i am speaking purely performance mate
otherwise you concede maxwell was way better than kepler because of its perf/watt gains

This post was edited by DCSS on Sep 21 2017 01:09pm

purely performance too look them up but with better thermals and energy

all you do is bitch about everthing

learn to be happy man

if efficiency and performance along with thermals isnt innovation then explain what is

This post was edited by yupitsmeh on Sep 21 2017 01:12pm

The gains in performance essentially come down to the node shrink, it saved them actually having to do any architectural innovation because they can achieve much higher clocks now with the exact same cores. And save power.

I am happy, I didn't buy this junk.

This post was edited by DCSS on Sep 21 2017 01:13pm

oh hmm
how many cores does the 980ti have vs 1080ti percentage wise?

jeez you sound sour about everything like its owed to you

i'm glad i bought 2 just wish I could use them more though

This post was edited by yupitsmeh on Sep 21 2017 01:24pm

The core count is higher on pascal for obvious reasons. That was accounted for and normalized by AdoredTV when he tested maxwell vs pascal IPC at the same clocks and found they were identical. Of course you have to throttle a pascal card pretty hard to reach the clocks its 28nm template struggles to peak at.

I'm not sour about it, just bored, which for someone like me who is usually enthusiastic about new generations is a big deal. Found myself yawning almost as hard as I did when AMD released the r9 280 (o hello again hd 7950) and most of the r9 3xx series. At least performance is decently better with Pascal even if it brings nothing new to the table, but they could have charged the same amount for them as they did with Maxwell or even less considering chip sizes + what the R&D must have looked like compared to previous gens.

Now I'm cynical and pretty sure Volta is just going to be a complete Pascal rebrand thanks to Vega being ass.

But that's consumers for you, if they can open their mouth wide enough the turd will always expand as necessary to fill it.

The one good thing is that 28nm is finally dead, it was long overdue. That node came into use when I was still in high school ffs. Meanwhile Intel's manic tick tock cycle caused a die shrink every few seconds on the cpu side of things, which was for the best in the end. It is mildly interesting to see what the best arch we had on the 28nm process does if you merely shrink it to 14(ish, lolfinfet), gives us a solid reference point moving forward to whatever comes next, maybe even acts as insurance against them half-assing it, could you imagine if they made an arch weaker than Maxwell to introduce the 14nm node but thanks to the superior process the cards still pulled ahead by 10-15%? That would've been frustrating, tho ppl would still buy them like hot cakes.

This post was edited by DCSS on Sep 21 2017 02:03pm

I dont care how that get the gains as long as its not intels typical 7% if that
thats why i'm still on 4770k
i'm sick of them not going to graphene or something else already,
sad thing ibm has showed us 5nm I believe and 7nm before but we arent seeing those

one of the biggest issues with CPU's is that core counts essentially don't change as software isn't demanding more of them, it's hard to see gains when you can't change the number of cores, and are just trying to make them 'better' instead. Not an issue with graphical workloads. However, if all nvidia gpu's basically had 200ish cuda cores like they did in the Tesla days right up til now with Maxscal.. perf gains would have been pretty small. Parallel processing is superior in every conceivable way to trying to beef up individual cores, but software wasn't and still isn't written with that in mind, at least not anywhere near the scale that GPU's utilize it. Ryzen was a step in the right direction but I think it won't be til post-x86 that we really start to see CPU's push the envelope and even compete with GPU's in throughput. At the end of the day, it's a dreadfully inefficient and antiquated instruction set that demands far more power than a GPU to do the same amount of math.

ARM when?

im guessing dcss is rocking a 980ti?

Dude 4 of them, why do you think im so SOUR SALTY over pascal? >2x sli is just one of the many features they removed with this 'generational leap' as they call it. Well i'm no expert on computer hardware but it seems to me you can't have less features and call it an upgrade. God they tried to pull this by dropping the hardware scheduler from Fermi too because it was 'too hot' but all they did was end up with a painfully shallow command queue that ultimately ruined any chance they had to be fully dx12 compliant.

Screw AMD Nvidia, I will be buying a Matrox card next.

think amd makes matrox