facebook rss twitter

Review: GeForce GTX 680 vs. Radeon HD 7970 at same clocks

by Tarinder Sandhu on 30 March 2012, 09:21

Tags: NVIDIA (NASDAQ:NVDA), AMD (NYSE:AMD)

Quick Link: HEXUS.net/qabekz

Add to My Vault: x

Overall performance, power-draw, and closing thoughts

Overall normalised performance

Taken over our six games and calculated by adding up the percentages and then comparing them against one another, based on an equal weighting in each title, here's what we end up with.

Normalised performance at 19x12 GTX 680 HD 7970 @ 680 speeds HD 7970
100% 94.2% 86.6%

The better-than-GeForce showing in Aliens vs. Predator helps drive up the comparative performance of the HD 7970s. But even with the handicap of sub-par numbers in AvP the GeForce GTX 680 is still the fastest single-GPU card out there.

Normalised performance at 25x16 GTX 680 HD 7970 @ 680 speeds HD 7970
100% 96.8% 88.4%

Clawing back a per cent or two once the higher GPU load of the 2,560x1,600 settings begins to bite, the same-clocked HD 7970 almost matches the GTX 680's performance.

Power

For what it's worth, the GeForce card is also slightly more power efficient than the Radeon in either frequency guise.

Implications for the gamer

We deliberately use very high-quality settings in our games that are expressly designed to exact the greatest toll on the GPUs. GeForce GTX 680 and Radeon HD 7970 are both consummate performers at the full-HD, 1,920x1,080 resolution common on many monitors.

Playing through the games shows us that it's actually very difficult to categorically say that one GPU is significantly faster than the other; a Radeon HD 7970's performance feels like a GTX 680's.

And this feeling is reinforced in the crazy-resolution (2,560x1,600) numbers. Such is the load at this setting that even these high-end monsters begin to buckle; we notice the occasional slowdown and lack of silky-smooth rendering.

The GeForce GTX 680 is the fastest single-GPU card around and nothing we've witnessed in this editorial changes our view of that fact. It's also faster, on average, than a same-clocked HD 7970, though it's almost close enough to call it a draw.

GeForce GTX 680 has been lavished with bombastic praise since its arrival last week. Most of this is deserved, because it is a mighty fine high-end GPU. Our numbers and real-world playing also tell us that it would be foolish to dismiss the Radeon HD 7970, for it's also a very good enthusiast-orientated card.

The corollary of this evaluation is startlingly simple: one cannot buy a bad high-end card right now. Yet spending £400 on a single graphics card entails making a hard choice. Knowing what we do and having played game after game and seen more benchmark results than are advisable, our advice would be to opt for the GeForce GTX 680, unless your particular game du jour happens to run better on Radeon hardware.



HEXUS Forums :: 13 Comments

Login with Forum Account

Don't have an account? Register today!
Nice review. This is why I stay at Hexus so much. :)
Fantastic article :) I think AMD might have gone to heavy with the compute power on the top level chips while nvidia have gone the other way. Think this means nvidia will also win the duel crown this time round unless AMD can pull something big out of the bag.
This is quite interesting, as for AMD this is a new arch, whereas Nvidia it is modification of fermi. If you look back at where the 6000 series came from, this new arch from AMD could hold a lot of promise. I'm still amazed at the power draw figures for the Nvidia card, certainly well done by them.
It will be interesting if AMD will be able to boost Tahiti somewhat more,in the next few months. There has some been some vague noise about them raising the default clockspeeds somewhat:

http://forums.anandtech.com/showpost.php?p=33184497&postcount=1

http://www.rage3d.com/board/showthread.php?t=33987782

OTH,the dual GK104 is going to be less TDP constrained than the dual Tahiti card,so should be faster too.
They'd have to raise it over 100MHz to performance competitive alone. I'm not saying this isn't doable, we know they're highly-clockable (under-clocked?) chips and they did 100MHz on the 4890 but it would be a bit ‘meh’ if they clocked them higher and cost more to boot - whilst being slower.