Pascal: What do we know? Discussion, Latest News & Updates: 1000M Series GPU's

Yes it has been a while , nice see you too!

 

So... I've been missing out on news, it seems. 1080 is out and you guys think it's rather disappointing...

We shall see what GP106 brings to the table.

 

GP106 won't be any better than GP104, that much should be obvious. If they just basically did a die shrink of Maxwell 2 to make GP104 then you can be sure that GP106 is going to be a disappointment as well.

 

Of course. What I meant was with respect to GM107 or GM206 - for example, @Prema can get his GTX 860M to (nearly) exceed 5000 on Fire Strike 1.1.

If GP106 is a disappointment, then the GTX 1060M wouldn't reach 9000 on FS, and in that case I might as well get a notebook with a GTX 980M and OC that.

 

AMD uses Dynamic VRAM, which is basically sending the data that might not be needed for the moment to the system memory to use as cache if later needed. They optimize drivers for newer games that need Dynamic VRAM and so far from HardOCP tests they haven't really any performance issues.

 

As posted earlier in this thread, Internet is reacting faster too !!
http://videocardz.com/59718/nvidia-gtx-1080-gtx-1070-founders-edition-explained

Get rekt green troll, stupid marketing sauce.

 

GP106 should at least be as fast as 970m when not bandwidth limited.

 

That Chinese chart has the 1060M at 75% faster than the 960M. Fast enough, if true?

An OC'd 980M will still be faster though, because the 1060M isn't designed to be faster. The x60M part will always have a 128-bit bus.

 

NDA on Reviews lifts on 17th
"NVIDIA and Alienware were busy clearing stocks of GK104 GPUs by giving away 1000 GeForce GTX 770 cards during Dreamhack (to celebrate announcement of GeForce 1000series)."...770!?
http://videocardz.com/59695/nvidia-geforce-gtx-1080-reviews-on-may-17th
http://wccftech.com/nvidia-gtx-1080-cards-shipping-reviewers/


Sent from my Micromax AQ4501 using Tapatalk

 

By the time DX12 takes off, we will be using volta/navi.

Sad thing is, I am still tempted to pick up 2 x 1080 aftermarket for 650 USD each since there is absolutely no competition at high end coming from AMD until 2017.

 

http://www.tomshardware.com/news/id-software-doom-vulkan-gtx-1080-nvidia,31756.html

Sent from my Micromax AQ4501 using Tapatalk

 

They claimed it was faster than 980 SLI. It is only 61% faster, which means that proper-scaled 980 SLI with its 95% bonus over 980 SLI is much faster. And I don't care if overclocked helps, because you can OC 980 SLI too. To great speeds. They are correct in calling out nVidia's lies, because they're lies. The only time the GTX 1080 is faster than 980 SLI is in VR, where SLI scaling is limited to about 50% over single GPU... but that's not what they said. Perf per watt is also only 2x Maxwell in VR... but that's not what they said.

If the card can't hold its 1700MHz boost, then it sucks. And as for that 2000MHz overclock... that's not even very far if its base boost clocks are over 1700. 2100MHz is less (percentage-wise) overclockable than Maxwell compared to base. +300Mhz was almost guaranteed with Maxwell which had anywhere between 1000 and 1180 base clocks at reference, and many people managed 50% OCs on 980Tis. Without much downside either. Pascal probably is going to be much hotter and will likely draw much more power. "67c on air with overclock" in an AC room, where people can't see the PC, likely with fans on max, maybe with an AC vent blowing on it? >___> I don't believe anything.

Why? It uses Crossfire. Windowed mode = act like Fury X. No Crossfire Profile = act like Fury X. It's a worthless card unless somebody wants to run Quad CrossfireX on a mainstream intel board, which is even MORE stupid in itself than the card's own design.

Yeah, engineering-wise it's disappointing as hell. While the facts are all there, people either don't know about them or aren't reading them. All the big websites aren't reporting on their lies.

GP106 is going to be like GP104 in its relation to GM206. It's going to be more expensive and under 2x as fast.

960M is GM107. If 1060M is GP106, then it'd be a different class of card. You should compare it to the 965M which is GM206.

If it's only 75% faster than a 960M, I'd say they're skimping badly. 965M in itself is much faster than a 960M; it has 60% more cores. 970M should be nearly double the speed as well, with double the cores and 50% more memory bus width. If a 1060M is only 75% faster than a 960M, then by my count it's worse than a 970M, which is garbage for a jump that's supposedly so big.

Only way they'll get out of that is if it's GP107, which falls in line with what we know about the 1080 vs 980.

 

You forget one important factor. When Nvidia start to optimize the drivers for Pascal and start to destroy drivers for Maxwell you will see that Gtx1080 will be faster than Gtx980 Sli!! This is plain fact.

 

As much as I hate and bash nVidrosoft for their crap, I'm fair. Since I've gotten this machine, I've only seen performance go up, or remain the same, with all my games. I've never seen a game run worse after a driver update; at least not significantly (maybe 1-2fps here and there if anything, and if so they usually go back up with later drivers).

I expect Pascal to improve somewhat with drivers like Maxwell did, but I don't expect GP104 to go from 61% to 100% over GM204 because of drivers alone. 970s went from ~780 level to ~Titan level. 980s went from under 780Ti/Titan Black level to ~780Ti/Titan Black level. A few % increase. They were never even as weak as regular Titans.

I DO know that Fermi has performed worse in new games using new drivers; Dark Souls 3 and the Doom Beta are the specific ones I recall. They also claimed to have fixed it with a recent driver after 364.72, meaning it was likely unintentional. I also noticed that Linus did game testing where games basically didn't get much better or worse over three years of drivers being tested, and IIRC it was Fermi he used. Yes, I do remember correctly: here is the video in question, where he used a 480 and tested all sorts of drivers, with a 5930K used to remove CPU bottleneck. After Fermi's performance improvements in general with earlier drivers, most games didn't have many improvements to write home about. The only game where it made a real difference seems to be Far Cry 4, where the 337.88 driver (which IIRC was around the time they optimized their DX11 drivers for CPU overhead) seems to spike ahead of the old drivers in any significant way. Witcher 3 I believe uses more cores in general so it mattered less.

I understand that many of the users on this forum including yourself tend to test via benchmarks like firestrike, and any performance drop is usually taken to mean "discard drivers", but in terms of gaming, I have seen little such evidence of intentional nerfing of older stuff. What I DO find stupid is the fact that they somehow managed to, somewhere inbetween likely 355.60 and 364.72 (or maybe further back, and it was tied to a form of rendering that DS3 and Doom use and no other bigger games use so it was "missed" by the public) they "accidentally" broke Fermi's performance for those games. But they fixed it fast enough (even though r364 is cancer, the next driver branch, if less cancer like 362.00 is, is probably going to contain said fix) for me to assume it was a mistake.

So no, I don't expect maxwell gimping or kepler gimping to occur... at least for gaming. What I *DO* expect is that new tech, or optimizations for varying games, is not going to happen for the older cards any longer. The only time we saw older cards get something from newer cards is when they added DSR to Kepler, as far as I remember.. and that was probably because it launched with Maxwell and custom resolutions were already a feature present in prior GPUs, so they couldn't get away with saying it was Maxwell-only.

 

Yep, I've not seen any decrease in performance from drivers for my Kepler card - I've been tracking synthetic & game benchmarks since I had this card (& before) for years. I agree with what you said about Pascal in your previous post - not impressive gains considering shrinking by 2 nodes & a new architecture too - I ran some numbers that I posted earlier in this thread after the NVidia Announcement & I calculated only a 65% increase in performance per Watt with Pascal over Maxwell - this I extrapolated from their Gaming Performance vs Watts slide they displayed (note not VR).

 

Yeah. All their claims about Perf per Watt and speed improvements are talking about VR usage alone. And hell, for all I know, that article where they claimed it had a 1.7x to 1.8x boost in Witcher 3 is most likely because they probably put a better tessellation engine in the cards, as I know Witcher 3 is a huge tessellation monster. Just like all their post-maxwell-launch Gameworks games. They can't render tech any other way because they gimp their own cards for everything but gaming and cuda performance, so they rely on tessellating more and more and more for new tech. Want better hair? Tessellation. Want god rays? Tessellation. Want anything in the world? Tessellation. #Gameworks. And then when older cards, even considering performance ratios, suck in newer games? People are mad. Why did the GTX 960 match or beat GTX 780s in Project CARS? Because there was ridiculous amounts of tessellation used. AMD cards had it so much worse it was a joke; R9 290X cards were performing like GTX 770s or something.

 

It can hold the clock speed, what are you talking about? Did I not use the word "stable" multiple times? You speak as though you never wanted Pascal, lol.

Here's another video from LTT:

 

Well technically it's still not out yet and there haven't been any reviews so far (maybe I'm wrong though), so don't see a point of being disappointed or hyped about it.
It will be out, reviewed, and then we will be able to draw proper conclusions instead of writing 200 pages worth of speculations...

 

Weren't you the one who said that numbers of ~1600MHz were seen? If the card, for any reason, hits ~1600MHz in a GPU-bound scenario, that means that it has problems with boost. Either there's not enough power or something else is the issue. Boost should be a constant, from what I've come to hear/see about these cards.

 

I can get an amazing deal on a msi Titan with dual 980 desktop versions in sli. Should I pull the trigger on this ? I don't think we will see a laptop with pascal beat this setup anytime soon .

 

65% boost is pretty good in my books. Something running at 30fps on the 980M will run at 50fps on the 1080M. Something running at 40fps will hit the magic 60fps on 1080M.

980Ti performance in mobile with same power consumption as 980M sounds good to me

 

What do you even have? Note that those 980s are pretty gimped and considering max OCs and power limitations, if you have a P870DM-G with 980Ms that OC well you're probably in a better state.

It's not that 61% boost is "bad". It's that it's "bad because there is a new architecture plus TWO die shrinks". Remember, Kepler was new arch + die shrink and had GK104 beating GF110 SLI. This is GP104 losing to GM204 SLI, far less GM200 SLI. But it costs more than Kepler did.

 

I would imagine 1080M sli would beat it when it comes out.

 

I currently have nothing , and I have a feeling I will be waiting a very long time to get a notebook with 1080m's. The Titan with desktop 980s has more than enough power for 1080p and should be upgrade able to pascal of we are to believe msi. But 4600 bucks is a lot to spend . I just wish I had a timeframe for mobile pascal. If it's more than 3 months out , I would just buy the Titan now

 

I see. If I were you and I played a lot of games that don't like SLI or enjoy new titles a lot, I'd quicker pick the P870DM-G with the 200W 980 and 6700K and basically keep a 4.5GHz 6700K and a ~1500MHz 980.

If you know what you're going to play plays well with SLI and stuff? Then maybe the MSI might be better. But in such a high end machine I'd think you'd want the better CPU and cooling.

As for mobile Pascal... I still have this ridiculous feeling it's going to be hotter than Maxwell; at least as hot as kepler.

 

I like both machines and many of my games do use sli. However dropping 4.5k now and then having 1080m notebooks next month would kill me lol

 

Well if they haven't announced it, it probably won't launch the same time. But maxwell resale value is going to TANK real bad when pascal comes out, for certain. But $4500 for that is... something heavy to dump. But I don't know. That's your choice. If you had something and wanted an upgrade I'd have waited. But even if something new comes out, a Prema mod is going to take a while still (and thus is boldly assuming Prema will make one, he does not HAVE to) to draw out its strength.

 

all of this is good, nice for desktop, but my main concern is about the upgradability of our machines (MXM).

 

We are all concerned about this. At least those of us who would rather upgrade their notebooks instead of sell them and buy completely new ones. But there's nothing we can do about that except speculate.

No info on mobile at all means we can't even make educated guesses as to what might be compatible (like will mobile get gddr5x? Will that change the layout and thus force a new heatsink? How will that affect models that have a shared heatsink between the cpu and gpu, etc)

 

I fear the end is drawing near, but hopefully that's not the case.

 

I want demand P870DM with desktop 1080. Nvidia, for the love of God, listen and do it right for once!!!

 

http://videocardz.com/59725/nvidia-gtx-1080-polaris-10-11-directx12-benchmarks
Well, the GTX 1080 performs well Ashes of Singularity.

 

That does not seem well to me lol, the fury is right behind it

 

1080M 2048 shader 16gb vram, calling it now, or 8gb gddr5x

103W.

 

With acer just releasing the desktop 980 predator laptop in June , maybe there is no mobile coming for a long while. This is definitely the most frustrating time to buy a laptop or desktop lol

 

All I can say is I hope that newer Pascal and Polaris cards work with MXM 3.0b.

 

And if it could be compatible with my P375SM-A ... it will be great

 

I'm sure if it wasn't by default, Prema could update BIOSes and add compatibility.

 

omg dude relax, we hear about your story about how you want to upgrade on every page

 

Every time I see "Acer" and "laptop" I must do this. That company does not understand how to make a decent laptop.

 

Maybe they're just trying to milk the current gen GPU's for all their worth. I'm thinking we will see mobile Pascal in August or September.

Sent from my SM-G935T using Tapatalk

 

wait were you guys saying 1060m should be around 980m performance?

what would the 1060m power draw be?

 

I had seen worse.

I had seen better.

For price / performance ratio, they have been good so far. They are the cheapest for what they offer as far as I know.

Well, that does not do it for us, enthusiasts, but it is good to know that they are not the worst. Just need more power... More cooling... more pixels... more speed...

Joking aside, they make only BGA laptops, and are doomed to BGA forever. I am willing to accept them as good for the money, but they are as far as BGA goes.

 

Man, is it me or do you ask the same questions over and over again? Not trying to be rude, just having deja-vu.

It won't be at 980M performance... It will probably be 30% better than the 960M. Low-end GPU's hardly ever improve more than that.

 

If it's GP107, it should be around 60% faster, give or take. Power draw? Heat? Those are entirely different things to consider.

If it's GP106, it'll be a different class of GPU and the 1050M is probably going to be the new entry-level card.

 

Cant remember when was the last time acer ever had a good product...

 

Do you have a world record? Because I have one with my Acer. There are far worse in terms of laptops from so-called boutique brands, but you don't see me going all pretentious about it.

 

Aftermarket cards coming soon...

 

I've run through enough Acers in my time, including my sister's now-obsolete (thankfully) netbook, to know that Acer mainstream products are generally garbage.

I don't think there are any worse in terms of laptops. Because places like HP don't make laptops. They make garbage and business laptops, and we aren't talking about business laptops here.

Fujitsu, HP, etc I don't even consider. As far as gaming laptops and cheaper mainstream laptops go, I recognize Dell, Acer, ASUS, Toshiba, Lenovo, Clevo, MSI, Gigabyte, *shudders* Razer *shudders* and I suppose EVGA.

 

You are not being fully fair to them

Dell have mostly low voltage CPUs these days, not even considering them worthy of seeing eye to eye with me. More expensive than Acer. Alienware, well they are so very expensive. And latest ones, really are looking grim. 2 slots for RAM in total, BGA GPU and CPU. Really?

Asus. really ? They cost up to 50% more than Acer variant of same specs. Usually worse quality control, especially on high end machines. Read theads complaining about screen bleeding. I wanted to get one, I know what I am talking about.

Lenovo. Does not make nothing high end. They stop at 960M as far as they go. I would avoid them, had one, and build quality was good, but not extraordinary,.

MSI - Expensive. Does not compare to Acer price wise, these are so much more expensive.

Gigabyte - I do not know anything of them, and neither about Razer.

Clevo - I think they are the only ones making a high end machine.

Acer - fully BGA, hard to open the things. Good quality components, they don't skimp to cheapest on market. Build quality very good, screen very good, cooling - eh. GPU never gets over 70C but CPU gets up to 95C. Cheapest high end laptops, yet not the worst. Very good performers. Not the best either, but a cheap and not the worst is a win if you want the best price / performance ratio. I had not tested a Predator yet, cannot speak for the series.



Depends what side of the coin you look at. Acer is a good cheap performer, not the best, neither the worst. Acceptable.