knightsilver
Guess the new non_Ti RTX 3080 Super $1,000.00 , y’all know I’m right!

hard to say what nvidia has for logic when the ampere surface 
 
most likely the are having the same issues amd is having  which is to consider smaller devices 
 

NVIDIA GeForce RTX 3070/3080/3080 Ti: More VRAM, Way Faster, & Cheaper!
 
NVIDIA's next-gen Ampere: huge ray tracing performance upgrades, more VRAM, less power, and also on 7nm. The news is coming from the purported fact that NVIDIA has been discussing Ampere and the next-gen GeForce RTX 3000 series cards with AIB partners; there are a very few, but juicy tidbits here, that if true, would be great to see.
 
 
from Linustechtips
 
 
 
--------------------------------------
 
if Titan Ampere RTX - has 4 stacks of 12GB chips of hbm3 you could be looking at 48GB on Titan A
usually we get about 1/2 that for gaming - minus 1 GB - so you don't NVlink two 2080ti into a cheaper titan (don't want to cannibalize your high end market)
that could mean 16gb 3070,3080 and 22 GB 3080ti
 
 
they could also leave chip size relatively the same the die shrink could allow for two turings in the same space with improve performance clockwise and power reductions too
they could really step on AMD right now that way
 
they need it to stay ahead of both AMD and Intel
it's intel Im worried about...they have enough resources to catch and and surpass in a few years if Nvidia doesn't step on it

I'm wonder if Nvidia will go back to releasing the Non-Ti units & then 3 - 6 months later the Ti version
 
Seeing as they have no real competition on the high end GPU .... Yet

as for architecture  wasn't nvidia going to go chiplet design soon- when?? who knows... now or next gen???
the shrink from 14nm to 7nm offers under 4x  the transistor density - because it's a full shrink - 1/2 node is 2x
 
so you could put 2 turings on a chip and still shrink about 50% or offer a few more tensors and RT cores and still shrink 25%
the shrink also offers up better clocks for less power
 
wasn't mGPU compute going in big time to windows 10 with intel stepping in the discrete GPU market
 
the future is chiplet and offering more Vram to the card as well as SSD(m2 drives) directly to the cards soon  offering like 4 or 8TB of M2 directly to card 
cut down on latency and unzipping and compressing graphic files - keep the data direct nearby - so you avoid PCIe traffic,CPU traffic and slow mass storage on slower drives
 
which is why Arm was and maybe considered for future GPU's to cut down on use of Main CPU 
 
Didn't I post a diagram of such a card a few years earlier I gleaned from an Nvidia presentation
 

 
the one i posted showed 2 arm chips with four m2 drives on the board as well
 
as Jen Hsun said in the last GTC - latency is the problem now
and why buy mellenox - to have your own infinity fabric and cut down on latency issues
 ------------------------------------------------
Since the beginning of this month, Nvidia has been giving the upcoming Ampere GPU's, they are looking for the next reference boards and the next one AIB cards for the arch.
Ampere should be just about everybody, some of the big points that we've communicated to us:
-Significant raster performance increases (higher than Turing gains)
-Massive RT performance increases
-Expanded Vram capacity on the full range of SKU's.
-Dialed down TDP's on many cards in the range (high end remains unchanged at 250w)
-Increased core clocks (healthy 100-200mhz bump over Turing)
That ends the good news.
Bad / neutral news:
-Prices, notes on the most popular cards, slight decreases on the unpopular ones. (x80ti and x80's will see a price drop, while x70's and below will remain UNCHANGED)
-Overclocking, core clocks stock will be higher than ever, 100-200mhz higher depending on SKU, however this does not mean the 2100-2200mhz ambient cooling upper limit of frequency will be broken when overclocking, which brings me to the next point.
-Voltage limits decreased. We went from 1.2v of Maxwell and previous generation to 1.093v to Pascal (maintained in Turing), which now seeks to decrease to sub 1v in ampere, which will severely limit overclocking frequency, but has a positive impact on power consumption.
Many points are highly sensitive to market forces (Prices, TDPs and vram capacity especially).
Do not ask specifics about the information, thank you.
 
from 3dcenter.org
 

 

Xavier Zepherious
NVIDIA GeForce RTX 2080 Ti SUPER Rumored To Launch In Early 2020 As Next-Gen Ampere GPUs Delayed
 
It looks like the rumor mill is now suggesting that NVIDIA is indeed working on its flagship 2020 product, the GeForce RTX 2080 Ti SUPER. The rumor comes from Twitter user Kopite7kimi who previously reported accurate specifications for NVIDIA's GeForce RTX SUPER and GeForce GTX 16 series lineup prior to their official introduction. Still, since this story has a major twist to it, we would advise readers to take it with a grain of salt.
 
 Recent reports have pointed that NVIDIA has plans to go with Samsung and use their advanced 5LPE EUV process for their Ampere GPUs but that remains to be seen.

 

GA101 cancelled— kopite7kimi (@kopite7kimi) November 7, 2019

 
The same source also alleges that one of NVIDIA's Ampere based GPU known as GA101 has been canceled. We don't know the details of this GPU but the source has previously stated that it was a smaller variant of the top GA100 GPU core. It was said to be half of the GA100 and would allegedly feature a 3072-bit bus interface and a much smaller die compared to the GA100 which is rumored to have 55 Billion transistors, 2304 Gbps bandwidth speed from 6 stacks of HBM and a die size of around 800mm2.
 
 
https://wccftech.com/nvidia-geforce-rtx-2080-ti-super-q1-2020-launch-rumor/
 
 
-----------------------------
big honking chip even at 7nm
 
that also 48GB of HBM3 guys or 96GB if it's dual die- see the underline portion - im also gonna assume the HPC GA100 is dual variant- with 2 chips and 96GB of hbm3
so half would be 1 chip and 48GB of hbm3
 
 
 
Im also assuming Nvidia knows more about AMD radeons that are about to come out...and that they can delay ampere and put out another super card- they know something we don't?

Id like to change my estimate from mid to late 2020 to NVIDIA launching in only late 2020: "You’ll have to put your credit card on ice for the time being though because we won’t be seeing actual hardware with DisplayPort 2.0 inside it until late in 2020." 
https://gizmodo.com/displayport-2-0-is-finally-here-and-this-is-why-it-mat-1836453442

Do know that smaller than 8nm means lower boost frequencies in the silicon world.
Seems like 8 to 12nm is about the sweet spot for performance, unless they're able to cool down the GPU further, by perhaps cooling the board around the GPU better, or cooling the back of the GPU (or perhaps even find a way to cool the sides of the GPU block??.
 
While I don't really need all the RAM, I do hope we can see 2x 2080Ti-like GPUs on one card!
While it may seem overkill for just about anything, it should be necessary for 8k gaming, and very beneficial for distributed computing!

castrator86

ProDigit
While I don't really need all the RAM, I do hope we can see 2x 2080Ti-like GPUs on one card!
 

 
I miss the days of double-GPU-die cards. I'd love it if they brought them back.

They were cool and all but they were still gimped while having additional thermal issues being on a single PCB.  You were better off buying two separate GPUs in SLI than a single dual GPU.
 
I don't mind a powerful 8K ready GPU. 

castrator86
I know but SLI seems to have its  own set of issues.
 
I'd love a card that uses both sides of the PCB instead of mounting everything to just one. Fans & heatpipes on opposing ends to help with heat dissipation. Even if you only put 2 of the "lesser" cores on it, two 2070s on a single card should wallop a 2080Ti in performance.

Ya know that's a darn intriguing concept.

Imagine a hole in the board and the chips are double sided during manufacturing making it one large gpu with built in vias from one side to the other as opposed to two separate gpus in sli.

Board to chip interface happens around the edge of the chip where it rests on the board. Heatsinks on both sides clamping the GPU between the two with a second x16 with no pins to provide support to the massive card.

Dang I should patent this and send the board design to Nvidia for licensing.

Now you're cooking with gas!
 
I'd love the idea of a 4 or even 5 slot GPU with fan shrouds on top & bottom, one PCI-e connection in the center, and then opposing laid out GPU die/memory on each side of the board.
 
No idea what you'd be looking at for power requirements though!!

You guys are giving GPU sag a new meaning. 

boylerya
 
I owned the 9800GX2 and it didnt take long before problems occurred to where either one of the GPUs broke or something in the card broke to where only one GPU would function properly. -- until EVGA finally gave me a GTX275 instead.

 
Man I still have a working GeForce GTX 295 CO-OP Edition  , that runs about every day in an auxiliary system for media & such .... 
 I always thought the dual chip cards were such a great idea....  

ProDigit
A heat plate on the back has been done by some manufacturers, to efficiently cool the base of the GPU die.
But a dual GPU system (I'm only in for raw GPU performance), would mean each GPU block would be located under a fan in a dual fan system.
And if we're talking about 2070 or 2080(Ti) GPUs, chances are we'd need a quad fan system, as even current 3 fan system is really only good enough for up to 200 Watt, 300Watt for a 2,5 slot design.
3 slot designs don't really work efficiently.
 

My idea would be dual fan design on each side of the card for 4 in total. I'm talkin this PCB/card itself would be 2 slots thick and taper into a single PCI-e connector; then dual slot heatsink/fans on both sides of the cards to keep it cool. You're probably looking at a 400+ Watt card though.

Bruno747

Ya know that's a darn intriguing concept.

Imagine a hole in the board and the chips are double sided during manufacturing making it one large gpu with built in vias from one side to the other as opposed to two separate gpus in sli.

Board to chip interface happens around the edge of the chip where it rests on the board. Heatsinks on both sides clamping the GPU between the two with a second x16 with no pins to provide support to the massive card.

Dang I should patent this and send the board design to Nvidia for licensing.

 
I think the concept of cooling is gonna require tunneling lasers to put thermal cooling pipes for water cooling or water vapor cooling inside the silicon package itself
 
this was not gonna be acheived til 5nm or smaller i believe
 
https://www.researchgate.net/publication/228931688_Thermal_Management_Solutions_Using_Electron_Tunneling_Through_a_Nano-Scale_Vacuum_Gap
 
https://www.technologyreview.com/s/530911/nanotechnologists-discover-how-to-carve-tunnels-beneath-the-surface-of-silicon-chips/
 
 
 

• New lithography. Extreme ultraviolet (EUV) lithography is required for 7nm/5nm. Beyond 3nm, though, there may be a need for a next-generation EUV technology called high-numerical aperture (NA) EUV.
• Selective processes. Chipmakers also a need a broader array of selective deposition and etch technologies, enabling vendors to deposit and remove materials in precise locations.
• New interconnects schemes. The wiring schemes in chips are too congested, requiring new materials in the arena.

Then, at each node, the defects are becoming smaller and harder to find. “Lateral scaling, namely denser transistor layouts, drives the need to detect smaller defects and increases the need for design-aware inspection and review. Vertical scaling drives the need for detecting and verifying buried defects,” said Mark Shirey, vice president of marketing and applications at KLA-Tencor.
 
 

castrator86

ProDigit
A heat plate on the back has been done by some manufacturers, to efficiently cool the base of the GPU die.
But a dual GPU system (I'm only in for raw GPU performance), would mean each GPU block would be located under a fan in a dual fan system.
And if we're talking about 2070 or 2080(Ti) GPUs, chances are we'd need a quad fan system, as even current 3 fan system is really only good enough for up to 200 Watt, 300Watt for a 2,5 slot design.
3 slot designs don't really work efficiently.
 

My idea would be dual fan design on each side of the card for 4 in total. I'm talkin this PCB/card itself would be 2 slots thick and taper into a single PCI-e connector; then dual slot heatsink/fans on both sides of the cards to keep it cool. You're probably looking at a 400+ Watt card though.

 
 
there is a process of making the whole PCB a vapour chamber and cooling from underneath ... with a liquid cooling solution on top

Xavier Zepherious
 
NVIDIA CEO: Our Leadership in Ray Tracing Turned It into a Standard, Almost Every Platform Will Have to Have It
 
 https://wccftech.com/nvidia-leadership-raytracing-has-turned-it-into-standard/

https://wccftech.com/nvidia-introducing-ampere-in-march-2020-releasing-rtx-3080-in-june-analyst-indicates/