nVidia 2015 mobile speculation thread

Read the first response in this thread.

Edit: Nvm NBR censored me

 

... EVERY customer when he looks at the price tag #1999.98$

EDIT: Oh, I see, so this picture is as much fake as 970's BUS being 256bit, I get it)

 

Likely these two Baidu threads

 

Ah, that makes more sense then.

 

cool I guess, won't be picking it up tho

 

Whoa that's crazy... How can mere mortals follow this insane tech evolution speed? '-'

 

WOAH! Have you seen this?! Nvidia decided to create another 6xxM GPU based on Big Maxwell! See yourself!


Also Half-Life 3 finally confirmed

 

Nvidia bought AMD CONFIRMED!

 

Come on, at least call it GCN 3.0 or something.

 

lol wtf!

 

It was a test guys. Relax. I already knew it was likely a fake but wanted to see if other people could spot the irregularities.
There are many things thats wrong with the screenshot:

1. Nvidia ID is 10DE, not 10DS
2. Clevo Subvendor ID is 1558 not 2858
3. Release date of Titan X which this screenshot is gathered from is March 17th, not 13th
4. Texture fillrate is wrong based on the specs
4. Pixel fillrate is wrong based on the specs
6. GPU-z doesnt support GTX 980MX yet, so there are many fields that should have been blanked out.


@Ethrem @Mr. Fox and the rest of the mods: As for the moderating on this forum, unless something change, I`m not returning to this forum. octiceps need to get a 1 week ban because of constant childish behavour and calling out members like "heads are in the clouds" etc. He have done it numerous times now and does nothing but post troll posts and flame bait whenever he feels like attacking members. At many other techforums you get infractions and ban if you call out members.
Yes, I know I am calling out octiceps now, but it needs to be raised awareness publicly how sloppy the moderation on this forum have become.

Forums like this should be discussing tech, not acting childish and attacing other members. Joking is fine by me, but not if its directed at someone personally. I dont give a damn about octiceps and his opinions, but behaviour like that should not be allowed.

 

Try reading the first post again.
There are two sources that have covered the Hasee presentation.
The first link explains that GTX 990M is coming.
The second screenshot is from a different source and is also from the presentation where it was said 990M = 980M SLI. Look at the timestamp.

 

Meh, I was too lazy to be creative. It was 2am with a little booze in me.

 

You forgot to read my other post to n=1.

You also didn't link the other source, so did affect my perception of this 990m == dual 980m.
However, the time stamps are throwing me off a lot. The GTX 990m was mentioned at 16:03:47, but this "phone source" cites it at 10:48. Granted, the 5-8 must describe the date of the conference, yes?

 

Well my point was more about how they never fail to increase the series number every year.

900M - 2014
800M - 2014
700M - 2013
600M - 2012
500M - 2011
400M - 2010
300M - 2010
200M - 2009
100M - 2009
Etc...

So it makes sense for 1000M or some new naming scheme entirely for 2015. It'd be bucking the trend to keep the 900M series nomenclature throughout 2015 and into 2016.

That's all I meant. It's not very important I know. What matters is what it actually is, not what they decided to brand it

 

The GTX 1000-series is for the Pascal architecture which should arrive some time in 2016.

 

Says whom? It will probably be late Q3 2016 by the time Pascal is available and they're certain to do a refresh (maybe two) between now and then with Maxwell, and will bump to 1000M series or name it something else entirely. Who knows.

 

It's not Q3. More likely at Computex 2016 or some other major event before. It will be 1H 2016.

NVIDIA will probably launch another Maxwell chip for mobile by the end of the summer or early fall.

 

Maybe. Pascal is going to be a big deal for scientific computing. This space has been dominated by AMD's GPUs. Pascal is NVIDIA's opportunity not only to take the lead but to do it big enough to convince existing AMD shops to switch. NVIDIA shuffled their roadmap to accommodate Pascal for this reason. Delaying Pascal at this point would be a mistake and I don't see NVIDIA making it.

 

Dawg you need to chill out

 

This is false. Nvidia has always dominated the professional GPU space.

 

I was going to delete this and leave it at that but instead I'm going to respond publicly since you felt the need to call out the moderator team publicly.

If there is a post that you do not agree with, you report it and we decide as a team what to do with it. I am very active on the forum and I do remove posts that I deem to be completely out of line and I issue warnings and infractions as I deem appropriate for each situation but myself and the rest of the moderator team don't want to step in for every little disruption in the forum. We are all adults here and while we won't always agree with each other, we should know how to get along.

With that said, unless a post gets reported to us, you can't expect us to know that it is offensive to anyone and if we went through and dealt with every single post from the perspective that someone might get upset by the content, we would be deleting posts and threads all day.

We have the PM system available as a tool to reach any of us moderators or the super moderators should you have any concerns about the conduct of a member. These concerns should never be made public, they are private matters.

 

I dislike purely speculative discussion threads. There has to be some fact behind it. It leads to arguments/disagreements among peers, which tends to "get out of hand." Not that I don't like speaking to what could happen. At least a moderator is actively moderating, as they should.

 

Yep AMD even admits to it themselves.

Hopefully by "invest more in software" that includes consumer-oriented drivers and not just professional ones. But definitely refreshing to hear.

 

I believe these threads are problematic too which is why I closed a few of them when they got out of hand before but I am trying to deal with issues on a case by case basis now rather than shutting down an entire discussion because a few people need to modify their behavior. There's no doubt that nVidia will release some refresh with Skylake coming so I believe this thread has a purpose but I'm going to change the topic name to make it a more general nVidia mobile rumors thread.

 

The next 900M flagship will have shaders in the range of the desktop 970, or maybe one more shader cluster than it. They won't blow the full load yet, when they may need to stretch 2 more flagship mobile cards out of GM204, with Pascal so far off.

 

Professional does not mean scientific although you might not see the distinction if you're not in research. Scientists use what their colleagues use so that they can share code and data. Right now that's predominantly ATI/AMD because of better 64-bit floating point processing and multi-GPU scaling.

 

I think we are looking at GM204 full die next and a rebrand with an overclock ala 880M (but hopefully better executed) and then Pascal. They need to make it attractive to upgrade to Pascal, I can't see them shooting themselves in the foot with a large jump with a Maxwell refresh.

 

I lump workstation and server/HPC under the professional graphics umbrella because of the differentiation between Nvidia and AMD's respective lines of professional GPU products (Quadro/Tesla and FirePro W/S) and their gamer/consumer lines (GeForce and Radeon).

On both pro and consumer sides, AMD has great hardware but software and drivers are lacking. AMD has even less market share in pro graphics than it does in consumer graphics because Nvidia jumped out to a very early lead with CUDA. Lisa Su admitted as much during FAD '15 and said pro graphics will be one of AMD's areas of focus moving forward since that is a high margin market they are currently missing out on big-time.

 

HPC is definitely nvidia. AMD is for poor man cad, graphic, ray tracing workstation. It is hard to make a margin off a poor man though.

Sent from my 306SH

 

Still, scientists currently are using more FirePro than Quadro. I don't want to sound like I'm playing the appeal to authority card, but one of the things I've done in recent years is spec out GPGPU rigs for dark matter researchers. I made all of these arguments in favor of Quadro: better tools, better drivers, etc. They still went with FirePro because that's the code that their colleagues were using and its easier for them to adapt colleagues' existing code than it is to start from scratch even with NVIDIA's better tools. As their work spread to other colleagues, so has the investment in AMD's FirePro.

 

Please don't use the term scientists to refer to one group. There are many different types of scientists out there. It depends heavily on which field you're in.

Astronomy/physics simulation is a every special field where FP64 is minimal requirement. No amount of hacking on truncated precision or space tree encoding or whatever can save you from low FP64 throughput. FP64 on GK200 is not bad (better than Hawaii on paper actually), but many problems in this field are N-N which out right suck on anything Kepler/Maxwell. In this case, Radeon/FirePro is the better hardware, and your clients are making their decisions accordingly.

In my field which is geophysics/geodynamics, it's kinda split in halves. We have Teslas for some problems and FirePros for others. None is going to replace the other on a large scale any time soon.

This is a consumer forum and most workstation users here are DCC specialists. In that field Nvidia is dominating in everything software and at least in some cases hardware as well. So don't expect to see many FirePro users/supporters here.

There's also the public cluster club where most decision makers would prefer Nvidia due to fear of losing CUDA customers.

 

You are correct. It was arrogant of me to do that. I work for theoretical and particle physicists which I guess tends to blind me to the other sciences.

 

With the Oculus requirements released, I wonder how much longer it'll be before mobile machines can run it fluently.

 

I dunno about you, but I clear the spec XD

 

Right now the Clevo P77xZM/P75xZM is the best laptop for Oculus Rift. Fastest mobile CPU, 980M meets the GPU spec, and lowest latency due to no Optimus or SLI.

 

980M is not as fast as a desktop 970

780M SLI is not as fast as a desktop 970

Neither meets the spec and SLI is likely to cause more issues so my system won't meet the spec either.

 

Close enough when you OC the 980M

 

Overclocking does magical things... But I'm talking about stock vs stock.

 

Then wait for the mobile refresh. I don't want to split hairs.

 

Actually, stock v stock, 850/5000 780M SLI should beat a 970, which is close in power to a 780, because I school my friend's (overclocked) R9 290X, which in turn is more powerful than a 970/780. If you said a 780Ti/980 spec, then yeah, I'd step backwards. I'd need 100% scaling here to match stock for those cards and that's only sometimes.

 

The only issue with that is, SLI is useless with VR right now. And I don't know that Kepler cards in SLI will ever be viable if Kepler doesn't get the VR Direct improvements that Maxwell 2 has. So a single 980M w/o Optimus is as good as it gets currently.

 

Considering that nVidia still hasn't even given us DSR...... Yeaaaa

 

Why is SLI useless with VR? In stereoscopic 3D one card can render one "eye" and the other can do the other "eye", so why that cannot work for VR is beyond me.

DSR can't work with optimus though, and adding "DSR" as a "feature" for mobile will just have everybody making noise as to why they can't have it (in their 1" thick notebooks).

I doubt we'll ever get that feature unless nVidia discards Optimus on the whole, or delegates it to low-end systems which are not really meant to run heavy 3D apps.

 

VR SLI (one GPU per eye) is part of VR Direct. Which, like I said, I don't know if Kepler will ever get since it was announced with Maxwell 2.

 

Well, Kepler can for a fact do this with Stereoscopic 3D, so Kepler by default MUST be capable of doing it with VR, and if nVidia does not enable it, they are specifically gimping products. There is absolutely no reason why one technology that is already available and working for one card in one 3D solution cannot apply to another 3D solution.

 

Found this: http://www.reedbeta.com/talks/VR_Direct_GDC_2015.pdf

Page 48:

However, on page 46:

 

Well, there we go. That I can understand. But at least we know it'll work.

 

But without perfect scaling, your 780M SLI doesn't meet Oculus' recommended 970/R9 290 level of performance.

BTW about your earlier point, stock 780M SLI does not beat 970. 970 is as fast as 780 Ti and 290X, and 980 is another 20-25% ahead.

 

970 has gotten faster then. But yeah, I suppose. Either way, with an OC it should work. I can handle 1006/6000 without issue. Should be fine.

Also, games that don't sit at 90%+ scaling in SLI but rather 60%-70% (with acceptable framerates) should work fine.