GTX 780 performance revealed (kepler)

Jakus · 18 Dec 2011 at 12:26

Well for me, 10% more performance and 50% less heat and power would be nice but GPU's don't seem to scale as well as cpu's with these die shrinks, maybe as the architectures are already so highly developed in comparison to CPu's

Dauthi · 18 Dec 2011 at 13:02

bru said:
estimates could mean anything from these are just hopefully guesses by Nvidia of the performance gains or the article writer used the word without realising how it could be interpreted or even that he did it deliberately.

Feels like Nvidia just trying to stall a few people from buying AMD cards when released hoping for this estimated*performance. It is just this time they can't say their cards are due very soon.

*in optimal conditions, heavily overclocked, on a good day with the wind helping them J/k.

ALXAndy · 18 Dec 2011 at 13:12

Note they are not pointing out what performance they regard to !

Could be heat, noise, speed, memory bandwidth, tessellation..

It's like AMDs chart that looks similar.. It's for tessellation.

Note all of those things are all DX11 ready.

Dauthi said:
Feels like Nvidia just trying to stall a few people from buying AMD cards when released hoping for this estimated*performance.

It's just a chart to give the fanboys reason to hold on. That's all. Any one who wants a 7970 due to compulsive personality will get one, regardless.

Jakus · 18 Dec 2011 at 13:16

ALXAndy said:
It's just a chart to give the fanboys reason to hold on. That's all. Any one who wants a 7970 due to compulsive personality will get one, regardless.

Exactly, then swap for a nvidia later if their better

ALXAndy · 18 Dec 2011 at 13:22

JAKUS said:
Exactly, then swap for a nvidia later if their better

Eeeyup and lose a fortune in the process.

I confess, I totally live in the past now. I buy from said people with upgraditis.

One man's trash is another man's treasure.

trisp · 18 Dec 2011 at 13:33

Does no one remember the 2 > 4 series jump?

Jakus · 18 Dec 2011 at 13:33

About the same as buying a 6 month old Car really

Jakus · 18 Dec 2011 at 13:35

trisp said:
Does no one remember the 2 > 4 series jump?

No, can You hum it ?

Ryan-3 · 18 Dec 2011 at 13:36

I'm thinking the 600series is to be the new 300series. So, it's not an Nv fail.

Roll on the GTX770!

robfosters · 18 Dec 2011 at 13:50

I'm looking forward to the GTX1000

That should be IMMENSE

ALXAndy · 18 Dec 2011 at 14:15

trisp said:
Does no one remember the 2 > 4 series jump?

I bet the CEO of BFG does. I bet the poor sod will never forget it.

Dangerous_Dave · 18 Dec 2011 at 14:20

I'd say the reason they're going from 5 to 7 is so the lay consumer doesn't think it's an inferior or older generation to AMD's cards which will start with a 7. Unless of course AMD do something really devious and start their cards with an 8 instead...... No, I've decided: just to be on the safe nvidia should skip 6, 7 AND 8 and call their kepler cards the GTX-9## series.

Take THAT amd!

ALXAndy · 18 Dec 2011 at 14:27

Maybe the 6 series will all be mobile and low powered?

That was what they were supposedly doing first, with the high end to follow at the end of 2012.

Just goes to show what happens when you think you have no competition. Talk about drag it out. AMD must have really stirred a hornet's nest in the green camp

trisp · 18 Dec 2011 at 15:17

ALXAndy said:
I bet the CEO of BFG does. I bet the poor sod will never forget it.

I bet he does not!

and no i can not hum it

Scougar · 18 Dec 2011 at 15:28

Twice the performance... sounds suspect to me.

Lightnix · 18 Dec 2011 at 15:29

Agnostic said:
What are the units on the Y-axis?

Frames per second.

DarrenM343 · 18 Dec 2011 at 15:54

JAKUS said:
maybe as the architectures are already so highly developed in comparison to CPu's

I would have thought CPU's are much highly developed than GPU's. Maybe that's why they don't scale so well, the designs are not optimised to take advantage of the die shrink.

Mind you I think if a new product is only a little faster than an old product but saves a chunk of energy (say 20%) that is very good also but ultimately most would prefer more performance for same power usage.

StonedPenguin · 18 Dec 2011 at 16:00

Probably a horrendous question but I've always wondered why GPU designers cant just stick the equiv of say a q9550 in there, Would it be 400% faster?

Sorry for random (probably v.stupid question)

bhavv · 18 Dec 2011 at 16:09

http://www.techpowerup.com/156709/GeForce-Kepler-104-and-100-GPU-Specifications-Compiled.html

3DCenter.org compiled a few specifications of the GK104 and GK100. They go like this:

GK104

640 to 768 CUDA cores
80 to 96 TMUs (depending on what the CUDA core count ends up being)
384-bit GDDR5 memory interface, 48 ROPs
Built on the 28 nm TSMC process
Products based on this will launch in the first quarter of 2012

GK100

1024 CUDA cores
128 TMUs
512-bit GDDR5 memory interface, 64 ROPs

Seems true, GK104 will be anywhere around 25-50% faster than a GTX 580, GK100 will be 100% faster.

This is why I dont buy high end cards, the next midrange is always better, and / or much more efficient.

Unless the price on the Gk104 is <£200 for the full 384 bit and maximum shader version, I'll deffo be skipping. 2 x £180 cards is the maximum I am willing to spend for some lovely Gk104 SLI goodness.

I'd also have to wait until April as my earliest upgrade time, I need my ISA interest.

Jakus · 18 Dec 2011 at 16:41

DarrenM343 said:
I would have thought CPU's are much highly developed than GPU's. Maybe that's why they don't scale so well, the designs are not optimised to take advantage of the die shrink.
.

No the GPU's have way more transistors than CPU's. I'm not the man to ask about architecture comparisons but I would say that = more complex