NVIDIA Provides a Statement on MIA RTX 3090 Ti GPUs

[Raevenlord]

NVIDIA's RTX 3090 Ti graphics card could very well be a Spartan from 343 Industries' Halo, in that it too is missing in action. Originally announced at CES 2022 for a January 27th release, the new halo product for the RTX 30-series family even had some of its specifications announced in a livestream. However, the due date has come and gone for more than half a month, and NVIDIA still hadn't said anything about the why and the how of it - or when should gamers hoping to snag the best NVIDIA graphics card of this generation ready their F5 keys (and bank accounts). Until now - in a statement to The Verge, NVIDIA spokesperson Jen Andersson said that "We don't currently have more info to share on the RTX 3090 Ti, but we'll be in touch when we do". Disappointed? So are we.

While the reasons surrounding the RTX 3090 Ti's delayed launch still aren't clear - and with NVIDIA's response, we're left wondering if they ever will be - there were some warning signs that not all the grass was green on the RTX 3090 Ti's launch. The consensus seems to be that NVIDIA found some last-minute production issues with the RTX 3090 Ti, which prompted an emergency delay on the cards' launch. The purported problems range from issues with the card's PCB, BIOS, and even GDDR6X 21 Gbps memory modules - but it's unclear which of these (or perhaps which combination) truly prompted the very real delay on the product launch.



View at TechPowerUp Main Site

 

[Space Lynx]

good, make the elites suffer like the rest of us

 

[HD64G]

Power draw was too high for many components, maybe even for the pcb itself to withstand the heat. My 5c.

 

[Nater]

I still say it's because no Radeon 6950XT showed up. There's no reason for it in this market, especially if they're having issues producing one, which it seems they are.

 

[Dr. Dro]

Gee, I have no idea why

From my perspective as a 3090 owner - and well aware of its issues and shortcomings - is this really surprising to anyone? TGP is too high. Everything is too high, and taken too far. Memory density has doubled so they could halve the amount of devices used, and the G6X controller on GA102 itself already came at pretty much its maximum as it was - 3090's 21Gbps memory shipped at 19.5Gbps to give end users an illusion that it could overclock, yet even the cards with excellent controllers (mine would count, probably) die off at exactly that 21 Gbps or thereabouts mark and begin showing very heavy signs that they are past their limit (the v/f curve shoots to the moon and it begins erroring out heavily).

Add extra power consumption that arises from the extra 2 SMs enabled and the jacked up clock speeds plus NVIDIA's hilariously bad automated power management system, this Ti card is a disaster. It will take a lot of adjustments, extreme binning and I would honestly not be surprised if this model ends up cancelled altogether.

 

[Oberon]

Power draw was too high for many components, maybe even for the pcb itself to withstand the heat. My 5c.

More like 105C, amiright? lololol

... I'll see myself out.

 

[mb194dc]

Power draw was too high for many components, maybe even for the pcb itself to withstand the heat. My 5c.

Igor labs suggested power draw from the board would be out of spec (8 amps instead of 5.5 iirc), no way to prevent it without redesigning the pcb.

 

[ir_cow]

They probably realized releasing another video card that cost 3k is stupid. Especially when 40 series and AMDs cards are due out later this year.

 

[owen10578]

I'm calling it. The 21gbps GDDR6X and fully unlocked cores make fore too high power draw and being uncoolable for the Founder's Edition. The design starves the PCB of any direct airflow.

 

[mama]

Hardly "a Statement". More like a non statement.

 

[Dr. Dro]

I'm calling it. The 21gbps GDDR6X and fully unlocked cores make fore too high power draw and being uncoolable for the Founder's Edition. The design starves the PCB of any direct airflow.

The 3090 already has 21 Gbps chips, this is not the problem. Fully unlocked core, by itself, should not present much of a difference (it exists, but it should be minor in the grand scheme of things and if they just kept the 3090 settings with the full 84 SM die, it would be feasible), since they would technically be of a higher quality bin, but jacking up the v/f curve and adding an extra 100 watts of power limit by default is a BIG difference.

I have a friend that owns a Strix OC 3090, he bought it about a month after I bought my TUF OC (that was on launch day), his card has a 105 watt higher power limit compared to mine (375 W vs. 480 W), and his card has pretty much always shown signs of instability and getting _very_ moody when running at the 450W+ mark or thereabouts, and that is because he runs a 1600 T2, next to which even my 1300 G2 is a low end PSU. It just isn't feasible, imo, to run these on aircooling or even ambient-temperature water at such high wattage ranges. They become unpredictable fast and even though some people might be able to get a stable system, I doubt they would pass the strict validation for 24/7 operation that the install-and-forget setting that NV goes for requires.

I just wish BIOS editors for these boards were released. I hate NVIDIA's GPU boost mechanism, it's a blight and it can never beat a human operator... the workaround for setting the boost target so low that it pretty much always operates in the highest bin is not really optimal, IMO.

 

[TheoneandonlyMrK]

"We don't currently have more info to share on the RTX 3090 Ti, but we'll be in touch when we do"

That's a statement, No comment, we might later. Wtaf.

 

[zlobby]

Gee, I have no idea why

From my perspective as a 3090 owner - and well aware of its issues and shortcomings - is this really surprising to anyone? TGP is too high. Everything is too high, and taken too far. Memory density has doubled so they could halve the amount of devices used, and the G6X controller on GA102 itself already came at pretty much its maximum as it was - 3090's 21Gbps memory shipped at 19.5Gbps to give end users an illusion that it could overclock, yet even the cards with excellent controllers (mine would count, probably) die off at exactly that 21 Gbps or thereabouts mark and begin showing very heavy signs that they are past their limit (the v/f curve shoots to the moon and it begins erroring out heavily).

Add extra power consumption that arises from the extra 2 SMs enabled and the jacked up clock speeds plus NVIDIA's hilariously bad automated power management system, this Ti card is a disaster. It will take a lot of adjustments, extreme binning and I would honestly not be surprised if this model ends up cancelled altogether.

Preach!

 

[TheoneandonlyMrK]

Igor labs suggested power draw from the board would be out of spec (8 amps instead of 5.5 iirc), no way to prevent it without redesigning the pcb.

I'm calling it. The 21gbps GDDR6X and fully unlocked cores make fore too high power draw and being uncoolable for the Founder's Edition. The design starves the PCB of any direct airflow.

I'm calling it, it's the power's

 

[zlobby]

Igor labs suggested power draw from the board would be out of spec (8 amps instead of 5.5 iirc), no way to prevent it without redesigning the pcb.

And the neighborhood's power grid as well...

 

[Vya Domus]

These products are in the pipeline long before they're actually released but sometimes that doesn't quite work out very well. I bet they simply can't find a reason in today's market state for this thing to exist. They make money hand over fist on every single product because they have a class of customers that will buy as much as they can manufacture, miners. It's highly possible they just can't be bothered, from a pragmatic perspective who cares about this, why waste money on marketing and working with AIB for a completely irrelevant product.

 

[windwhirl]

Power draw was too high for many components, maybe even for the pcb itself to withstand the heat. My 5c.

Wasn't this going to be the 900 W peak power draw card or something?

 

[trsttte]

Nvidia provides a statement by not providing a statement!?

C'mon guys, is this wccftech now?

 

[DeathtoGnomes]

The excuse for the delay could be anything, even as simple as bad solder joints.

 

[mechtech]

That guy has one in his hand, how could they be missing??!!

 

[HisDivineOrder]

The best part is how you know that Nvidia has openly stopped caring about most enthusiasts when they don't even attempt to lie and tell us they cancelled the product to focus on delivering more cards to the masses. Like, the lie is there, so easy. Just waiting for them to take it and spin it and you'd have an army of fans ready to fall on that hill.

Instead, they're CROATOAN'ing that card.

 

[AusWolf]

No surprise here considering that the whole 30 series is a flop in terms of efficiency and production yields. So far, they've probably managed to make two fully working GA102 chips, and even those are waiting in some warehouse for stronger power delivery and cooling solutions to be invented.

I'm not sour - I just miss the days when <75 W meant entry-level, 75-200 W mid-range and >200 W enthusiast-class. I guess, it's still true in one sense, just model numbers have shifted towards x50-60 still being mid-range, everything above enthusiast-class, and entry-level completely disappeared. It's even sadder to look at it from here.

 

[nguyen]

Maybe Nvidia should shove these 3090 Ti into Geforce Now servers, like they did with the supposed 2080 Ti Super.

 

[Hyderz]

its already 16 months since the launch of rtx 30 series, time to bring out next series.
Please make the card run cooler and draw less power please, so that 3rd party doesn't have to mount triple slot coolers

 

[eidairaman1]

Gee, I have no idea why

From my perspective as a 3090 owner - and well aware of its issues and shortcomings - is this really surprising to anyone? TGP is too high. Everything is too high, and taken too far. Memory density has doubled so they could halve the amount of devices used, and the G6X controller on GA102 itself already came at pretty much its maximum as it was - 3090's 21Gbps memory shipped at 19.5Gbps to give end users an illusion that it could overclock, yet even the cards with excellent controllers (mine would count, probably) die off at exactly that 21 Gbps or thereabouts mark and begin showing very heavy signs that they are past their limit (the v/f curve shoots to the moon and it begins erroring out heavily).

Add extra power consumption that arises from the extra 2 SMs enabled and the jacked up clock speeds plus NVIDIA's hilariously bad automated power management system, this Ti card is a disaster. It will take a lot of adjustments, extreme binning and I would honestly not be surprised if this model ends up cancelled altogether.

Face it, Ampere certainly sucks Amperes.

its already 16 months since the launch of rtx 30 series, time to bring out next series.
Please make the card run cooler and draw less power please, so that 3rd party doesn't have to mount triple slot coolers

Fermi 3.0 anyone