Nvidia Pascal Announcement

http://videocardz.com/64303/rumor-nvidia-geforce-gtx-1080-ti-with-10gb-memory

1080 ti rumor with 10 gb of vram.

 

So are we thinking that Nvidia determined that Vega won't arrive for a while, so they are preparing a 1080 Ti based on a decently cut down GP102 so the Titan X remains relevant?

It's relatively clean-cut to market an 8gb card at $600, a 10gb card at $800-1000 and a 12gb card at $1200.

Then shortly after Vega, do a 1100-series pascal refresh in mid-late 2017 with a fully enabled 24gb gp102 titan and a beefier (if not fully enabled) 12gb 1180 Ti and so on. Kinda like the 700 series.

 

Its possible, still kinda against the refresh though doesn't make much sense (doesn't seem like they are going to increase clocks much more then what they have already and keep the power envelopes the same as they are now), possibly refresh certain cards, but I don't think they will go to the 11xx number scheme.

 

That kind of product policy would not convince me as a high(er)-end customer. Early on, you only get severely cut-down parts which will last you ~1 year longer (since launched earlier) and in the end, you will get the more desireable cards which are fully enabled (780 Ti > Titan) only to have them become obsolete even more quickly because of the new generation following it (Titan X <= GTX 1080/1070).

I for one will probably be driven towards more sane choices in the 300-EUR-range.

 

I surely agree that the ~300 EUR range generally has the most sane options, no doubt.

But I'd contend that if Vega isn't around for the next few months, then Nvidia had no incentive to back of their ridiculous pricing.

And I agree that the "refresh" parts like the 700's 780 Ti generally have (relatively) superior value compared to initial titan offerings. However, in the example of the 780 Ti, we can't forget that the Titan Black came out like 3 mo after the 780 Ti to regain the performance crown (at a ridiculous Titan-esque price). That's why I wouldn't be surprised to see a fully enabled GP102-based 12gb "1180" Ti coexist with a fully enabled 24gb Titan-2017 with slightly higher clocks.

But it all depends on Vega. Even a mediocre Vega in late-2016/q1-2017 would probably be enough to make Nvidia be at least slightly less aggressive with their current gouging. Or if Vega is performance-competitive at the high end, then I wouldn't be surprised if Nvidia pulls another 980 Ti trick to release a GP102 at a (relatively) aggressive price to deny cushy margins to Vega.

 

DGX SATURNV (125 * DGX-1s) is Nvidia in-house supercomputer used to design next gen GPU by Deep learning !!!

and a free pick at Xeon Phi...

sources:
https://blogs.nvidia.com/blog/2016/11/14/dgx-saturnv/?ncid=so-fac-dxsv-1176
https://www.nextplatform.com/2016/11/14/nvidias-saturn-v-dgx-1-cluster-stacks/

 

Laptops with Nvidia GTX 1050 & 1050 Ti graphics to launch at CES 2017

http://hexus.net/tech/news/laptop/99829-laptops-nvidia-gtx-1050-1050-ti-graphics-launch-ces-2017/

 

A little cross-posting from the Vega-10-thread since it also applies to GP100:
http://i.imgur.com/79Vyzbr.png
SK Hynix now lists only the faster HBM2 stacks but instead of Q3`16 availability is now set to "TBD". Nothing from Samsung after their PR from january.

 

Also amusingly Micron it seems no longer list their GDDR5X, been checking for nearly a month now since I notice it went AWOL and either they have removed it since then or moved it somewhere else that I cannot see.
https://www.micron.com/products/dram/gddr/gddr5x-part-catalog#/
Cheers

 

Interesting and hopefully only a database glitch, since GDDR5X still servers as "features product" on their GDDR landing page:
https://www.micron.com/products/dram/gddr

 

Yeah only other reason is glitch or hiding status.
There are alternative paths to the GDDR5X data and each one of them shows zilch, but strangely it is only the GDDR5X that is affected, meaning either something interesting has happened and not reported or they have totally changed part numbers for the 3 products they offer and it is messed up.
But for over a month now, that would be some glitch.
Or they have decided to hide the product status due to the expected 11 and 12gbps products going from sampling to production.

Cheers

 

Over a month is too much for a glitch, definitely agreed. Haven't seen an apparent shortage of G5X products lately though (except for Titan X-P, which changed between in and out of stock quite often lately, but that's hardly a high volume product).

 

Mention of 1080 Ti in Nvidia LinkedIn job posting

https://www.linkedin.com/jobs/view/220941257

 

Comparison of NVIDIA Tesla (P100) and NVIDIA GeForce GPUs

https://www.microway.com/knowledge-...of-nvidia-geforce-gpus-and-nvidia-tesla-gpus/

 

Update on this.
Does look like all the info was removed due to upcoming product updates for the other GDDR5X models.
The 11 and 12Gb/s has now gone into manufacturing order status and seems to have happened recently: https://www.micron.com/products/dram/gddr/gddr5x-part-catalog#/

What is interesting is that their DRAM part number document late 2016 also shows the 14Gb/s (-140 with clock frequency of 1.75GHz) GDDR5x product, but that is not even showing sampling status on the url.
So wonder when/if that will be pushed.
Cheers

 

Thanks for sharing this. I happened to notice that the gddr5x catalog was empty like a week-ish ago (and it looks like it had been that way for a while based on this thread). Good to see that it was probably done so micron could update with progression on the speedier parts.

And just to be clear, before the wipe, they listed three gddr5x options: one "contact factory" 10Gbps option and two "sampling" options (11Gbps, 12Gbps), right? Just want to make sure I'm remembering that correctly.

 

Yeah you are remembering right.
Just need to keep an eye to see if the 14Gb/s ends up at some point on the list in sampling status, would still mean 6-8 months wait before product order status.
Cheers

 

I think 14Gb/s is canceled. Micron already talked about GDDDR6 at the end ot the year/early 2018. GDDR6 will then take the 14GB/s speed. They also expect G5X Volume to grow significantly, so i expect Pascal refresh with more G5X in mid of the year. Especially the 1070 is perfect suited for a refresh with G5X.

http://www.anandtech.com/show/11100...ce-roadmap-updates-forecasts-and-ceo-retiring

 

Thanks for sharing. I missed that.

Does someone have a comparison piece for gddr5x vs gddr6?

Micron said last year that gddr5x had the theoretical potential to hit 16 Gbps.



But then at hot chips, Samsung did they were working on 14 Gbps gddr6 that could get as high as 16 Gbps.

https://www.computerbase.de/2016-08/samsung-gddr6-14-16-gbps-2018/

I think someone explained it to me once, but I forgot. :/

 

Yeah I am wondering if it is cancelled but keeping an eye just in case and waiting to see if it is officially cancelled or goes into sampling.
Micron is meant to be accelerating the schedule for GDDR6 as anandtech points out but still would need to go into sampling and I doubt we will see any products with it until some point 2018 if all goes well, and the article points out Micron is expecting GDDR5X to grow significantly.
Cheers