Nvidia Pascal Speculation Thread

Hah
Funny you say that I only thought of checking out the mobile part after I posted
And yeah with you on that possibility.
Thanks

 

The problem I see with this card is not only that it would be made obsolete if high power/large die 14nm was available, but why would developers/pro use a card limited to 4GB (combined memory only happens in special development under DX12).
Only reason for the card IMO is that the higher end 14nm (maybe includes 390x equivalent we do not know but the fact is only the very low power/efficient model has been touted to date) will not be available until Q3 earliest.
Fingers crossed the rumours are all wrong but it is not looking promising IMO for enthusiasts who are looking for a good performing (lets say increased performance 390x replacements/980 replacements) 14nm or 16nm by the summer.

 

It's been synchronized with VR-headset release since beginning, and they got all (the relevant ones anyway) delayed > dual Fiji got delayed too. AMD actually said it themselves that it's lined up to be released with VR sets

 

So the product was completely ready for launch for months (it was announced almost 10 months ago) yet they let it sit there, depreciating over time as a whole new silicon generation is about to hit the market.

I've read their explanations, but they don't make a lot of sense IMO.

 

Been ready since Q4 apparently. Anyway, as someone said (I'm thinking AnandTech/ @Ryan Smith but could remember wrong), it would get slaughtered on todays market if it was released as traditional dual-GPU product, CF(/SLI) support on new titles is hit'n'miss, often plagued with issues even if the performance scales ok. VR on the other hand is completely different, it gets nothing but benefits from having 2 GPUs

 

That's probably right. But that's no different than 2 discrete Fijis. But I'm sure there were a sufficient amount of potential buyers for dual Fiji, the same market as this for 2 discrete Fijis. If you're going to sell a dual board anyway, why not go for those as well?

I think so as well.

 

Seems to me even AMD do not know what to do with this GPU, as a pro developer would you use 4GB HBM memory or one closer to 12GB GDDR5 when it comes to visual related designing?
This is further compounded on the VR front with Roy Taylor saying:

This card is only being released IMO because we are not going to see a serious enthusiast/pro developer GPU until very late this year on the 14nm generation (or maybe late 2017/early 2018 according to latest presentations for pro 14nm products); bear in mind also AMD's schedule now for HBM2.
Waiting to see how depressing the news is going to be for enthusiasts when NVIDIA do their presentations soon as well, although as they have probably gone with Samsung for HBM2 that aspect may be a surprise but I am still concerned for their large die/high power 16nm GPUs as well.

Cheers

 

What Taylor said is just about bringing the cost of VR down, nothing more or less.
4GB memory is a bit on the low side, yes, but it's not necessarily too low.
Late 2016/Early 2017 is for Vega with HBM2, so high memory, finfet etc.

 

sorry, wrong thread.

 

In a general sense 4GB might be low, but I wouldn't say that it is low for VR though. Consider that the Rift has a rendering resolution of 2700x1600 ¹ and the Vive 3024x1680 ¹. This means 1350x1600 (2.16MP) and 1512x1680 (2.54MP) per eye respectively. This is pretty much 1080p (2.07MP), so unless 4GB has been obsoleted for 1080p (and it appears to be holding up okay even at 4K ²) I'd say that as far as memory is concerned I wouldn't worry too much. Rendering at min 90fps is another matter, although if a Fury per eye can't do it, I don't see how a single 290x could.

¹ According to this guy at least https://www.reddit.com/r/oculus/comments/41vuf8/rendering_resolution_in_the_rift/cz5thxq
² http://techreport.com/blog/28800/how-much-video-memory-is-enough

 

SMW,
Although it means it is running two displays simultaneously both with same level of fps action; no idea how this impacts memory/performance.
Cheers

 

Unfortunately Vega has been pushed back beyond 2016 now according to the information yesterday (which I was hinting at).
So ironically Taylor is pushing 14nm Polaris for consumers and the professional world will have a dual older gen card at 4GB (albeit with HBM) when it comes to VR.
However our discussion strengthens my earlier and main point; this suggests there will be no 14nm (maybe same for NVIDIA as well) GPUs that are upgraded equivalents to 390x or 980 any time soon as their target seems below this from their marketing.
Cheers

 

Sorry just to add (needed to check the information), AMD positioned the presentation of the 290x/970 VR discussion even below that of a 390.....

So raises question on expectation on when we can see actual upgraded enthusiast level of performance on these next generation GPUs; my concern is that the older cards will still comfortably compete on performance (only benefit for NVIDIA fans is that maybe their lower bit bandwidth will be raised but then this would probably only be applicable to the 980 equivalent and higher enthusiast cards) so the gains will only be size/efficiency and not performance until I feel at least Q3 and maybe closer to end of year.
Cheers

 

No it hasn't.
I wasn't referring to any rumours or any such, but the actual roadmap AMD released - Vega is scheduled for late 2016/early 2017, it's impossible to say exactly obviously, but it's right on where the year changes, while Polaris is midway between 2016 and 2017

 

Well all news from yesterday reported AMD saying Vega (with HBM2) is now 2017 (no suggestion it will be even Jan/Feb) and after Polaris which could take us easily up to end of year depending upon how the large/power intensive GPUs progress.

This is not a rumour.
Cheers

 

News can say whatever, I should know, I write them
(also AMD didn't actually reveal anything about the schedule other than the roadmap)

On the roadmap, if we assume new year starts from the middle of "2017", Vega is set for late 2016 release. It's block starts from quite a bit before the "2", and ends just right of "1", of "2017"

 

Why all the off-topic AMD this, AMD that?

Please post about Pascal.

 

Sorry, offtopic train is hard to stop
There's nothing new to post about Pascal, there's GDC event going on just about now but nothing from it yet.

 

To me there is a trend for both AMD and NVIDIA and that is it seems the only solid news from both is that the releases we may see before summer will either be mobile or low power/efficiency cards, cards equivalent (even for NVIDIA) to the 390 sadly seems to be pushed back IMO to Q3/Q4.
Raises the question when will we see the performance replacements (not efficiency so actually better fps-latency) for either the 390x and 980, the presentations from NVIDIA this week may help to answer all/some of these speculations.
The rumour news regarding the 1080 seems a bit questionable as mentioned earlier in the thread.

Cheers

 

No this thread is NOT about AMD it is about Nvidia's upcoming Pascal. Unless your posts have something about Nvidia's Pascal in them then they ARE off-topic.

If you want to post about AMD try one of the many AMD threads.