980 Ti vs 1080 - WHAT???

[D.P.]

A lot also depends on the games. the 180 is about 25% faster than the 980ti in older games but in newer games, especially DX12 is is more like 40-45% faster. There is then the overclock lottery. People often compare the mature GPUS of the previous generation with the new cores of the next. Mature GPUs always clock much better.

there will also be some good driver enhancements to come with Pascal to optimize for the new architecture.

 

[IvanDobskey]

The 1080 just isn't a true high end part.

 

[geordiejc]

Just played Fallout 4 for 5 minutes on ultra its rock steady at 2067mhz core. here's the data:

 

[Caracus2k]

The 1080 just isn't a true high end part.

Stop baiting/being silly, the only single GPU to card currently able to beat a 1080 is a Pascal Titan X

 

[Five.Stars]

And an overclcoked 1080 is approx 20% faster than an oc 980ti again.........

Not seen what hes getting his 1080 overclocked to. It needs to be 2050 to 2100 to match the 1500 of the 980ti though.

True, but a 980Ti at 1526 is higher % overclock that the 1962Mhz (only going by clocks reported in OP's pic) on the 1080. So, potentially that 20% becomes closer to 10% at those clocks. Still think 8.6% on the graphics score is a bit low though.

 

[geordiejc]

Its running exactly as it should...

https://www.techpowerup.com/gpudb/2839/geforce-gtx-1080

GDDR5 uses two separate differential clocks (unlike its precursor DDR3 which uses only a single clock).

The first clock is used for command and address operations. The second clock is for reads and writes. The second clock is twice the frequency of the first. Data on the IO bus is transferred on the rising and falling edge of the read/write clock.

The GTX 1080 has a memory transfer rate of 10 gigabits per second. Multiply this by the width of the memory bus which is 256 to get the bandwidth in gigabits per second, and divide it by 8 to get the bandwidth in gigabytes per second which is 320 gigabytes per second.

The 1251 Mhz is the command operations clock the read/write clock is double this speed and is 'quad pumped' with some trickery involving using the rising and falling cycles of the clock...

so 1251*2 = 2502 and quadrupled again gives you your 10 Ghz or so effective memory speed clock (with one bit being transferred on every clock cycle to give a 10 gigabit per second data transfer rate)

Thanks for that, that solves the memory speed difference reported in 3dmark.

 

[Strife212]

Marketing hype, it's obviously a better card but it's barely better.

 

[Rroff]

Stop baiting/being silly, the only single GPU to card currently able to beat a 1080 is a Pascal Titan X

Still doesn't make it a true high end part.

 

[Ste]

Takes mine 20 mins at 100% for the core clock to stabilise

 

[geordiejc]

i've just been playing overwatch 4k/epic settings and it stabilised at 2050mhz

still going to buy a 1080 ti when it comes out.

 

[IvanDobskey]

Stop baiting/being silly, the only single GPU to card currently able to beat a 1080 is a Pascal Titan X

It's a replacement for the 980, which it is a good deal faster than.

There is no replacement for the 980ti as of yet.

I'm not sure why you'd buy a 1080 to replace a 980ti and expect a huge jump in performance.

 

[Caracus2k]

Still doesn't make it a true high end part.

What exactly does make a 'high end part'? Is it performance respective to its contemporary peers? Is it the die size or memory bus width? Or are you actually saying its not high end - in the expected eventually available Pascal line-up including cards as yet unreleased and in announced?

Pascal's only likely (based on past recent history) to get one more card a '1080Ti' and possibly a Pascal Titan 'X' 'Black' or something similar with the full die enabled unlike the current Pascal Titan X.

Perhaps it comes down to semantics but the Titan X (and if released the Titan X 'Black') are very high end with the 1080Ti (if released) being next then the 1080.

Ultimately surely the best test of what is high end at any one time is performance... what else matters? You cant play memory bus size or die size now can you? It also a rather arbitrary distinction to talk about performance only with regards to the 'Pascal' branded GPU's because guess what shortly after your beloved 'high end' Pascals are all out a 'mid range' (in your eyes) but yet faster Volta card will most likely be out.

I have already explained why a pricing structure based on your definition of 'mid rage' would be a complete non starter for Nvidia or anyone else here.

It's a replacement for the 980, which it is a good deal faster than.

There is no replacement for the 980ti as of yet.

I'm not sure why you'd buy a 1080 to replace a 980ti and expect a huge jump in performance.

Its the next high performance card released by Nvidia after the 980ti which it beats comprehensively stock vs stock and oc vs oc across the board the fact that its branded an 'XX'80 card is all rather arbitrary and does not necessarily indicate anything other than the marketing departments current whims.

The single GPU fastest top end NVIDIA cards for each generation used to be a 'X'800GTX or an 'X'80 (i.e. 7800, 8800GTX/ 8800GTX Ultra, 280GTX 480GTX, 580GTX etc) so in that sense the 1080GTX would be the very best single GPU card for Pascal right? No its not because the marketing people bought in the 'Ti' and 'Titan' branding

I'm not suggesting that a 1080 is a good replacement for a 980ti. Why on earth would you think that given the relatively small incremental gains from one top end card to the next that it would represent a 'good' upgrade? It is an upgrade none the less across the board and so until the Pascal Titan X came out was most definitely the 'top end' performance wise.

And judging by the general forums and MM lots of 980ti owners DID think the 1080GTX was a worthwhile upgrade over the 980ti just like when I swapped out a 'high end' Titan 'mk1' for a 'mid range' 980GTX

 

[melmac]

I have already explained why a pricing structure based on your definition of 'mid rage' would be a complete non starter for Nvidia or anyone else here.

Gotta love Nvidia's marketing. Don't tell me you believe that R&D line about Pascal costing billions in research? It's plainly just Maxwell spiced up and with Volta coming in early 2017 according to the latest rumours, Pascal is just a stop gap. Look at how quick they are releasing the Pascal parts, getting them all out before more solid info about Volta appears and cuts into their sales.

Rroff is right, but, nobody wants to hear about their £600 cards been mid range.

 

[tyler_jrb]

Something a bit wrong Max OC to Max OC on my 1080 vs 980Ti was 20-30% about 1501 vs 2101. Not a huge jump no. Not up to the hype anyway.

TXP is a good 30% faster still.

 

[titaniumx3]

That 1080 score is fine, I get a similar score with my 1080.

Also the 980ti is clocked quite heavily. Both cards clocked heavily will be very close in performance.

 

[NirK]

It's a replacement for the 980, which it is a good deal faster than.

There is no replacement for the 980ti as of yet.

I'm not sure why you'd buy a 1080 to replace a 980ti and expect a huge jump in performance.

Its about 25% faster or more in newer games, this is a pretty big difference... consider that people are getting a TX at 1100GBP with a reference cooler (worse than all custom coolers) which is only 25-30% faster than a 1080... so upgrading from a 980ti > 1080 is relatively good upgrade.

980ti vs 1080 in timespy for example... massive increase in score with 1080... but it seems that in firestrike the 1080 does not score as well, compared to gaming or timespy etc.

 

[RanxZy]

And an overclcoked 1080 is approx 20% faster than an oc 980ti again.........

It's not, the 1080 gets less gains per extra MHz than the 980Ti. Works out less than 20% faster with a 980Ti at 1455MHz and 1080 with 2050MHz. Both of those clocks are not impossible to sustain with superior cooling from aftermarket cards either.

http://www.overclockersclub.com/reviews/nvidia_geforcegtx_1070_overclocking/2.htm

 

[NirK]

It's not, the 1080 gets less gains per extra MHz than the 980Ti. Works out less than 20% faster with a 980Ti at 1455MHz and 1080 with 2050MHz. Both of those clocks are not impossible to sustain with superior cooling from aftermarket cards either.

http://www.overclockersclub.com/reviews/nvidia_geforcegtx_1070_overclocking/2.htm

Its funny how people just make up results depending on what card they have... "less than 20%" is not correct. You have a maxwell card so it is "less than 20%", if you had a 1080 it would be 25-30%. IIRC the results from the review you linked are 23% faster overall for the 1080 if you add them all up.

 

[RanxZy]

IIRC the results from the review you linked are 23% faster overall for the 1080 if you add them all up.

Because you have a Pascal card

 

[Rroff]

What exactly does make a 'high end part'? Is it performance respective to its contemporary peers? Is it the die size or memory bus width? Or are you actually saying its not high end - in the expected eventually available Pascal line-up including cards as yet unreleased and in announced?

Pascal's only likely (based on past recent history) to get one more card a '1080Ti' and possibly a Pascal Titan 'X' 'Black' or something similar with the full die enabled unlike the current Pascal Titan X.

Perhaps it comes down to semantics but the Titan X (and if released the Titan X 'Black') are very high end with the 1080Ti (if released) being next then the 1080.

Ultimately surely the best test of what is high end at any one time is performance... what else matters? You cant play memory bus size or die size now can you? It also a rather arbitrary distinction to talk about performance only with regards to the 'Pascal' branded GPU's because guess what shortly after your beloved 'high end' Pascals are all out a 'mid range' (in your eyes) but yet faster Volta card will most likely be out.

I have already explained why a pricing structure based on your definition of 'mid rage' would be a complete non starter for Nvidia or anyone else here.

My metric for high end is pushing the limits of what is possible - between using barely half its available thermo/electrical budget, GP100 showing much bigger cores are posible and that aside from the VRAM modules most of the parts on the PCB are more commonly found on mid range GPUs I struggle to take it seriously as "high end".

That pricing structure you talk down is the way it has always been done upto Kepler and even to an extent with Kepler and Maxwell it is only really with Pascal that anything has changed (no one was fooled into thinking the GTX980 was "high end") and at the end of the day people tend to buy GPUs based on a compromise of what they need and what they can afford - if you trickle out updates by and large people will just wait until they can see a significant increase in performance for their money anyhow so you still end up with people skipping a generation or buying a cheaper card to get the performance they want by waiting one way or another. (One aspect you are skipping over here is that often a new generation of hardware especially on a new node comes hand in hand with newer hardware feature support which reduces the ability for people to stay on older hardware and do a jump from mid-range to mid-range instead of spending out for high end as they will eventually be forced to buy sooner rather than later when they need hardware that supports the latest version of DX, etc. to get the best out of games).