Just a week after their flagship NVIDIA GeForce GTX 780 launch, NVIDIA added the more affordable GTX 770 to their line-up. The green camp have since added the more mainstream GTX 760 to the mix (and we will get to this one soon, promise!). We’re now only just getting around to taking a look at the reference GTX 770 card, but as the developers of Duke Nukem Forever must have said countless times – better late than never.
The GTX 780 is for all intents and purposes, a really well executed part. The only moot point is a strong retail price that takes it off the list of many end users budgets. This I suppose is where the NVIDIA GeForce GTX 770 is meant to come in. For those with GTX 680′s, your upgrade cycle (if looking for more performance) is fixed squarely on the GTX 780, which offers significant, quantifiable gains. You are certainly not the target audience of the GTX 770. The GTX 770 will, however, look enticing to those from a few generations past, perhaps those with a GTX 480/580 or AMD Radeon HD 5870/6970 where the performance gains and price may urge many to consider now as a good time to upgrade.
If this looks familiar, chances are you saw it in reviews from 15 months ago. This is the same architecture reference image sent out with the GTX 680 press pack. The new NVIDIA GeForce GTX 770 uses the same GK104 GPU found in the GTX 680, only repackaged with higher clocks and beefier voltages. We gave an overview of the different GPU’s found in NVIDIA’s high-end cards (flavours of GK104 and GK110) in our GTX 780 review here, but the main points are that the GTX 780 uses a cut down 12 SMX block GK110 GPU – a chip which prior to the GTX 780 launch could only be found in the GeForce GTX Titan (14 SMXes) and NVIDIA’s HPC Tesla line of cards. The GTX 780 was the first true consumer card to use it, and still is. For the GTX 770, NVIDIA could have gone a different route and cut down GK110 even further with less SMX blocks. but Instead they’ve stuck with GK104 and taken it to it’s knees.
|Max Configuration Potential|
So here is the situation. GK110 even to this day still has headroom to expand above and beyond any card it has in the channel (including Titan, which uses 14 SMXes). GK104, on the other hand, is and has been at its peak configuration since the GTX 680 launch 15 months ago. There is nowhere to go with this chip other then core and memory clock increases, and that’s what we have here today.
From both technical and business perspectives, this is a smart move on NVIDIA’s part. They can continue to use this tried and tested part in a re-branded guise that really does fit well within their product stack. Unlike previous and highly controversial re-branding policies at NVIDIA, this one seems perfectly legitimate and mindful. Another good reason for staying with GK104 has a lot to do with how NVIDIA generally like to design their high-end GPU’s. NVIDIA has always had a strategy of producing very large die-sized GPU’s of which tend to be expensive to produce but almost always guarantee them the single fastest GPU on the market (not always, but generally they do). After the mediocrity that was the GTX 480 and 580, NVIDIA changed their approach and thus GK104 — a chip that was only 294 mm2 in size (GTX 580 is 520 mm2) — was born. AMD, on the other hand, has always tried to keep their GPU’s a little leaner (around 320 to 380 mm2) and this normally gives them scope to price their cards aggressively and in turn hurt NVIDIA’s profit margins. Tahiti (352 mm2) found in the HD 7970 follows this predictable trend, only this time it’s bigger than NVIDIA’s flagship, GK104, thus there is no easy way for AMD to hurt NVIDIA. This is likely a significant factor for why prices have stayed quite high this generation and haven’t really come too low as AMd are usually the driving force of the price cuts we routinely see. If this was chess, NVIDIA has AMD in checkmate. At any time NVIDIA has had the ability to come way down in price as part of their strategy, but market conditions never required them to do so. Today is that day. I’m speculating now, but if Tahiti had been significantly bigger than GK104, i suspect NVIDIA would have been real aggressive with its GeForce 600 series pricing just to hurt AMD, but I think they were close enough in size for that decision to leave the table.
It’s quite interesting to note that GPU architectures do take a long time to develop, so some very smart people at NVIDIA and AMD have to try and second guess each other well in advance with future architectures in an effort to surprise or outwit. We won’t know exactly how well AMD has done this time round until they launch their delayed 8000 series, but one thing we can say is that NVIDIA have gone back to producing a huge chip in the high-end but have a small one as their number two. There is plenty of scope to launch counter parts depending on what AMD eventually launch. But only time can tell how this plays out.
|GTX Titan||GTX 780||GTX 770||GTX 680||GTX 580||HD 7970 GHz Ed.|
|Memory Clock||6GHz GDDR5||6GHz GDDR5||7GHZ GDDR5||6GHz GDDR5||4GHz GDDR5||6GHz GDDR5|
|Memory Bus Width||384-bit||384-bit||256-bit||256-bit||384-bit||384-bit|
|Memory Bandwidth||288 GB/s||288GB/s||224 GB/s||192 GB/s||192 GB/s||288 GB/s|
|Die Size (mm^2)||551||551||294||294||520||352|
|Manufacturing Process||TSMC 28nm||TSMC 28nm||TSMC 28nm||TSMC 28nm||TSMC 40nm||TSMC 28nm|
|Launch Price||£830 /$999||£550/ $649||£319/£399||£430/ $499||£430 /$499||£430/£$499|
Looking at the specifications table, you can see just how similar the GTX 770 and 680 are. But still, there are healthy increases to note. The core clock in tandem with GPU Boost 2.0 means not only is the GTX 770 clocked higher than the GTX 680, but the way GPU Boost intervenes has been updated. GPU Boost 2.0 was first introduced on the GTX 780 and is a method used to control TDP and temperature by manipulating the voltage and core clock in real time. GPU Boost 1.0 found on the GTX 680 was designed to prioritise TDP more so than temperature, and had a hard limit at stock of 170W. Boost 2.0 has changed the algorithms used to now favour temperature more than TDP, so has increased this limit to 230W. The end result of these changes means that not only is the GTX 770 clocked higher than a 680, but it will likely reach its Max Boost clock far more regularly than the GTX 680 could ever have hoped to achieve.
Moving on to memory, we see a majestic speed bump from 6GHz on the GTX 680 to a whopping 7GHz. This bumps up memory bandwidth by 16.6% over a GTX 680 and will most definitely differentiate performance. Perhaps the most pleasing change here is not one of increases, but of decrease. Performance is certainly going to be up over a GTX 680, but the price has come way down. It is now squarely facing the AMD HD 7970 GHZ Edition, so those two are the ones we will focus on most today.The Card: NVIDIA GeForce GTX 770