a full PC would potentially be pulling as much wattage as my 1200 watt microwave. Although again, that would be running for potentially hours on end rather than just minutes at a time.

Absolutely ridiculous if true.

Regards,
SB