techspot Rumored flagship Nvidia card 48GB GDDR6X, 18,176 CUDA cores, and 800W TDP

[UltraMega]

Quote

Regular leaker kopite7kimi posted a tweet about what he appropriately nicknamed "the beast." The card is supposedly based on a new PG139-SKU0 board design, rather than the PG137 expected to be utilized by the RTX 4090, and could be the first to feature a triple-fan reference cooler from Nvidia.

 

kopite7kimi believes the card will pack an AD102-450-A1 GPU featuring 142 streaming multiprocessors (SMs) on 18,176 CUDA cores. That's still a cut-down version of the full-fat AD102 GPU, which packs 144 SMs on 18,432 CUDA cores.

Rumored flagship Nvidia Lovelace card to pack 48GB of GDDR6X, 18,176 CUDA cores, and 800W TDP

WWW.TECHSPOT.COM

---

Regular leaker kopite7kimi posted a tweet about what he appropriately nicknamed "the beast." The card is...

If this is where the 800w rumors are coming from, that's makes it more reasonable. 

 

 

[Sir Beregond]

I just hope gamers quit trying to call it a "Titan card". Prosumers don't want 800w cards that don't have Titan drivers which are a bit of an in-between GeForce and Quadro.

 

Will be a ludicrously expensive halo GeForce product I am betting.

 

On another note, 48GB does make sense. Up until Pascal/Turing/Ampere, really starting with Fermi -> Kepler -> Maxwell -> Pascal, each tier of card was usually doubling in VRAM gen on gen. For instance, the midrange cards of each: GTX 460/560 at 1GB, GTX 680 at 2GB, GTX 980 at 4GB, GTX 1080 at 8GB. Yes I am calling those mid-range based on the chip.

 

Same was happening with the 80 Ti cards and Titans for a while, until it wasn't. They really stagnated the VRAM after Pascal.

[Avacado]

47 minutes ago, UltraMega said:

Rumored flagship Nvidia Lovelace card to pack 48GB of GDDR6X, 18,176 CUDA cores, and 800W TDP

WWW.TECHSPOT.COM

---

Regular leaker kopite7kimi posted a tweet about what he appropriately nicknamed "the beast." The card is...

If this is where the 800w rumors are coming from, that's makes it more reasonable. 

 

 

[pio]

Until they bring their prices back down to Earth levels, I really have to say I'm out.  $1000-$1200 was a tough spend for a high end GPU, but okay, fine...inflation and all that.  $3000+ these days, yeah no.  Those price ranges were okay for professional level cards, but not for gaming cards.  I consider the Titan more of a gaming card than a pro card anyway, it wasn't a Quadro.

 

Of course as an enthusiast, I don't care as much about the power consumption.  So that's not really an issue.  If its 800w, its 800w, whatever.  It's not like other GPU's haven't drawn a ton of power, or CPU's either for that matter.

 

Either way, guess we'll wait and see how things go.  There's rumors going around on youtube that Nvidia might end up getting stuck with the next gen cards and not being able to sell them off.  *shrugs

Edited July 26, 2022 by pioneerisloud

[ozlay]

It probably needs 200w just for the vram. Well i guess the 3090 was around 82w. So maybe 100w. 

Edited July 26, 2022 by ozlay

[bonami2]

Wonder how folding will run on those. And at 800w if it really the rated power the power bill is going to be insane.

[J7SC_Orion]

I am wondering if those 48 GB are using 4 GB VRAM modules, or 2 GB VRAM per...if it is the latter, it might be back to double-sided VRAM special backplate cooling, a la 3090... 

[Sir Beregond]

I do care about power draw. If a card is twice as fast as a previous gen card, but uses twice or more power to get it, that just tells me the card sucks as a gen on gen improvement architecturally. You're just brute forcing at that point.

Edited July 27, 2022 by Sir Beregond

[bonami2]

6 hours ago, Sir Beregond said:

I do care about power draw. If a card is twice as fast as a previous gen card, but uses twice or more power to get it, that just tells me the card sucks as a gen on gen improvement architecturally. You're just brute forcing at that point.

I think they are just bruteforcing. It like intel throwing e-core at tdp limited mobile cpu.

Edited July 27, 2022 by bonami2

[Fluxmaven]

21 hours ago, Sir Beregond said:

Will be a ludicrously expensive halo GeForce product I am betting.

 

Same was happening with the 80 Ti cards and Titans for a while, until it wasn't. They really stagnated the VRAM after Pascal.

 

Yup... Used to just be able to do a lil strange for some change to afford a new GPU. Now you gotta be a full time onlyfans model on top of a day job to keep up with the latest tech. 

 

On a more serious note, I am glad to see them pushing more VRAM. Helps keep cards relevant for much longer. I still have a Titan Black that I was actively using up until a year ago. The 780ti with only 3GB became a paperweight much sooner.

[Avacado]

15 hours ago, pioneerisloud said:

Until they bring their prices back down to Earth levels, I really have to say I'm out.  $1000-$1200 was a tough spend for a high end GPU, but okay, fine...inflation and all that.  $3000+ these days, yeah no.  Those price ranges were okay for professional level cards, but not for gaming cards.  I consider the Titan more of a gaming card than a pro card anyway, it wasn't a Quadro.

 

Of course as an enthusiast, I don't care as much about the power consumption.  So that's not really an issue.  If its 800w, its 800w, whatever.  It's not like other GPU's haven't drawn a ton of power, or CPU's either for that matter.

 

Either way, guess we'll wait and see how things go.  There's rumors going around on youtube that Nvidia might end up getting stuck with the next gen cards and not being able to sell them off.  *shrugs

Flagship card at launch is usually a 90. If the 3090 costs $1299 new now, I would expect to see a 4090 settle in the $1499-1599 range at launch assuming perf increases hold true. Don't forget, GPU demand is at an all time low right now and release pricing will reflect that. I wager we'll see a $1899 Ti/KPE variant. I do not think prices will crest 2k.

Edited July 27, 2022 by Avacado

[Sir Beregond]

15 minutes ago, Fluxmaven said:

 

Yup... Used to just be able to do a lil strange for some change to afford a new GPU. Now you gotta be a full time onlyfans model on top of a day job to keep up with the latest tech. 

 

On a more serious note, I am glad to see them pushing more VRAM. Helps keep cards relevant for much longer. I still have a Titan Black that I was actively using up until a year ago. The 780ti with only 3GB became a paperweight much sooner.

 

Yep, its why I stopped upgrading every gen. Just not worth it to me.

 

6 minutes ago, Avacado said:

Flagship card at launch is usually a 90. If the 3090 costs $1299 new now, I would expect to see a 4090 settle in the $1499-1599 range at launch assuming perf increases hold true. Don't forget, GPU demand is at an all time low right now and release pricing will reflect that. I wager we'll see a $1899 Ti/KPE variant. I do not think prices will crest 2k.

 

Yeah, I've been wondering about this too the past few days. I'm on the fence about how Nvidia will handle pricing, but what you say does make sense.

[UltraMega]

It's interesting to see this rumor and think about the huge jump in vram, but it's not relevant to gamers IMO. 48GBs of vram is just pure overkill and by a lot. Modern game engines are proving that you can take a huge amount of super highly detailed objects made with photogrammetry and run totally wild with that, and the vram amounts we have now work fine for that. There are some Quixel demos running on a 1080Ti that are extremely impressive. We just don't need that much vram for games with how things boil down now. Plenty of ram already to have tons of assets on screen and SSDs are fast enough to make loading more assets in real time a non-issue. Already being at the point where 3D models are as detailed as we need them to be, the rest mostly comes down to lighting and artistic design. 

 

As @Sir Beregond pointed out though, 48GBs is not that surprising, especially when you consider it's only a doubling of what the 3090 has now, and that the rest of the 3000 series cards were supposed to launch with double the amount of ram. But by the time we get anywhere close to needing that much vram for games, the GPU won't be anywhere close to holding up because it will be a very long time. 

 

Pricing though, if they do release a 48GB card, I think they will definitely use it as an excuse to charge a massive premium for it. They might price some of their other cards more reasonably this time around, but they also want to have something out there that's mostly just bragging rights for people with more money than sense. 

Edited July 27, 2022 by UltraMega

[Avacado]

14 minutes ago, UltraMega said:

Pricing though, if they do release a 48GB card, I think they will definitely use it as an excuse to charge a massive premium for it. They might price some of their other cards more reasonably this time around, but they also want to have something out there that's mostly just bragging rights for people with more money than sense. 

You could be right. I mean GDDRX6 has been in fab for awhile now. I don't know what current cost is per GB, but I wager it's cheaper to produce now than it was at launch. I imagine if it holds true, well be seeing 2GB VRAM IC's (For the first time) 12 front/12 back. 

 

*Edit I could not find manufacture cost for GDDRX6, but I found it for GDDR6. It's not THAT costly per chip.

 

https://www.guru3d.com/news-story/gddr6-significantly-more-expensive-than-gddr5.html

Edited July 27, 2022 by Avacado
Added info

[UltraMega]

20 minutes ago, Avacado said:

You could be right. I mean GDDRX6 has been in fab for awhile now. I don't know what current cost is per GB, but I wager it's cheaper to produce now than it was at launch. I imagine if it holds true, well be seeing 2GB VRAM IC's (For the first time) 12 front/12 back. 

 

*Edit I could not find manufacture cost for GDDRX6, but I found it for GDDR6. It's not THAT costly per chip.

 

https://www.guru3d.com/news-story/gddr6-significantly-more-expensive-than-gddr5.html

I dont see AMD releasing a card with 48GBs meaning it will be a unique high end product, and this is Nvidia we're talking about. It will probably be the worst price/performance card by far and still be exuberantly priced in a way that doesn't reflect the manufacturing cost. 

Edited July 27, 2022 by UltraMega

[Sir Beregond]

17 minutes ago, UltraMega said:

It's interesting to see this rumor and think about the huge jump in vram, but it's not relevant to gamers IMO. 48GBs of vram is just pure overkill and by a lot. Modern game engines are proving that you can take a huge amount of super highly detailed objects made with photogrammetry and run totally wild with that, and the vram amounts we have now work fine for that. There are some Quixel demos running on a 1080Ti that are extremely impressive. We just don't need that much vram for games with how things boil down now. Plenty of ram already to have tons of assets on screen and SSDs are fast enough to make loading more assets in real time a non-issue. Already being at the point where 3D models are as detailed as we need them to be, the rest mostly comes down to lighting and artistic design. 

 

As @Sir Beregond pointed out though, 48GBs is not that surprising, especially when you consider it's only a doubling of what the 3090 has now, and that the rest of the 3000 series cards were supposed to launch with double the amount of ram. But by the time we get anywhere close to needing that much vram for games, the GPU won't be anywhere close to holding up because it will be a very long time. 

 

Pricing though, if they do release a 48GB card, I think they will definitely use it as an excuse to charge a massive premium for it. They might price some of their other cards more reasonably this time around, but they also want to have something out there that's mostly just bragging rights for people with more money than sense. 

 

Yeah gamers won't need a 48GB card. But if its not actually a Titan (and I don't think it will be given power draw rumors), it will be a 48GB, expensive, gaming GeForce card. Which is exactly what the 3090 was. Jensen sure did a good job calling it "Titan class" and so many fell for it being a "Titan replacement", when it was never that. Marketing at its finest...

 

Still the rest of the cards have stagnated for multiple gens now. 8GB 3070 was kind of ridiculous.

[J7SC_Orion]

...I still wonder about 4090 vs 4090 Ti, and related pricing (never mind heat management of the rumored  wattage). I also still think that AMD's mGPU approach for next gens might have spurred NVidia into these monsters (if they materialize), even if AMD's tile approach will initially be different and not threaten NVidia's 'GPU performance leadership'.