Nvidia GeForce GTX 760 Review Roundup

An anonymous reader writes "Earlier today, Nvidia released its latest graphics card: the Geforce GTX 760. A followup to last month's GTX 770 launch, the new GTX 760 is the fourth 700-series card since the company launched the GTX Titan back in February. Sporting 1,152 CUDA cores, 96 TMUs, 32 ROPS, a 256-bit memory interface that effectively runs at 6 GHz, a base clock of 980 MHz, and a Boost speed of up to 1,033 MHz, the newly-minted GTX 760 is offered at a price point of $250. Benchmark results are available from all the usual suspects: AnandTech, HotHardware, PC Magazine, PCPer, and Tom's Hardware. To make a long story short, Nvidia's new card edges out AMD's equally-priced Radeon HD 7950 Boost Edition, and even goes toe-to-toe with the $300 Radeon HD 7870 GHz Edition. Factoring out AMD's incredible Never Settle game bundles, and looking purely at performance, the GTX 670 allows Nvidia to cinch up the mainstream gaming price point." Reader crookedvulture adds, "The $250 card is an updated spin on an existing GPU, so it doesn't raise the bar dramatically. In fact, the GTX 760 achieves rough performance parity with the Radeon HD 7950 Boost, which costs just a little bit more. The situation is similar at around $400, where the contest between the GeForce GTX 770 and Radeon HD 7970 GHz Edition is a toss-up overall. These price/performance scatter plots paint the picture clearly. AMD has largely resolved its previous frame latency issues with new drivers, making the battle between GeForce and Radeon more about extras than performance. Nvidia offers software to optimize game settings and record gameplay sessions, while AMD includes download codes for recent games. You really can't go wrong either way."

Comments

tl;dr:

[Guspaz]

It's the GeForce GTX 670 for a hundred bucks cheaper. That's what all the reviews boil down to.

Re:tl;dr:

[amiga3D]

I can't even look at AMD cards. Thanks to their drivers it's Nvidia only for me and my linux tower. Since Nvidia came out with their linux drivers years ago it's been one Nvidia card after another in every machine. AMD doesn't care of course since Linux is such a tiny slice of the customer pie which means when I build a windows box for friends I slap an Nvidia card in there too out of sheer resentment.

Re:tl;dr:

[Hadlock]

Now that AMD powers all three consoles, it's likely that windows and BSD drivers will improve. Recompiling a BSD ps4 game for the Linux stream console probably will be less difficult than it has been in the past.

I'll still be buying NVidia this round as well.

Re:tl;dr:

[Guspaz]

AMD powered two of the three consoles this past generation (Wii and 360), the drivers are still terrible (getting Enduro drivers working on Windows is a nightmare). Why would adding one more console change anything?

Re:tl;dr:

[Guspaz]

It has same performance as the GTX 670, at the same TDP as a GTX 670, with the same board/shape/etc as the GTX 670, at a price that is $100 cheaper. It doesn't matter what it's based on.

Re:tl;dr:

[adolf]

The Wii and the 360 weren't x86.

Re:tl;dr:

[ikaruga]

Unfortunately that is not how things work. Unless Sony decide to open-source, or at least, release the binary for someone to reverse engineer them, having the PS4 using BSD+AMD combo will have no impact on the open development scene at all. It's going contribute to the BSD driver library as much as OSX, which is also based of BSD, did, i.e. nothing.
Either that or someone jailbreaks the PS4, dumps all it's system files and starts reversing.

Bloat

[causality]

Nvidia offers software to optimize game settings and record gameplay sessions

Did anyone else read that and think, "this does not belong in a device driver"?

Maybe it's a great idea that many people will use and there is no other possible way to accomplish this task in userspace. I'm open to that idea, but right now I just don't see the merit.

Re:Bloat

[Hadlock]

Perhaps it's a rebranded copy of FRAPS, out whatever third part equivalent? If it is driver level, it'll be nice to see a standard way to output video for services like ustream or corporate users like webex out gotomeeting instead of some goofy Java setup through the browser.

Re:Bloat

[blankinthefill]

Alternately, for those that do use it, it seems like a really good way to cut OUT a lot of bloat. By taking out the middle man, I would think that you would be able to get better quality for cheaper than if you used a 3rd party program like fraps. I could be wrong, but considering these cards tend to be targeted towards gamers, and gamers like to make videos of them playing games, it seems like there may be something to this.

Re:Bloat

[epyT-R]

The best way to make it easy to capture video from a game is for the developers to add demo record/playback capabilities in their titles. Many games already support this. In windows, they can be run (with a command to load the demo) through something like kkapture which wraps the timing, gfx, and sound apis and captures the output to a codec of your choice (I use lagarith). From there, edit the file as you please. The framerate can be set, and as the game thinks it's always time to render the next frame, it runs as fast as your leftover CPU cycles can encode and you lose no frames. It's a perfect capture.

kkapture is a free opensource utility written by ryg, a member of the demo group farbrausch. He originally wrote it to capture the output from win32 demos, but it works for many games too.
http://www.farb-rausch.de/~fg/kkapture/

Re:Bloat

[Nemyst]

It's good because it uses the GPU's internal H.264 encoding hardware to record seamlessly at no cost to framerate. Fraps and other screen recorders are known to often halve frame rates or more. By doing the recording on the GPU itself, you can extract the framebuffer much more efficiently, transcode it straight in the GPU without much (if any) involvement by the CPU, and save the much smaller file to the hard disk at the very end, thus avoiding the use of the comparatively slow disk to store the very large non-compressed buffers.

Prior to that, the only way of recording the screen efficiently was to use HDMI recorders which would just take the entire output and transcode that externally, which is far less practical and much more expensive.

Re:Bloat

[TheLink]

I don't see anywhere in the summary (who reads the articles anyway? ;) ) that implies it would be driver level. So I'd just assume they aren't that crazy.

I use MSI's Afterburner to record video from games and it works well enough for me (price seems right too): http://event.msi.com/vga/afterburner/download.htm
The output file ends up quite big and I need to use virtualdub or vlc to compress it. But that's because my PC isn't powerful enough to do on-the-fly H264 video compression of the capture at 60fps.

Needed for streaming to Shield console anyway

[Blaskowicz]

Such a feature was needed anyway and already exists. It's also the stuff behind the "Geforce GRID" (racked servers with GPUs that stream 3D accelerated software, like CAD, GIS, visualization or games to multiple thin clients, compatible with desktop virtualization, with choice between a "geforce" or "quadro" software configuration for each user)

Kepler GPUs include a H264 encoder used for that streaming purpose. Dumping to disk is a simple and interesting option, which I hadn't thought about but is obviously useful as well.

Gotta love Nvidia fans

[tyrione]

Laced with broad strokes of bs like any other zealot they come out in droves to defend their brand. Nvidia CUDA is a dead end future. The LLVM/Clang driver Target for the R600/future AMD set up will make AMD Radeon/FirePro solutions on Linux/FreeBSD rock solid in the next 6 months. Nvidia continues to ignore reality: OpenGL/OpenCL are married together. APU designs are the future and having a crappy OpenCL presence far behind AMD is the reason Apple dumped Nvidia from now on.

Gotta love Apple fans

[dutchwhizzman]

As if Apple won't go back to NVidia when they have a more competitive offering. They have been switching back and forth several times since the radeon 7000 was a hot video chip.