Nvidia Pascal Announcement

I thought Dolly Parton reached that limit years ago...

-FUDie

 

Full ACK. Even though I see the fundamental problem in how integrated circuits are currently designed: Planar.

I see little room for improvement as long as you can place transistors only directly on the substrate, as you are forced to accept longer paths on at least a couple of traces inside each single function unit. Even approaches such as stacking whole dies are just ugly workarounds, not that they would even be suitable to compact anything but entire subsystems.

So what I actually expect for mid term, is a move to TFT like designs, and a change of paradigm, by trading a significantly larger footprint instead for multiple (functional) layers. Or in other terms: Ditching the idea of processing/including the substrate, and instead try to form anything in the deposited layers.

 

"Silicon". Sorry for being a nag, but silicone is a type of synthetic silicon rubber compound.

 

Is that even possible? The wafers are after all cut out of an ingot or whatcamacallits, specifically grown to be a single crystal of silicon; can you even form monocrystalline silicon in the deposited layers?

 

http://videocardz.com/60289/breakin...rs-confirm-pascal-gp102-gp106-gp107-and-gp108

 

If founder's edition GP104 is 600 bucks, you can be sure that a founder's edition GP102 will be a thousand bucks plus easily. This is nvidia we're talking about. They could wring money out of stones if they really wanted to...

 

But it's not.. (700)

 

I don't know if I fully understand your comment TBH but they could also try the stacking dies. companies already have the experience on this area with HBM memory(Yes something completely different but the basics are there) . But then again we will be getting limited by the efficiency since you would need to add hardware for interconnect and logic to do and an additional substrate which would decease the efficiency.

All of this is a really interesting topic. Would be nice to crate a thread to analyze all the ideas and information in detail.

 

The price delta has been 30% and 20%, historically, so a $850-$900 price is a reasonable projection.

The Kepler and Maxwell releases can be used as a basis for timing and price projections for Pascal releases.

Historical GTX 680 (GK104) / Titan (GK110) / GTX 780 (GK110) releases were 3/2012, 2/2013, and 5/2013.
Prices were $499, $999, and $649 respectively. Tesla K20 (GK110) was earlier in 10/2012.

Historical GTX 980 (GM104) / Titan X(GM200) / GTX 980 Ti(GM200) release dates were 9/2014, 3/2015, and 6/2015 respectively.
Pricing was $549, $999, $649 respectively.

So Kepler was 7 months between GK104 and GK110 (14 months mainstream!). Maxwell was 9 months between GM104 and GM200 mainstream.

Beginning speculation here, a similar timing and pricing pattern for Pascal might project a $1500 GP102 Titan P release at the beginning of December in time for some luxury Christmas sales and more importantly as an AMD Vega comparison spoiler if needed.

That timing would also pair well with initially releasing a GP102 FP32-focused Tesla at SuperComputing 2016 in November. (This is where GK110 debuted with the K20 Tesla).

Finally, $850 GP102 GTX 1080 Ti in April 2017 at PAX East.

(Dear wccftech.com, this is all armchair speculation, not Sooper Dooper Secretz)

 

actually you should delete that last sentence and see if they pick it up lol.

 

I will eat my hat if GM102 launches at $850-900. GP104 is way more expensive than GK104 was; the upper midrange has gone way up in price in a few years, I don't see reason why the top end would then come down to meet it.

 

I think I'm not wrong if I say that 1080Ti timing will only depend on Vega availability in Nvidia's style to spoil AMD launch.
Same thing can be said for price...

 

Yes I think you're right. NV hold all the cards at the moment.

 

The extra challenge they will have with their 'case' for higher price for better components is that both the 980ti and Titan already use a blower+vapor chamber, I appreciate NVIDIA will point to other components but this is possibly a large chunk when 1080 was compared to 980 (and still overly inflated).

Will they dare pricing a FE 1080ti above a TitanX RRP...
No competition so wonder if this is going to be a wince moment for all us customers.
Cheers

 

Okay, it's May 27th.

What are the stores selling the GTX 1080?

Overclockers.co.uk has all 1080 models on "Available soon". They're not taking pre-orders.
There are some pre-orders available in ebay. For $1200. With an estimated delivery by the end of June.

 

I think Overclockers mentioned they are allowed to sell them at 2pm UK time.
They are taking pre-orders in a different way where you get a ticket and then somehow apply that ticket at time of purchase.
On the more positive side they do seem to have a broad range of FE by various AIB partners (according to their employee who sources the products for sale).

Cheers

 

Yeah but overclockers appear to be adding a £60-80 premium over the already over priced FE!

Scan appear to have a few models in stock, but no prices, maybe after 2pm?

 

This is kinda strange... beginning to smell like a paper launch, but the day has a couple more hours.

 

Overclockers has pallets of cards, some places may be low but I think the larger stockists will have a moderate bunch.
Still it looks like there are plenty interested in the FE who do not want to wait for custom designs.
If pre-order interest is so high, I can see some retailers will take advantage.
I think they also saw a massive interest in the 980ti that had them sell out in the beginning.
Cheers

 

The Asus was a higher price than the other FE cards according to overclockers, you thinking of that or all of them in general being £60-80 higher than the official rrp?

As there are so many posts there any idea where you saw this?
Thanks