Caporegime
- Joined
- 8 Jul 2003
- Posts
- 30,063
- Location
- In a house
Turned out it wasn't maxed, and the Polaris 10 cards are only going to be around the 390s performance.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **
- Thread starter Boomstick777
- Start date
Turned out it wasn't maxed, and the Polaris 10 cards are only going to be around the 390s performance.
Soldato
- Joined
- 28 May 2007
- Posts
- 10,203
I believe it turned out that nobody knows the settings unless i missed something. All we know is that it held a steady 60fps.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Considering both Samsung and TSMC's 20nm are themselves different, that is complete nonsense. The important layers are smaller and ignoring that is just people not knowing what they are talking about. The metal layers are relatively easy and non complex anyway, them being 20nm really doesn't make a whole lot of difference, the significant electronic characteristics of the chip come from switching speed, voltage required for effective operation which is almost entirely based on the complex layers which are 14 or 16nm.
20nm metal layers has far less relevance than the 14/16nm complex transistor/finfet layers in terms of how the chip performs. 14/16nm are marketing names, but they are significantly more accurate than 20nm. The size is and has always been the name of the smallest parts made on a chip. Every chip ever made on 28nm has features built way way bigger than 28nm, that is just the absolute smallest possible. On 14/16nm finfets, the smallest feature size is 14/16nm.
Curiously it seems that TSMC 16nm is capable of higher speeds than Samsung 14nm but the Samsung process can sustain good speeds at very low voltages (better than TSMC) which seems to a degree counter intuitive.
Caporegime
- Joined
- 9 Nov 2009
- Posts
- 25,339
- Location
- Planet Earth
The 14NM process is also denser too,IIRC.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Not sure why you think that is counter intuitive, it's anything but. If one process is optimised for a lower operating range for voltage and is stable at say 0.85v where another process works better at 0.95v, then you'd absolutely expect the latter chip to clock better. If as a chip designer you are forced to use 0.95v over 0.85v, you create a less efficient chip by using the same clock speed on both, if you already have to take the extra voltage and the power usage that comes with it then taking the clock speed that comes from that voltage increases efficiency.
Though I've seen little to suggest which clocks better, frankly unless AMD produce some of the same architecture at both fabs we won't get a great idea of clock speeds. Different architectures are designed often with the process in mind and will have their targets changed based on process characteristics and outside of Intel, AMD and Nvidia, there isn't an awful lot of overclocking going on simply as there are few devices with as much control over the hardware as PCs.
That could be interesting in the future actually, Arm chips aimed at desktop on the same socket, would be nice to see an K12(or probably 1-2 gens later before we see something aimed at more consumer based). Seeing some Arm chips with stonking heatsinks and motherboards with overclocking in mind and seeing what happens with them.
- Joined
- 2 Jan 2012
- Posts
- 12,230
- Location
- UK.
What sort of performance level are we expecting the GTX 1070 to be at?
I mean if for example if it can beat the GTX 980 / Radeon Nano level of performance but uses less power and still retains that ~ £250 price range, than I could see that being quiet decent. Might even be my next card. Just don't have the need for the big guns any more, a lower power GPU but with good performance is all I want tbh.
Chances of 1070 beating the 980 / Radeon Nano are pretty good?
Thinking a GTX 1070 and a Kaby lake CPU could be nice cheap upgrades this year. Urge for the highest end has gone 0.0.
I mean if for example if it can beat the GTX 980 / Radeon Nano level of performance but uses less power and still retains that ~ £250 price range, than I could see that being quiet decent. Might even be my next card. Just don't have the need for the big guns any more, a lower power GPU but with good performance is all I want tbh.
Chances of 1070 beating the 980 / Radeon Nano are pretty good?
Thinking a GTX 1070 and a Kaby lake CPU could be nice cheap upgrades this year. Urge for the highest end has gone 0.0.
Won't happen unless you call 30fps playable
OR
You want play on medium settings which is worse than 1440p ultra.
OR
you play dots LoL and rocket league. Which can be done now on a single 980
(I have both)
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
The urge for the highest end has gone, says a guy with a Nano and a Skylake and you want to get a new 1070 and Kaby lake the second they are out..... really?
1070 won't be awfully priced but it also won't be cheap and Kaby lake certainly won't be cheap. A quad core with maybe 5-10% over your current CPU which is no where near maxed out for gaming for basically the same price. £200 for 5% for no end benefit is not a cheap upgrade.
Last edited:
I think the 1070 will be around 980ti performance for £300-£350.
If it isn't, it's not because it can't be, it's because Nvidia don't want it to be.
As always with Nvidia they'll cynically tune the card to market conditions, not the cards actual capabilities.
You can express disagreement without the low effort condescending response here man.
Personally, I agree with him. Anything less would be disappointing, frankly. We're still looking at a 300mm+ die on a successive node shrink(after skipping 20nm) that's been a long time coming. If they can get a 970 to basically match a Titan/780Ti on the same process, why couldn't they do it now? And that was with a £270 pricetag.
This is how it works for anyone. There will most likely be a cut down GP104 and a full fledged GP104. This isn't 'tuning' to market conditions out of greed or cynicism, this is simply filling in spots in their range of products. AMD do the same thing.
Nvidia certainly aren't going to gimp the GP104 as a whole. They'd either end up lacking in competitiveness against AMD or they'd miss out on a crushing blow if it turns out they cut things close. And Nvidia is certainly not going to miss an opportunity to secure or gain more market share.
Also if the current chip size rumours and leaks are correct Polaris is 100mm^ smaller than GP104 and that is expected to match a FuryX in some situations, then the GP104 being much bigger should definitely surpass a 980ti, leaving plenty of room for the cut down 1070 to match the 980ti.
Indeed. If the 1070 doesn't at least match a 980Ti we are in for one very disappointing generation!