** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

[r7slayer]

Nope, even the G1 was only £299.

http://forums.overclockers.co.uk/showthread.php?t=18626392&highlight=900+series

An Inno 3d one there for £260 and a palit for £251

Average price on launch was around £270.

So it was! I remember it being £300 + for some reason.

Even still i doubt this time around we will see the same price due to it being new mem tech not only a node drop which appears to be much more expensive to fabricate chips on.

Id be expecting a high price point this time around.

 

[Jono8]

So it was! I remember it being £300 + for some reason.

Even still i doubt this time around we will see the same price due to it being new mem tech not only a node drop which appears to be much more expensive to fabricate chips on.

Id be expecting a high price point this time around.

I agree. I reckon we might see a return to the ~£325 mark for the X70 card and the X80 will probably be £450+, assuming the X70 matches a Ti and the 1080 beats it. Especially considering it looks like the X80 will have GDDR5X.

 

[Tommy_Here]

the 1070 will be using standard GDDR5 so no new tech there. 570 to 670 40nm to 28nm price increased £30 so take the most expensive card I listed the 670 at £330 add £30 so at most a £360 card that could match or just edge out Ti thinking titan x performance for that price it'd be a fantastic card for nvidia 2 of them for less cash than a titan X and not that much more than your higher end 980 Tis

 

[drunkenmaster]

Paper launch according to WCCF, but beating AMD to launch.



http://wccftech.com/geforce-gtx-1080-gtx-1070-launching-6th-may-market-june/

Not really, because their definition of a paper launch, which they claim from the headline is the NDA press briefing on the 6th of May means AMD launched in December so Nvidia clearly weren't first.


Honestly first up I'd say this is the usual from websites, they take new rumours(more so if they are official/confirmed events) and add them into old made up rumours. What we know is an NDA event is happening on May 6th, wccf and others have been pulling June/July launch rumours out of their behinds for a long time(well since their GTC fake rumours were proven false), so they take those old rumours and add in the new piece of real information and package it all up as completely real.

Remember when they said Titan was launch at GTC when they knew Nvidia would be talking at GTC, then they said GP104 would be spoken about at GTC.

Outside of the May 6th event I wouldn't presume the rest is true, but this is what these websites do, take a shred of truth and use one true thing to attempt to gain credibility for the rest of the story. Could it be true, possible but I don't see why this is more credible than the dozens of stories proven false they've come up with on Pascal. Put it this way if they were accurate and had good sources how are they wrong 100 times more often than they are right?

If it does launch in June/July, which I wouldn't put it past Nvidia, lets see what volume is. The last time Nvidia was insistent on 'launching' first and showing fake products they were miles behind and used events to deflect from the delays. If there are dozens or even a few hundred cards available in June then it's not quite a paper launch but also not a real launch.

 

[tyler_jrb]

Well the 970 did pretty much match the 780TI for ~£270....

No it didn't. On release the 780Ti was equivalent to a 980. In some cases still faster. That performance gap has only increased because of nvidias lack of support for the 700 series.

GTX 1070 is looking to be about £400 and the 1080 £500.

New node, new gddr5x VRAM. Ain't going to come cheap.

 

[Seanspeed]

No it didn't. On release the 780Ti was equivalent to a 980. In some cases still faster.

Not true, though there was the odd title a 780Ti beat a 980. Not many at all, though. 980 was about 10% faster on average. With the 970 only being inches behind.

https://www.techpowerup.com/reviews/MSI/GTX_970_Gaming/27.html

That performance gap has only increased because of nvidias lack of support for the 700 series.

It's because there was a lot more room for improvement optimizing drivers for GM204, which was brand new. Whereas GK110's drivers already reached most of its potential at that point.

Driver gains are biggest in the first year and then taper off massively afterwards.

 

[Rroff]

Not true, though there was the odd title a 780Ti beat a 980. Not many at all, though. 980 was about 10% faster on average. With the 970 only being inches behind.

Yeah my 780GHz came out the box at stock 780ti performance and I had to take it upto its max overclock to match a stock 980 at the time (things stack up a bit differently these days though).

 

[Jono8]

No it didn't. On release the 780Ti was equivalent to a 980. In some cases still faster. That performance gap has only increased because of nvidias lack of support for the 700 series.

GTX 1070 is looking to be about £400 and the 1080 £500.

New node, new gddr5x VRAM. Ain't going to come cheap.

970 is very close, trading blows, even faster in a few tests.

http://www.anandtech.com/bench/product/1595?vs=1441

980 is faster across the board ( on average at least around 10% faster from a glance at these numbers)

http://www.anandtech.com/bench/product/1442?vs=1441

You can't bring drivers into it, as that other guy pointed out. The 780Ti had the maximum potential brought out of it that it could during its time, just like the 970 and 980 have had now.

 

[tyler_jrb]

I should have said. Yes a 980 is and was faster in the majority of games but not by a huge margin. 5-10% in games on release.

It was and still is faster than a 970 though in the majority of games. Not forgetting the price drop the 780Ti's took after the 980/970 release. £260 odd?

1664 cores vs 2880 though. That puts a 1664 core 1060/70 type card in a very good position for pascal. Especially if the 1080 supposedly has 2560 and gddr5x.

 

[seklas]

I'd say these two rumoured/confirmed x70 and x80 GPU are pointless at this point and very limited once again, because nVidia knows just how much everyone is eager to get one of them. It'll be selling really well anyway. In the last gen (Maxwell), the fastest affordable GPU was GTX 980 Ti - had 6 GB VRAM, however the GTX 980 for laptops (not Mobile version) already had full 8 GB of VRAM and in tests it performed a few frames better than GTX 980 (desktop).
We all know that the Pascal can have up to 32 GB VRAM, but if they are planning to use only 8 GBs at launch, it's pointless. In this case, if it would become the truth, the x80 Ti would have 16 GB of VRAM and the new Titan - 32 GBs in Q1, 2017, because the generation will end in 2018, so they need to get everything from the Pascal before then. But the gap between x80 - Ti - Titan would be too wide, especially if they would utilise HBM 2 as well. So could it be that Nvidia is trying to change their marketing strategy on desktops this time around and introduce even more versions of graphics cards just like what they've done with mobile GPUs in the past? If we put the new Ti version as x90, then it could be as follows:
x60 - 6 GB GDDR5
x70 - 8 GB GDDR5
x80 - 8 GB GDDR5X
x85 - 12 GB GDDR5X/HBM 2
x90 - 16 GB HBM 2
Titan - 32 GB HBM 2

It looks pointless doesn't it? In this case, I'd say those "confirmed rumours" might still be wrong and we'll see something like:

x70 - 8 GB GDDR5
x80 - 12 or 16 GD GDDR5X
x80 Ti - 24 GB HBM 2
Titan - 32 GB HBM 2

There is still a possibility that we won't get 32 GB VRAM at all, because nVidia might want to drag it to next the generation after Pascal, but it should hit them quite hard, because they've promised those 32 GB VRAM capabilities to their customers.

 

[DarrenM343]

Given that that the Titan X's full potential is realised with 2+ cards (due to memory), 32GB for the next generation seems a little extreme. I personally doubt it's true unless this has been mentioned officially. More like 20-24.
I expect HBM2 memory will be a lot more expensive so with 32GB the next TItan X successor would be very expensive, even compared to the high price of the curent TX.
I thought I read somewhere that HBM2 works differently anyway so less memory is required? Maybe it was the marketing pitch for the Fury X I don't know.

 

[Seanspeed]

I'd say these two rumoured/confirmed x70 and x80 GPU are pointless at this point and very limited once again, because nVidia knows just how much everyone is eager to get one of them. It'll be selling really well anyway. In the last gen (Maxwell), the fastest affordable GPU was GTX 980 Ti - had 6 GB VRAM, however the GTX 980 for laptops (not Mobile version) already had full 8 GB of VRAM and in tests it performed a few frames better than GTX 980 (desktop).
We all know that the Pascal can have up to 32 GB VRAM, but if they are planning to use only 8 GBs at launch, it's pointless. In this case, if it would become the truth, the x80 Ti would have 16 GB of VRAM and the new Titan - 32 GBs in Q1, 2017, because the generation will end in 2018, so they need to get everything from the Pascal before then. But the gap between x80 - Ti - Titan would be too wide, especially if they would utilise HBM 2 as well. So could it be that Nvidia is trying to change their marketing strategy on desktops this time around and introduce even more versions of graphics cards just like what they've done with mobile GPUs in the past? If we put the new Ti version as x90, then it could be as follows:
x60 - 6 GB GDDR5
x70 - 8 GB GDDR5
x80 - 8 GB GDDR5X
x85 - 12 GB GDDR5X/HBM 2
x90 - 16 GB HBM 2
Titan - 32 GB HBM 2

It looks pointless doesn't it? In this case, I'd say those "confirmed rumours" might still be wrong and we'll see something like:

x70 - 8 GB GDDR5
x80 - 12 or 16 GD GDDR5X
x80 Ti - 24 GB HBM 2
Titan - 32 GB HBM 2

There is still a possibility that we won't get 32 GB VRAM at all, because nVidia might want to drag it to next the generation after Pascal, but it should hit them quite hard, because they've promised those 32 GB VRAM capabilities to their customers.

I dont think they've ever promised they'd produce 32GB cards. Only ever said that HBM2 can potentially go that high.

That said, I think any 32GB cards they do make will be professional products, not gaming cards. There's just absolutely no need to have 32GB of vRAM for gaming. It's massive overkill, so it would just significantly increase the cost for absolutely no benefit.

In general, you're way overstressing the significance of vRAM at the moment. I'm doing just fine with my 3.5GB card still. I know we're going to see iterative consoles within the next year or so, but they will not have more memory, from what we've been told. So these games will still be limited to the 5GB or so of vRAM they've got available(unified memory, so not all 8GB's can go to graphics). We're simply not looking at any massive spike in vRAM usage anytime soon. Right now, 4GB is enough for the vast majority of modern games, 6GB is more than enough for anything at 1080p, and 8-12GB is really only necessary if you're gaming at 4K, and even then, 12GB is overkill in all but a very small minority of cases.

I expect the range to look something like

GP106 - 4GB GDDR5 standard, 8GB 'option' for higher end model
GP104 - 8GB GDDR5 standard, possible GDDR5X for x80?
GP100 - 8GB HBM2 for cut down model and Ti, 16GB HBM2 for potential prosumer card(Titan equivalent)

8GB of HBM2 will be beastly, dont get too hung up on the quantity. Very few games can produce a memory bottleneck with Fiji's 4GB of HBM1.

 

[D1craig]

Well i remember the 970 being a £300+ card at launch and wasn't the 980 a top end £400 card at launch? Only the Ref 970 maybe have been that price.

Your memory is definitely failing you the 970 was a sub £300 card. My EVGA sc was only 260bajd that was on the day it released and from OCUK.

The OCUK reference titan cooler might have been 299.but that and Asus where as close as you got.

 

[drunkenmaster]

I'm not sure you understand what sub £300 means, here's a hint, £260 is sub £300.

 

[AllBodies]

Seeing as the professional-grade P100 card has 16GB HBM2, I'd say it's almost certain the Pascal Titan won't have 32GB VRAM.

Should have 16GB though, they'll need to differentiate it from the 1080 Ti, which will likely have 8-12GB.

 

[tyler_jrb]

12gb is still overkill for 4K. I've ran 2 4gb 980s at 4K with no problems.

8gb is about the sweetspot at the minute for 4K for the next few years.

1080p and 1440p could still easily survive on 4gb.

New Titan won't have 32gb HBM2 that is for sure. That would be such an expensive card.

I'd think it will still have the same 12 or 16gb as a maximum. May have less depending on what hbm2 is like.

 

[tavtavtav]

I'm not sure you understand what sub £300 means, here's a hint, £260 is sub £300.

Isn't that exactly what he said?

 

[JediFragger]

I'm going with 8GB of GDDR5X for the Ti (if it does indeed launch) and 16GB of HBM2 for the Titan later this year/next. Seems reasonable to me

 

[Vertigo]

I'm not sure you understand what sub £300 means, here's a hint, £260 is sub £300.

And a hint for you, try reading his post again, and the one he was quoting

 

[seklas]

In general, you're way overstressing the significance of vRAM at the moment. I'm doing just fine with my 3.5GB card still.

In recent years the use of VRAM has actually spiked really high in such a short time. Knowing that Pascal and Polaris should allow for even more players to achieve a good 4K and VR performance for more reasonable price - 8 GB won't be enough, and I don't think 8 GB of HBM2 memory will change anything, because in that case it would mean that system requirements of games for VRAM would drop a little bit in numbers, but that has never happened before, it only goes up. So I think that if they'd stay with 8GB or 12/16 GB of HBM 2, developers would quickly start creating even more, better looking, unoptimised games, just to pointlessly get the full use of it. And if it's really so much faster than GDDR5 or GDDR5X, then it'll be the end for both very soon.

Knowing that at the end of each generations, the high-end card usually ends up being a recommended card, in this case GTX 980 and 980 Ti has already been listed for recommended or ultra specs in a few games, means that GTX 1080 will reach that point before Q1 2017. In that case the card probably won't be any faster than 980Ti even with GDDR5X, or the requirements will jump up that much. Of course that could be the case only with stock speeds, as overclocking capabilities are still unknown.