NVIDIA's next-gen Titan RTX specs teased: 48GB GDDR6X

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
10,785
"The cost of this card would be mammoth, somewhere in the $3000 to $3500 range. Keep in mind the Turing-based TITAN RTX with 24GB of GDDR6 costs $2499, and the new Ampere-based GeForce RTX 3090 with 24GB of GDDR6X costs $1499.

NVIDIA would be waiting in the wings with its huge Ampere-based TITAN RTX, just in case Bigger Navi launches and poses any form of threat to NVIDIA's reign at the top of the graphics card mountain.

Note: Micron is only just now debuting GDDR6X memory at 19.5Gbps on the GeForce, while the GDDR6X spec itself ramps up to 21Gbps. I will have an overclocking article within 24 hours of my GeForce RTX 3080 Founders Edition review -- expect some investigations in that."


Read more: https://www.tweaktown.com/news/7514...s-teased-48gb-gddr6x-and-over-3000/index.html
 
Holy moly.

giphy.gif


Too rich for my blood, but that's a lot cheaper than I was anticipating.
 
LOL, if this is true doesn't this make that affordable 3090 not a Titan but a marked up TI. Yup Nvidia would do that, wow the intent in their marketing makes me doubt if they day the sky is blue.
 
[The news is coming from Moore's Law is Dead]

[NVIDIA could use it on a next-gen TITAN RTX graphics card.]

So some YouTube dude's 100% speculation about what Nvidia might do with faster Micron memory gets twisted into "New Titan RTX teased".

Christ, I'm past my daily clickbait limit and it's only 9am
 
Last edited:
I’m still waiting to see what NVidia does for the new Quadro’s.... really hope they make some announcements there soon.
 

Gotta wonder why regular GDDR6 not GDDR6X?

Otherwise, that looks like what I'd expect a top tier Ampere Quadro card to look like (assuming 48gb is physically/electrically possible anyway). And with that in hand, assuming that Titan A will be the same specs as a Quadro, but not get the slower but more accurate rendering drivers and official CAD support drivers is at least plausible regardless of the general quality of the original source.
 
LOL, if this is true doesn't this make that affordable 3090 not a Titan but a marked up TI. Yup Nvidia would do that, wow the intent in their marketing makes me doubt if they day the sky is blue.
Even if it were all true. How did you come to this conclusion?
RTX3090
10496 Cuda cores
1700MHz Boost Clock
384bit Memory Interface
936GB/s
24GB GDDR6x

Full GA102
10672 Cuda cores
1860Mhz Boost Clock
384 Bit Memory Interface
768.10GB/s Memory Bandwidth
48GB GDDR6
 
Gotta wonder why regular GDDR6 not GDDR6X?

Otherwise, that looks like what I'd expect a top tier Ampere Quadro card to look like (assuming 48gb is physically/electrically possible anyway). And with that in hand, assuming that Titan A will be the same specs as a Quadro, but not get the slower but more accurate rendering drivers and official CAD support drivers is at least plausible regardless of the general quality of the original source.
GDDR6 uses less power and GDDR6x is experiemental. Unlike the gaming market servers have only so much PSU overhead to support GPUs. That number is 250w-300w per GPU supported. So if they made Ampere quadro 350w and all the servers in the world only support 300w then nobody is going to buy them in the enterprise segment.
 
Even if it were all true. How did you come to this conclusion?
RTX3090
10496 Cuda cores
1700MHz Boost Clock
384bit Memory Interface
936GB/s
24GB GDDR6x

Full GA102
10672 Cuda cores
1860Mhz Boost Clock
384 Bit Memory Interface
768.10GB/s Memory Bandwidth
48GB GDDR6
Simple full102 die, so that means more cuda cores, more memory in this case even with a lower memory bus its limited because they need to use double stacked chips that arent available for GDDR6X yet.
 
LOL, if this is true doesn't this make that affordable 3090 not a Titan but a marked up TI. Yup Nvidia would do that, wow the intent in their marketing makes me doubt if they day the sky is blue.

I don't remember a TI version ever getting over twice the memory of the standard version.
 
Good lord! Never in my wildest dreams would I have thought that a graphics card would have 48GB's of memory. Back in the day 1GB was considered extreme.
 
Good lord! Never in my wildest dreams would I have thought that a graphics card would have 48GB's of memory. Back in the day 1GB was considered extreme.

Depends on the day, my first two proper 3D cards were a 12mb voodoo 2 and a 32mb tnt2 ultra .

Before that we had the non 3D cards, plus special ones for higher resolutions or the crazy world of multiple monitors.

My watch is pretty much the same resolution as my first PC :)

48gb IS crazy though. Even In a work context I can’t really use that much. It’d be cool if they’d allow the gpu instancing on Titan though (never gonna happen) I’m not sure what I’d use 4 gpus on 1 card for, but I’d work something out.
 
GDDR6 uses less power and GDDR6x is experiemental. Unlike the gaming market servers have only so much PSU overhead to support GPUs. That number is 250w-300w per GPU supported. So if they made Ampere quadro 350w and all the servers in the world only support 300w then nobody is going to buy them in the enterprise segment.

Very few in the enterprise segment buy video cards only. Its a whole server replacement most of the time. Remember for enterprise hardware is the lesser of expenses. Support, and licensing... mostly licensing make up the bulk of thst upgrade spend.
 
I would love to own this, but THIS cost is the line I'm not willing to cross. I'll pay ~2K but 3? I can buy the better part of a 64-core TR for that.
 
Very few in the enterprise segment buy video cards only. Its a whole server replacement most of the time. Remember for enterprise hardware is the lesser of expenses. Support, and licensing... mostly licensing make up the bulk of thst upgrade spend.
With the current gen its actually been polar opposite.
 
no one is going to buy that as a gaming card at $3000, I don't care how much VRAM it has
 
But, but... Jensen said the 3090 was the Titan replacement, so how can this even exist?! Lol, speak out both sides of the mouth and watch everyone eat it up ;). The 3080 is the 2080 replacement, and 3090 is titan replacement, so benchmark,compare and buy these, then we'll release the actual titan replacement so everything we said is completely invalid.
 
But, but... Jensen said the 3090 was the Titan replacement, so how can this even exist?! Lol, speak out both sides of the mouth and watch everyone eat it up ;). The 3080 is the 2080 replacement, and 3090 is titan replacement, so benchmark,compare and buy these, then we'll release the actual titan replacement so everything we said is completely invalid.

Titan has always historically been the full fat die uncut which is what the 3090 is confirmed as having. They never get bigger as theres nothing left locked away aside from memory tweaks.
 
Titan has always historically been the full fat die uncut which is what the 3090 is confirmed as having. They never get bigger as theres nothing left locked away aside from memory tweaks.


Nope. 3090 has 82 of 84 possible SMs in a 100% GA102 die. A 100% die Titan would have 256 more cores.
 
Titan has always historically been the full fat die uncut which is what the 3090 is confirmed as having. They never get bigger as theres nothing left locked away aside from memory tweaks.
As mentioned, this isn't true ;). So they can increase/tweak both core count and memory interface.
 
Back
Top