The Mill Crunches Away: Alleged NVIDIA GTX 1170 Benchmarks Surface

[ppn]

980Ti XtremeGaming 6GB was +20% outof the box. My Gigabyte 780 Windforce throttled patheticly to 1Ghz. 2 years apart, same price, 2x perf. And that is on the same 28nm node. So we are either getting that or wait for 7nm.

 

[Fluffmeister]

It's impressive with what they achieved with Maxwell on the same 28nm node for sure. The 980 Ti was a gem of a card.

Certainly happy to keep mine for a while longer yet.

 

[efikkan]

well i really wish GTX 1160 to be great, 1170 to be awesome and dunno what to wait about 1180 and 1180ti... but i really doubt GTX 1160-70 having more than 8GB and GTX 1180ti havin 12 GB memory,

i do really hope GTX 1160 having a 256 bit memory bus, and possible 2x6 pin connector, possible match for GTX 1070 or ti, on the mid range cards,

Why don't you focus on finding out which performance level you need for the next ~3 years, and then pick the card that satisfies that, rather than considering cards before we know their performance

Don't be too focused on theoretical specs. If Nvidia puts 8 GB on "GTX 1180", then it will probably be enough. Memory is still expensive, and GDDR6 is slightly more expensive than GDDR5/X.

 

[peche]

better forget that 256 Bit now, for a 128 bit is much more likely

it died with the Glorious GTX 760, noble card,

nd the 980 was all around a bit faster than a 780ti even TPU's database shows that.

=
GTX 980 is the most forgotten card in TPU,

Why don't you focus on finding out which performance level you need for the next ~3 years, and then pick the card that satisfies that, rather than considering cards before we know their performance

Don't be too focused on theoretical specs. If Nvidia puts 8 GB on "GTX 1180", then it will probably be enough. Memory is still expensive, and GDDR6 is slightly more expensive than GDDR5/X.

dont take it that serious, im quite happy with my old 980, im playing a bit less and focused on other projects, thanks for the advise kid,

Regards,

 

[Blueberries]

2.5Ghz is definitely achievable... with LN2.

 

[[XC] Oj101]

Specs are wrong on the card, next...

 

[Vayra86]

I don't know why people remember the 780 to be much faster than it really was. A 970 was faster than it and the 980 was all around a bit faster than a 780ti even TPU's database shows that.

View attachment 104364View attachment 104365



Call it a rumor all you want but the gap between Kepler and Maxwell did nothing but widen over the years , 970s are now a touch faster than 780tis. The myth was that Nvidia is gimping performance through drivers on older GPUs which isn't true but they are certainly not paying as much attention to these older GPUs neither.

The gap did widen you're right but that is mostly attributable to VRAM and changes in game/engine design to cater to the current console crop. It has nothing to do with Nvidia or drivers and everything with cards that run into bottlenecks they didn't touch before in newer benchmark suites and games.

Maxwell's 970/980 offered both delta compression (higher efficiency in use of bandwidth) and additional VRAM over the 'late Kepler' cards. The 7970 suffers a similar fate. Another important aspect is increased use of tesselation over time. Maxwell features improved tesselation over Kepler as well.

Contrary to popular belief, cards never 'age well', but cards at a VRAM limit will not last as long as others without it.

And if you're wondering why TPUs relative performance database shows 'no change' across the years: check this out: we didn't use 4K yet at the time and the games tested are different:

https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_780/26.html

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

 

[Vya Domus]

Contrary to popular belief, cards never 'age well', but cards at a VRAM limit will not last as long as others without it.

The difference in VRAM between those cards ranges between 0.5 and 1 GB , that's not a huge and you can't blame the performance differential just on that. Many games still barley even touch the 3 GB mark at 1080p , yet like I said a 970 now egedes out a 780ti. With or without things such as tessellation , hell tessellation isn't even used much anyway these days.

Some cards do age well , or rather it's that some age poorly. Kepler was a radical redesign with many deficiencies at the hardware level which were mitigated through software , software which is no longer maintained as well through drivers updates. It's no mistake that "game ready drivers" became a thing around that time.

and changes in game/engine design to cater to the current console crop.

Exactly , but these happen all the time and it's taken cares of through driver updates. There are API tools to inform the driver what engine is used by the application, why do you think such feature exists ? So that specific optimization may be applied.

 

[Vayra86]

The difference in VRAM between those cards ranges between 0.5 and 1 GB , that's not a huge and you can't blame the performance differential just on that. Many games still barley even touch the 3 GB mark at 1080p , yet like I said a 970 now egedes out a 780ti. With or without things such as tessellation , hell tessellation isn't even used much anyway these days.

Many games do hit the 3GB mark because they want to load as much as possible into VRAM and when they do, its beneficial to frame times. Its part of the reason the 6GB 1060 leaps a bit further ahead than it should according to cuda counts and its also the reason 1050ti ships with 4GB on its super tight bus. We see a radical change in VRAM usage across the past five years, as engines are tailored to streamed content rather than 'loading a map'. That has little to do with 'driver updates' to stay current and everything with a shift in hardware requirements. Average VRAM is growing almost exponentially over the past decade, 4GB in the midrange last gen > 6-8 GB fast becoming the norm now. As opposed to circling around the 1-2GB marker for many years.

Tesselation isn't even used much these days?! Are you blind, sir. Its everywhere.

And the 970 has 4GB it just has 0.5GB that is slower, but it still exists on the same PCB and the driver can still utilize it. So we're not talking about '0.5 GB' and 1GB is quite huge, its a whoppin 25% more. Sure, another bit can be explained with driver optimizations but you can still do just fine in every game without the game ready drivers and just getting the major ones. In fact since late Maxwell- early Pascal most of the 'Game Ready' drivers are bug fixes and fixes to the driver before it.

But, if you want to believe its all about driver TLC, be my guest.

 

[Vya Domus]

Tesselation isn't even used much these days?!

Yep , and when it is present it's used sparingly for a few surfaces in right in front of the player camera , for everything else parallax occlusion mapping is much more prevalent and you probably seen it dozens of times in games and thought it's tessellation. The reality is tessellation is still horrid in terms of performance and it's avoided as much as possible.

Its part of the reason the 6GB 1060 leaps a bit further ahead than it should

Than it should ? Compared to what ?

But, if you want to believe its all about driver TLC, be my guest.

It's not all about drivers or VRAM , that's what you don't understand. It stems from hardware whose fixed functionality doesn't do as well with the variety of new software that's written compared to other , newer and more capable hardware that does not require the same amount of work from the software side.

You think this is as cut and dry as chader counts and RAM capacity , it's not.

 

[Vayra86]

Eh? Let's go back to where we started, because what you're saying now contradicts your earlier statement about 'game ready' drivers altogether, what you're saying now is almost literally what I started off with: the statement that the 780 wás faster than it shows to be today compared to a 970/980 in recent games. As for the 'why': all you do is word it differently.

I said this:
The gap did widen you're right but that is mostly attributable to VRAM and changes in game/engine design to cater to the current console crop. It has nothing to do with Nvidia or drivers and everything with cards that run into bottlenecks they didn't touch before in newer benchmark suites and games.

You say:

It's not all about drivers or VRAM , that's what you don't understand. It stems from hardware whose fixed functionality doesn't do as well with the variety of new software that's written compared to other , newer and more capable hardware that does not require the same amount of work from the software side.

You think this is as cut and dry as chader counts and RAM capacity , it's not.

Its not a problem to agree on something And coming back to the driver TLC post Kepler, I think we can put that to rest as well, you literally confirmed it yourself that other factors are in play.

 

[Vya Domus]

Sure , let's go back when it started.

the statement that the 780 wás faster than it shows to be today compared to a 970/980 in recent games.

You said this though :

The 780 is equal to a 970 always has been, perhaps more recently it will lose out more because it runs into 3 GB VRAM and bandwidth limits.
780ti was 'about' equal to a 980 but the same 3GB VRAM limit applies here too.

So, 780 equal to a 970 and 780ti equal to a 980 which was actually never the case , not back then and not in present day. You attributed the differential in performance only to VRAM , clearly so. And I said it's not just VRAM , it's also the fact that Kepler is no longer a priority of software optimizations.

You also said this :

'Lack of driver optimizations' is another such copypasted rumor that was debunked over and over again.

you literally confirmed it yourself that other factors are in play.

You were the one who claimed there was just one factor namely VRAM , not me , I never once said there aren't other factors in play on the contrary I actually pointed out other things. What I did say though and perhaps I wan't clear enough is essentially that those factors which I mentioned aren't independent : hardware deficiencies => more work required on the software side => lack of software optimizations => a gap in performance

On the other hand VRAM is an independent factor from the above : not enough VRAM => bad performance. This is something that doesn't change over time , you push a GPU over it's VRAM limit it means bad performance back then and now with or without driver support , this is a constant of sorts. And a consistent widening performance differential as time goes cannot be explained by it under every circumstance.

 

[rtwjunkie]

So, 780 equal to a 970 and 780ti equal to a 980 which was actually never the case

Actually, back at release time, my own experience was that the 970 barely kept up with the 780, on two identical systems.