Nvidia GeForce GTX 680 £410
22nd Mar 2012 | 13:01
The top-end Kepler GPU is here for you, Mr Gamer
It may have been a long while coming but the Nvidia GeForce GTX 680 is on the way to balance up the next-generation graphics market.
AMD launched its Radeon HD 7970 in December (well, the cards were nominally available to buy in January actually) so it's a bit of gap that Nvidia has to make up with its latest top-end GPU.
So has it used the intervening time wisely, or has Nvidia missed its chance?
Well, actually this could easily have been one of those simple single sentence reviews.
The Nvidia GeForce GTX 680 is its latest £400-odd, top-of-the-line card and is now the fastest graphics card in the world.
Shut down the PSU, turn off the monitors and sit back with your pipe and slippers knowing that it was an easy job, done adequately.
We all kind of knew that would be the case, after all Nvidia has played the waiting game with AMD, letting the competition draw first and release its entire slew of HD 7000 graphics cards.
That meant Nvidia could see how the competition was performing and ensure its engineers finalised the GTX 680 specs and set the clocks to ensure the requisite 10% performance improvement.
If Nvidia had released the GTX 680 performing generally slower than the AMD Radeon HD 7970 then it would have had to be a hell of a lot cheaper and you can bet AMD's Austin HQ would've started looking like a frat house with non-stop high-fives and fist-bumping.
From this opening you could surmise that, because of the prior expectations we had of it, we were pretty non-plussed about this new card.
As it turns out this Kepler-based Nvidia GeForce GTX 680 is far more than just another big, power-hungry graphics card, relying on pure grunt alone to give it the edge.
This is actually a far more elegant card than people might give it credit for.
You could then say we're rather plussed about it.
High above the clean, sun-baked streets of San Francisco, in a hotel bar - which bears more than a passing resemblance to the Emperor's throne-room on the unfinished DeathStar – TR finds itself talking to a VP at Nvidia who shall remain nameless.
When challenged that Kepler was essentially just a die-shrink of the Fermi architecture his reaction was not one we expected.
The amount of alcohol imbibed may have had something to do with the vehemence with which he countered the argument, suffice to say he referred to the previous generation in less than glowing terms.
Terms which probably shouldn't find their way into a site so innocent or polite as TechRadar.
He's right though; while there are definite similarities to the last Nvidia graphics technology Kepler is still a different beast to Fermi, and is much more than just the actual hardware too.
But lets cover that die-shrink first.
Like AMD, Nvidia has taken the plunge and gone for a 28nm production process on this generation of its graphics cards. That means compared with the GTX 580 this new GTX 680 can manage a far smaller die size and still cram in more transistors than it ever has before.
We're talking about the GTX 680's GK104 GPU as a 295mm2 chip compared with the 520mm2 GF110 in the GTX 580.
That's a smaller chip than the GTX 560 Ti with 500 million transistors more than the GTX 580.
It's also a far smaller chip than AMD's 28nm Tahiti GPU at 352mm2, though AMD is claiming almost an extra billion transistors in its top-end chip.
From those simple figures you could infer that the Nvidia architecture is a lot less power-hungry than either its predecessor or the competition, and you'd be right.
The GTX 680 is actually a sub-200w part, operating at its base clock of 1,006MHz at 195w under full load, while the GTX 580 and HD 7970 are 244w and 230w respectively.
So many shaders...
All impressive numbers on the surface but what's actually going on inside, and why are we now talking about a GPU's base clock like it was a CPU?
Despite the ever-bombastic Randy Pitchford of Gearbox referring to it as 'a simulation CPU', because of the advanced PhysX and Apex effects, this is still very much a gamer's GPU.
But Nvidia has taken more than a leaf out of Intel's book; more of that later.
First let's take a look at the actual make up of the GK104 GPU. Like Fermi, the Kepler GPU is made up of multiple CUDA cores jammed into multiple streaming microprocessors (SMs).
These SMs act like simple processors, each concurrently taking on an action, making for impressive parallelism; a cornerstone of GPGPU computing.
But these are no longer called plain ol' SMs, oh no, they are now called SMX, which by today's typical nomenclature probably stands for Streaming Microprocessor Xtreme…
But compared with the old SMs of the Fermi days they could easily be deemed eXtreme. Previously each contained a total of 32 CUDA cores, in Kepler that figure stands at a whopping 192.
Even with half the SM blocks of the GTX 580 you're still looking at 1,536 CUDA cores/shaders spread out over 8 SMXs.
Nvidia is claiming a 2x improvement in terms of performance/watt compared with the GTX 580. That figure seems a little conservative considering the GTX 680 comes with three times the CUDA cores, but while they are physically identical to the cores in the Fermi architecture they are clocked much slower.
In fact they are half as fast because Nvidia has opted not to have a separate shader clock (historically set twice as fast as the GPU clock) instead we have one solitary base clock covering everything.
Boostin' makes us feel good
And there it is again, that base clock.
For every Kepler-based graphics card we are now going to be quoted two separate frequencies, one is the base clock and the second is the Boost clock.
Nvidia has been totally honest and admitted that it copied the idea from Intel, that it is "standing on the shoulders of a great company", as Drew Henry, General Manager of Desktop GeForce puts it.
So we have Turbo Boost for GPUs, the snappily-titled GPU Boost. Asus may have already tried to coin the term on its motherboards, but we doubt there'll be much of a fight over the name.
In previous GPU generations the final clockspeed was determined by the worst-case scenario of application power usage – typically taking the most power hungry app around and setting the clockspeed to match that power draw.
But the draw on the GPU varies massively between different programs, just taking Fermi as an example, power for apps could vary by as much as 50%.
So on lower powered apps there's a lot of GPU headroom not being utilised.
GPU Boost then analyses the amount of power an application is using and boosts the GPU frequency with the amount of extra headroom it then has at its disposal.
It's also completely application independent – you wont need a new Kepler profile when a game gets released so that GPU Boost can take advantage.
It's all based on measurements coming directly from the GPU's sensors in real-time.
Kepler can dynamically alter its clock and voltage every 100ms, essentially every few frames the GPU has a decision to make on clocks and voltage settings.
And as this is just the first generation of GPU Boost you can expect that to be done quicker over time.
This auto-overclocking though doesn't mean an end for traditional overclocking.
"GPU Boost is like overclocking's little buddy… our GTX 680 is a monster overclocker," says Tom Petersen, Nvidia's director of technical marketing.
Each of Nvidia's partner card manufacturers has been given all the API goodness with which to prepare their own overclocking apps, though we had a look at EVGA's Precision tool.
There are two sliders, one for altering the base clock and one for allowing the GPU to use extra power – up to just over 260w. And GPU Boost will still work when you've slammed both those sliders to max too.
Just as interesting is the Frame Rate Target.
You can set the software to target a particular frame rate – this is really only used to set a target, not to force the GPU to overclock to get there.
But if you give the GPU the frequency headroom and the capability to draw as much power as it can then it will push the clocks up to ensure it hits your specific frame rate.
This is arguably more interesting at the mid-range than the high-end. The prospect of being able to tell your £200 card to hit 30FPS in all games will give even the least savvy PC gamer console-like performance with minimal effort.
It's a tantalising prospect and shows that Nvidia is really focusing this release on the gamer, not the GPU compute all-rounder that Fermi had been touted as.
Compared to the performance of the HD 7970 the speed boost isn't huge, but on the whole the GTX 680 has the edge.
Looking at the Crysis 2 bench though the AMD card has the lead, but that is still within benchmark variance.
In overclocked trim, via the EVGA Precision tool, performance does step up a decent chunk, but the SLI performance is a little less impressive.
That's likely down to early drivers but we didn't see the longed for 2x speed boost we've sometimes seen from SLI combinations.
DirectX 11 tessellation performance
DirectX 11 gaming performance
All this talk about the Nvidia GeForce GTX 680 Kepler architecture is all well and good, but how does it actually perform?
As we've said there are two quoted clockspeeds.
The Base clock is the lowest clock speed of your GPU and the Boost clock is the typical clockspeed you'll see most of the time.
Except it really isn't.
The vapour-chamber cooling on the GTX 680 is good enough to ensure that most of the time you'll see considerably higher clocks than the rather conservative 1,058MHz it claims on the box.
In general testing we saw our GTX 680 hitting around the 1,100MHz mark in most games.
That extra boost allows the Nvidia GeForce GTX 680 to mostly outperform the competing AMD Radeon HD 7970.
There are a couple of exceptions, the HD 7970 gains a couple of extra FPS in Metro, and in Crysis 2 it is slightly ahead at the top, 2560x1600 resolution, but loses out again at 1080P.
Nvidia claims the extra performance of its 6Gbps GDDR3 memory gives it enough bandwidth to compete with AMD, but we still have a little niggling worry about the 256-bit memory bus at these rarefied resolutions.
In Heaven 2.5 and Batman: Arkham City however you're looking at around a 10-15% improvement over the competing AMD card.
Things get even messier for AMD when you take a look at the synthetic tessellation benchmark, Tessmark; that has the GTX 680 around twice as fast at tessellated OpenGL geometry.
While Nvidia has resolutely stuck to its gaming guns with the GTX 680, refusing point-blank to discuss compute performance, the Kepler GPU is still very much capable of those GPGPU shenanigans.
Even using the non-tweaked version of MediaEspresso the 1080P encoding of a short piece of HD footage took just 13 seconds compared with 44 seconds on the HD 7970.
So, the Nvidia GeForce GTX 680 is a pretty quick card then?
Yes, and it also bodes well for the cards that are yet to come.
The EVGA precision tool that Nvidia provided for benchmarking allowed for some serious overclocks (with GPU Boost giving even more on top of that), with most games running just under the 1,300MHz mark.
True the AMD cards, Tahiti in particular, have a huge amount of overclocking headroom, but you have to invalidate your warranty to get there.
The GTX 680 will give you automatic access to those higher clocks without any such worry.
You can also let the GPU draw up to 32% more power, which in turn means the GPU Boost will ensure even higher regular clocks.
But for the mid-range cards it's the Frame Rate Target that interests us most.
The average PC gamer may not necessarily want to overclock their new investment, but by letting the GPU decide when it needs to up the clocks and voltage - within its capabilities - then telling it to just try and hit 30FPS will make getting great performance simple.
The fact the GPU Boost is so temperature dependent means that aftermarket coolers, specifically water-cooled units, will deliver much higher performance.
That does though mean our Scandinavian cousins will have faster cards than us on average, and you're also likely to drop frame rates in the Summer.
But this is Nvidia back to its PC gamer-supporting best.
This is a card for gamers, this is a card that's built to be the best right now and built for the next generation of games. If Epic's Samaritan demo is anything like what Unreal Engine 4 can give us, the GTX 680 is well-placed for the future.
It's great to see the Nvidia GeForce GTX 680 coming in with relatively low power-draw. Eschewing the usual 8-pin/6-pin PCIe power connector combo is good news for your electricity bill.
Maybe not for the PSU manufacturers wanting to push their 1,500W beasts though.
It's fast enough too – hitting the 1GHz GPU clock that seems to be the 28nm norm. Nvidia has obviously used the time to engineer the GTX 680 up to a point where it can just about stay ahead of the AMD Radeon HD 7970.
We have to say we're also rather enamoured with the GPU Boost technology.
AMD's HD 7000 series cards may have serious overclocking headroom, but Nvidia's GTX 680 gives you access to it without necessarily needing you to mess around with your expensive new hardware yourself.
There are other impressive bits of surrounding tech too – like the Frame Rate Target and Adaptive VSync, a method of dynamically turning VSync on and off to avoid both tearing and the stutter associated with dropping frame rates with standard VSync on.
There's a niggling concern at the back of our collective minds about the fact it is such a low-powered card.
Sure it's quick enough to be called the fastest GPU of this generation, but a full-fat, 250w version, with a chunkier GPU housing more SMX modules must surely have been created in the labs.
That maybe should have been Nvidia's top-end, £400 card.
This relatively small GPU, with its lower memory-bus and power draw seems more like a GTX 670.
So essentially that makes us a little narked about the price.
It's £400 because that's the standard for top-end GPUs now, despite the fact that it probably is more of a mid-range card compared to the performance the technology could realistically manage.
It's £400 because Nvidia can charge that much, and because AMD's Radeon HD 7970 hasn't pushed it to create something even more spectacular.
That shouldn't take away from the fact that we've got a card in front of us that's cool, quiet and less power-hungry than Nvidia's normal top-end GPUs.
It's not the power-crazed GPU behemoths we're used to from Nvidia, but it's still got the performance chops and some neat extra tricks.