Nvidia GeForce GTX 1080 (16nm Pascal)

[kalniel]

Hi I am actually referencing to NVidia presentation of this card where they explicitly said they weren't just gunning to enjoy the performance gain from new node. They actually invested billions/ "several thousands of engineers worked for several years " to make even more. I do not see that. Do you?

I do see that. They've not only increased transistor density (new node) they've also done it with a massive increase in frequency. That takes a huge amount of work - you don't simply shrink transistor design and hope it still goes OK, let alone goes way faster.

You understand that gain they made comes from new node, no extra as they told.
Most of other thing presented were software solutions (where they are stronger than AMD), but nothing breathtaking.

We need higher performance on high end in order to have great performance at middle class, the one sells the most to gamers, and to have higher baseline.

I would say this is most certainly higher performance at the middle-high end: compared to a 980 the performance jump in such a short space of time is impressive.

They probably have GTX 1080 ti version ready, just wait for AMD to see how good (or bad they will be).
If they weak you will pay 1080 599USD for sometime. If AMD comes strong (at the end of the year) then nVidia will come up with 1080 Ti for that price.

They already announced GP100 some time ago so we know there is a bigger version of the chip, using HBM2 as well, so if you want the truly high end there is something coming for you.

So I will wait.

Waiting is the sure-fire way to get better performance for less

[CAT-THE-FIFTH]

The HD7970 was still around £400ish and had no competition from Nvidia for a few months - people are complaining that was overpriced. Even then the price matched the GTX580 3GB,which Nvidia never really dropped the price of,and yet everybody was blaiming AMD for the high price of that generation.

This card is more expensive than the GTX980TI. It should be matching the GTX980TI pricing.

Plus we had cards like the Titan Z which still cost more than the R9 295X2 and were worse cards.

At least in the last decade it has been Nvidia setting pricing tiers not AMD.

If anything it has meant AMD progressively upping the price of their own flagship cards now.

Should we not forget the GTX480??

According to TPU it was 10% faster than the HD5870 and was six months late and cost more.

[CAT-THE-FIFTH]

Another thing I noticed:
1.)GTX680 launch price $499
2.)GTX980 launch price $549
3.)GTX1080 launch price $699 then dropping to $599(maybe)

These are all first tier small GPU flagship cards.

I do see that. They've not only increased transistor density (new node) they've also done it with a massive increase in frequency. That takes a huge amount of work - you don't simply shrink transistor design and hope it still goes OK, let alone goes way faster.

It actually has less transistors than the GM204 - so whilst per transistor performance has gone up WHICH IS WHAT I FIND IMPRESSIVE,it is also the first time in a long time we have seen a replacement card have less transistors than the card it is replacing pricewise.

I would say this is most certainly higher performance at the middle-high end: compared to a 980 the performance jump in such a short space of time is impressive.

But it isn't middle higher end - even at the purported lower end price it is more expensive than the GTX980 in USD and the Founders Edition is true high end price.

It is competing against the GTX980TI and not the GTX980.

20NM had problems and it is most likely both the GM204,GM200 and Fiji might have been 20NM designs but made on 28NM. So they were kind of stop-gap.

They already announced GP100 some time ago so we know there is a bigger version of the chip, using HBM2 as well, so if you want the truly high end there is something coming for you.

Waiting is the sure-fire way to get better performance for less

Its not the same - if you look at the GP100 it has a fair bit of spaced towards dedicated FP64 performance. 1/3 of the shaders do 64 bit precision only which is not relevant to gaming.

That means a fully enabled GP100 would have the same number of FP32 shaders as a GP104.

There are hints of a GP102 which were leaked in some drivers. It could be an upscaled GP104.

[jigger]

This is what happens when real competition, for the time being at least, is absent from the high-end graphics space.

The real competition is to get this kind of money form peoples wallets for a graphics card.

[CAT-THE-FIFTH]

The real competition is to get this kind of money form peoples wallets for a graphics card.

Cheaper cards like this Galaxy for £500 should be released at some point.

[shaithis]

All the talk of prices.......do people forget all the news and graphs regarding node costs?

I for one, expected the price to be higher, as I do with Polaris when it gets here. Unless there is a great conspiracy over process shrinks, I can't see how anyone expects otherwise.

[DanceswithUnix]

Unless there is a great conspiracy over process shrinks, I can't see how anyone expects otherwise.

Partly because AMD said they will be making VR capable machines cheaper to make which sounds like a cost reduction to me: http://arstechnica.co.uk/gadgets/201...range-polaris/

So yes those charts said that each square mm of silicon might be more expensive, individual transistor cost isn't much better if at all, but with the bulk leakage cut right back you can divert all that power into clock speed, coolers aren't so expensive, power regulation for 180W is way easier and cheaper than for 300W so there is plenty to win with here.

There is an element of 14/16nm being capable of feats that just weren't possible at 28nm regardless of cost, but for a given level of performance I was expecting a cost drop.

[kalniel]

There is an element of 14/16nm being capable of feats that just weren't possible at 28nm regardless of cost, but for a given level of performance I was expecting a cost drop.

There sort of is? Take previous gen top tier performance and you now get it on a card with MRSP almost half the price. But first chips on new nodes always cost a lot - the price comes down in time. I suspect they're also selling fewer of these cards than they used to, so per unit costs are higher in terms of R&D share etc.

It actually has less transistors than the GM204 - so whilst per transistor performance has gone up WHICH IS WHAT I FIND IMPRESSIVE,it is also the first time in a long time we have seen a replacement card have less transistors than the card it is replacing pricewise.

GM204 had 5.2B transistors, 1080 has 7.2B doesn't it? They're pricing it against a higher tier simply because they can - as I said elsewhere, pricing is more market determined. Absence of competition = freedom to make margins.

[DanceswithUnix]

There sort of is?

Indeed, but I think we were expecting more than "sort of". But we got a bit of a performance improvement for a bit of a price hike and it all feels a bit feeble.

If AMD are indeed going to produce something no faster than before but for lower price it will be interesting how people react to that. That will be down to how low the price is though.

[CAT-THE-FIFTH]

Indeed, but I think we were expecting more than "sort of". But we got a bit of a performance improvement for a bit of a price hike and it all feels a bit feeble.

If AMD are indeed going to produce something no faster than before but for lower price it will be interesting how people react to that. That will be down to how low the price is though.

It also serves to artificially pump up the price of the rest of the range like the Fury X did,so to make them look better value.

If the GTX1070 is £350 to £400 and is only reference GTX980TI level performance,it will look better value than a GTX1080 but it would be a disaster for anyone buying sub £300 cards as according to TPU it would be barely 10% faster than a Nano,and probably worse value with regards to performance/pound than a GTX970 or R9 390.

It would also mean the GTX1060 and GTX1060TI will be a feeble improvement and give AMD a chance to jack up Polaris 10 and 11 prices.

So,what is needed is the GTX1070 be a bit better than a GTX980TI at £350(or maybe even cheaper ),if we have any chance for a decent improvement in sub £300 cards,ie,the more mainstream market.

People forget both are quite capable of jacking up prices quietly when it suits them.

[Corky34]

I for one, expected the price to be higher, as I do with Polaris when it gets here. Unless there is a great conspiracy over process shrinks, I can't see how anyone expects otherwise.

I thought that's why most nodes switched to FinFET, IIRC costs only started to rise when they started going under 20nm with the traditional planar technique, i thought FinFET postponed that cost increase per node shrink down to something like 10-12nm.

[HalloweenJack]

and still the 295X2 is fastest single card!"

[rjs675]

still don't understand reference cards being more expensive than board partners now...

[kalniel]

and still the 295X2 is fastest single card!"

Pro Duo might like a word, if they got the drivers sorted out..

[DanceswithUnix]

Pro Duo might like a word, if they got the drivers sorted out..

Pro Duo seems, well, a tad costly?

When Nano cards turn up at £350 every now and then, paying £1200 for a duo seems ott.

[kalniel]

Pro Duo seems, well, a tad costly?

When Nano cards turn up at £350 every now and then, paying £1200 for a duo seems ott.

Fastest is fastest. It wasn't qualified by price