nVidia X80, X80ti and X80 Titan rumoured Specs

[Citadelen]

4 hours ago, Valentyn said:

Polaris is going to be the "mid range series" to get the ball rolling with 16nm, Vega is their big boys. It's due very late this year, to mid next year.

Think of it as NV Mid Pascal to Big Pascal( next Titan )

All of AMD's CPU's and GPU's will be using 14nm FinFet LPP. Not 16nm.

[Sakkura]

52 minutes ago, Prysin said:

didnt micro say that 14GT/s wasnt going to happen before 2017 or 2018?

Probably. But what matters is whether the speed they want is achievable on GDDR5, or requires more expensive GDDR5X. 8 GT/s speeds should be achievable on GDDR5, since they've been shipping cards with 7 GT/s GDDR5 for a long time now. If they needed speeds like 10 GT/s, 12 GT/s or whatever, then GDDR5X might be a necessity (as and when it becomes available).

 

Of course there's some weighing of bus width vs. memory clocks vs. memory amount going on as well - eg. achieving similar bandwidth with fast memory on a 256-bit interface vs. "slow" memory on a 384-bit interface. And 256-bit lends itself best to "even" memory amounts, ie. 2-4-8-16 GB, while 384-bit would work best with 3-6-12-24 GB.

 

So eg. for the X80 they might have decided that 384-bit slow, plentiful, and cheap GDDR5 would be better than 256-bit fast, scarce, and expensive GDDR5X, which would "force" them to either step down to 4GB (arguably not enough) or up to 8GB (arguably overkill, and aggravating the price hike).

 

All speculation of course. These specs might all be fake, even if they mostly look plausible (the only odd thing is GP100 working with two different types of memory).

[Lazmarr]

6 hours ago, Paragon_X said:

Don't they have the Memory Interface (bits) and Memory Bandwidth (GB/s) mixed up?

Should't the Memory Interface be 1024 and the Bandwidth be 4096?

[Sakkura]

2 minutes ago, Lazmarr said:

Don't they have the Memory Interface (bits) and Memory Bandwidth (GB/s) mixed up?

Should't the Memory Interface be 1024 and the Bandwidth be 4096?

No. Look at the clocks. And welcome to HBM.

[Lazmarr]

Just now, Sakkura said:

No. Look at the clocks. And welcome to HBM.

Wasn't HBM supposed to increase bandwidth with each 'stack' ie 2 stacks would be 2048GB/s or have I got this wrong?

I must admit that I haven't looked into HBM for quite a while now lol and have forgotten most of it

[Sakkura]

1 minute ago, Lazmarr said:

Wasn't HBM supposed to increase bandwidth with each 'stack' ie 2 stacks would be 2048GB/s or have I got this wrong?

I must admit that I haven't looked into HBM for quite a while now lol and have forgotten most of it

Well, this is counting all the stacks, just like the cards with GDDR5 are counting all the GDDR5 chips together. In terms of performance, HBM just runs at lower clocks but with a MUCH wider bus, to deliver higher bandwidth.

[Mr_Troll]

those specs seem fake to me ... because there has never been a big diference between titan a ti version of a graphics card.... the ti version was always only like 10% slower than the titan.... 

[goaway]

4 minutes ago, Mr_Troll said:

those specs seem fake to me ... because there has never been a big diference between titan a ti version of a graphics card.... the ti version was always only like 10% slower than the titan.... 

So because Nvidia may have fixed a flaw in their GPU hierarchy, the rumours are false? Not saying the rumours are true, just seems like a weak argument IMHO.

[MEC-777]

1 hour ago, Sakkura said:

Probably. But what matters is whether the speed they want is achievable on GDDR5, or requires more expensive GDDR5X. 8 GT/s speeds should be achievable on GDDR5, since they've been shipping cards with 7 GT/s GDDR5 for a long time now. If they needed speeds like 10 GT/s, 12 GT/s or whatever, then GDDR5X might be a necessity (as and when it becomes available).

 

Of course there's some weighing of bus width vs. memory clocks vs. memory amount going on as well - eg. achieving similar bandwidth with fast memory on a 256-bit interface vs. "slow" memory on a 384-bit interface. And 256-bit lends itself best to "even" memory amounts, ie. 2-4-8-16 GB, while 384-bit would work best with 3-6-12-24 GB.

 

So eg. for the X80 they might have decided that 384-bit slow, plentiful, and cheap GDDR5 would be better than 256-bit fast, scarce, and expensive GDDR5X, which would "force" them to either step down to 4GB (arguably not enough) or up to 8GB (arguably overkill, and aggravating the price hike).

 

All speculation of course. These specs might all be fake, even if they mostly look plausible (the only odd thing is GP100 working with two different types of memory).

Didn't they not say that GDDR5X was/would be relatively easy to implement vs GDDR5? Didn't they say it only requires a few "minor" changes?

[SuperCookie78]

i don't believe this at all. 

[Dabombinable]

8 hours ago, Samfisher said:

Well, playing current gen games with a GPU 2 generations old, what do you expect.  VRAM usage has stagnated since Mordor and modded Skyrim, and any game that uses more than 4GB VRAM will be played at resolutions the 970 would have problems with anyway.

Skyrim at 2K DSR, all of the vRAM is used and its still won't struggle-at 1080p its only loaded around 50%.

[jollander]

Mini-rant with the assumption that this proves to be real.

 

What a ridiculous naming scheme; three x80's! Doing this is only to make the x80 feel more premium than it should be. It was bad enough naming 2 *80 variants (eg 980 980ti).

 

This entire set should be bumped down one level, starting at x50 instead of x60.

x50 (prev. x60)

x60 (prev. x70)

x70 (prev. x80)

x80 (prev. x80 ti)

x90 (prev. x80 titan)

 

The older logic that *90 versions means dual gpu cards is kinda invalidated since they stopped doing that several generations ago. Also, that old naming scheme only applied to two cards; 590 & 690.

[VerticalDiscussions]

Such an absolutely beatiful card, that X80 Titan! Yet, the price tag is a definite 1500€ tag .-.

[BuckGup]

9 hours ago, Starelementpoke said:

I´d still wait for Polaris numbers to drop.

Well of course. I think it will be like the gtx980ti and fury x again but I hope I am wrong

[BobbyG2]

I'm expecting the Titan to cost at least $1500, possibly even $2000.

 

The reason the Titan X was $500 less than the Titan Black was because they nerfed DP performance,  making it less useful for professional use.  Now that Pascal apparently beefs up the DP performance again,  I expect the price to hop back up again.

 

Add to the fact their using 16GB of HBM2, which might not have good yields yet,  makes me think it might even cost more than the TB.

[Sakkura]

2 hours ago, MEC-777 said:

Didn't they not say that GDDR5X was/would be relatively easy to implement vs GDDR5? Didn't they say it only requires a few "minor" changes?

Sure, it should be pretty easy. Doesn't mean it won't cost more, especially in the beginning.

[Briggsy]

10 hours ago, Starelementpoke said:

I´d still wait for Polaris numbers to drop.

yeah, if current performance of Polaris 10 and 11 are to be believed, the 232nm Polaris 10 (probably R9 480) is performing at around the level of a 980ti/FuryX in DX12. The big-dog Vega architecture is supposed to drop end of 2016 or early 2017, when HBM2 production will finally be in full swing. These are only canned DX12 benchmark numbers using Polaris engineering samples, which means the final production cards should be improved.

 

I don't doubt for a second that a huge performance gain is coming with 14nm, as we saw a massive performance jump from both Nwidia and AMD when they moved from 40nm to 28nm. Peoples GTX 580's were being stomped by cheap 660's, and AMD's 7000 series obliterated their 6000 series. 

[niofalpha]

All those cores... I'm ready for them to be jet engines...

[Tech Enthusiast]

8 hours ago, Notional said:

This is and always will be a fallacy. Textures take up a lot of VRAM, but requires very little GPU power. Rise of the Tomb Raider requires more than 4GB of VRAM for very high. Doom seems to need that too for highest setting. So the more the merrier.

 

 

 

 

Do you have any first hand experience on these claims, or just taken it from "someone"?

 

Not saying i fully understand how VRam works apart from "more is usually better", but on a GTX 960 with 2gb VRAM Tomb Raider runs at 55-65ish fps with textures maxed as long as some of the more taxing stuff is turned way down (custom settings mind you, presets usually screw over performance or quality... or both). 

 

So,.. why do i need more then 4 gb vram if 2gb seem to work just fine? I don't exactly understand how that works to be honest, just wondering about my personal results that seem to conflict greatly with what you just posted. Why would i care if a game CAN fill up more vram, if it does not run bad if it can not fill up all that extra vram? 

[LokiFire]

11 hours ago, Dabombinable said:

People kept on telling me that I'd never use the 2GB vRAM on my GTX 650ti, that its GPU wasn't powerful enough. And guess what? At 1080p I have it running out of vRAM before the GPU becomes the limiting factor. (Same games+settings, my GTX 970 runs at around 50% load and has 3GB vRAM used).

Full of poop my friend.

[zacRupnow]

So no "GTX 1080 designed for 4k" awwww man, I was so ready for those memes.

At least "Geforce X" sounds like "G for Sex" honestly Nvidia...

[Notional]

41 minutes ago, Rattenmann said:

Do you have any first hand experience on these claims, or just taken it from "someone"?

 

Not saying i fully understand how VRam works apart from "more is usually better", but on a GTX 960 with 2gb VRAM Tomb Raider runs at 55-65ish fps with textures maxed as long as some of the more taxing stuff is turned way down (custom settings mind you, presets usually screw over performance or quality... or both). 

 

So,.. why do i need more then 4 gb vram if 2gb seem to work just fine? I don't exactly understand how that works to be honest, just wondering about my personal results that seem to conflict greatly with what you just posted. Why would i care if a game CAN fill up more vram, if it does not run bad if it can not fill up all that extra vram? 

Yes, it's actually one of the reasons I picked a 290 instead of a 780 back in the day. I can run games with higher texture settings than people who have less than 4GB VRAM, but only in games that needs it. I hardly see any performance difference when changing texture settings, with the exception of settings that exceeds the VRAM I have on my GFX. When that happens, performance will tank extremely in choppy drops down to 1fps. That happens for me in Rise of the Tomb Raider when using the highest texture setting. I can't feel any difference in any of the settings below that.

Textures simply aren't taxing for a GPU at all. Geometry, physics, shadows, global illumination, post processing. Now THAT is taxing.

 

Because you can get higher resolution textures with more VRAM. Textures is almost the most important part in making nice graphics. It is the least taxing setting that makes one of the biggest differences in the game.

 

Try these examples:

http://images.nvidia.com/geforce-com/international/comparisons/rise-of-the-tomb-raider/alt/rise-of-the-tomb-raider-texture-quality-interactive-comparison-003-very-high-vs-low-alt.html

Note the wall in the background and especially the generator on the left (move the slider).

 

http://images.nvidia.com/geforce-com/international/comparisons/rise-of-the-tomb-raider/alt/rise-of-the-tomb-raider-texture-quality-interactive-comparison-002-very-high-vs-low-alt.html

Look at the mud part on the left.

 

http://images.nvidia.com/geforce-com/international/comparisons/rise-of-the-tomb-raider/alt/rise-of-the-tomb-raider-texture-quality-interactive-comparison-004-very-high-vs-low-alt.html

Painting.

 

The difference on a 980ti between those two settings in the examples above are a measly 3-4 fps out of about 50fps.

[Lays]

12 hours ago, TrigrH said:

why would the cut down titan not use HBM, such bs

 

 

12 hours ago, Master Disaster said:

Because HBM 2 yields have been very low and theres just not enough of it in existence.

 

 

Doesn't make much sense whatsoever though, is the GPU die produced at the same time the HBM is added onto it? Or is it an after production thing?   That means they will have to mess with the IMC most likely, but if it's the same chip, I don't know how that'd work..   

 

Also, HBM 2 yields probably won't be a big problem it seems for what dates keep getting leaked out in rumors for the TI variant, isn't it April of NEXT YEAR?

[thekeemo]

12 hours ago, jasonvp said:

Sign me up for two Titans please.  Just make sure they don't have those God-forsaken DVI ports on them, please!

cmon DVI is the best port after DP

[BuckGup]

3 hours ago, BobbyG2 said:

I'm expecting the Titan to cost at least $1500, possibly even $2000.

 

The reason the Titan X was $500 less than the Titan Black was because they nerfed DP performance,  making it less useful for professional use.  Now that Pascal apparently beefs up the DP performance again,  I expect the price to hop back up again.

 

Add to the fact their using 16GB of HBM2, which might not have good yields yet,  makes me think it might even cost more than the TB.

No way would it cost that much. The Titan Z costed around that but that was dual GPU. Seeing Polaris and the rumors we may have a price war. HHEHEH