680m vs 770m at 1920x1080?

I looked around alot about this and it seems that 680m is a bit faster yet the 770m is cooler, but most of these are outdated. Can anyone tell me if with the latest drivers the 680m is still a bit faster? How much faster is it actually?

 

680M is massively faster than 770M. The point about 770M being cooler is moot. If you underclock your 680M to 770M performance levels, it will be just as cool. Stoop to the level of 750M and you can fit one in your optical drive bay (jk ).

On the other hand it is pretty much impossible to overclock the 770M to match the performance of even stock 680M, not to mention one with a healthy 25% overclock.

 

NBC puts the two between 10-20% faster depending on the game. The numbers for the 680m on NBC use an older and slower processor but in general, you shouldn't see a difference of more than 25% when going from the 770m to the 680m.

If the prices (and configurations) for both the machines are the same, the 680m should be a no brainer.

 

Love it! I needed a good laugh. :thumbsup:

I have to agree with the others on this one. GTX 680M is on a different performance level from 770M. A highly-overclocked 770M might get close to a stock 680M, but 680M has quite a bit of overclocking headroom and there's no comparison once you push it to what it's really capable of.

 

The 680M can be overclocked 50-60% in the hands of an experienced person so yes lots of room lol. All that room goes into beating an overclocked 770M.

 

Does the 770m have nothing going in its advantage compared to the 680m? It's newer!

 

So? Newer is not always better, especially when you're comparing products from different tiers. 770M is really a higher-clocked 670MX. Its specs are weaker than 675MX, to say nothing of 680M.

 

This is such an over exaggeration.

 

770M outperforms the 675MX slightly. The 680M is definitely better, but not "massively faster" at all. Let's just say that the setting at which the 770M will have tendencies to dip slightly under 30 fps, the 680M will have no problem keeping 30+.
You can go to the notebookcheck charts and restrict the list to 680M and 770M to see the differences. Results may not always be the most accurate ones, but they're a good guideline nonetheless.

 

680+Ivy Bridge > 770+Haswell, as long as the laptop is plugged in, and i dont think 770M/680M laptops are made for Endurance.

Aye, and watch it run 20% hotter due to less thermal headroom

 

770M OCed can reach a stock 680M and maybe when running hot even outperform it. HOWEVER the 680M as already stated can massively OC. The 770M doesn't stand a chance. It doesn't really even stand a chance against the 7970M if you were to push both to their limits not to mention the 8970M. It stays firmly in 5th place.

1.)780M above the rest but high power requirements...
2.)680M amazingly close to the above when pushed hard
3.)8970M an updated 7970M so maybe slightly higher binned silicon (expect slightly higher OC potential maybe) still below the 680M
4.)7970M good, cheap performer. At release date very powerful and still a good card just no kepler lol
5.)770M (or 675MX with core OC slightly) a highly OCed 670mx. Even a 675MX could destroy it if OCed hard due to having a 256bit bus....

Don't forget 192bit bus is a severe limitation and crippling factor against the 770M at higher resolutions of 1080P and above...all the other cards have 256bit buses.

 

755M SLI



2x 128 bit bus = 256bit bus?

 

The effective memory bandwidth is doubled, so in a manner of speaking, yes it is an aggregate 256-bit bus.

 

That last comment is key. You need a significant vram and/or gpu OC with a 192 bit gpu to make up for the less bandwidth. NBC had to be taken with a grain of salt. Low, medium, high settings are primarily 1024x768 or 720p while ultra is 1080p with all settings maxed. There's no middle ground like 900p or 1080p at high settings. The 256 bit cards will eat the 770m at NBC's ultra though.

Beamed from my G2 Tricorder

 

Double post

 

Well, I am always looking at the Ultra column when checking NBC, the rest is just irrelevant nowadays. 680M and 780M are the only single Nvidia cards that beat it strictly fps-wise. I agree with the 192-bit limitation in other situations.

 

That is incorrect. My 770M SLI have been overclocked slightly above GTX 680M SLI

Other than that I agree, 680M is a better choice because A) it runs faster than 770M stock. B) It can overclock way higher than 770M and C) You can overclock the 680M to 900MHz+ on stock vbios.

There is a cap on how much memory bandwidth you really need. I overclocked the VRAM on the 770M to 1150MHz while running at 1100MHz on the GpU and fired up Valley. Everything was set at Ultra with 8xAA and it capped out at 80% memory bus utilization.
So you dont actually need sickly high memory clocks on the 770M to be able to keep up with 256bit cards in games.

The most important factor seems to be the core clock. That be gaming or benchmarks.

 

You have memory bandwidth to spare because you overclocked VRAM by 15% and you have SLI. Some games and benchmarks are more shader bound than ROP/bandwidth bound and vice-versa, and of course resolution, antialiasing, and other settings have their effects as well. At the end of the day it's about balance. 960 CC with 192-bit bus makes sense for 770M and 1344 CC with 256-bit bus makes sense for 680M but 768 CC with 128-bit bus really limits 765M at higher resolutions even though it has the highest core clock speed of the three. A 192-bit 765M would eat into most of the advantage 770M has over it, and that's exactly what we have with GTX 650 Ti vs. the significantly faster 650 Ti Boost.

 

I don`t think you are gonna hit much higher bandwidth usage than 8xAA, Ultra and 1080p.
SLI doesn`t matter. Each GPU is working on the same amount of data.

On stock, the 192bit memory bandwidth is more than enough for 1080p gaming. Which you can clearly see from NBR review of the card. What I`m showing here is the extra icing on the cake. Bandwidth isnt set in stone when you can OC the memory

At the clocks like this, I had like 55% bandwidth usage lol.






Anyhow, go buy the 680M. It is undoubtly a better card.

 

Supersampling and greater-than-1080p resolutions through external displays. Also Valley isn't the most intensive test you can do. Try Heaven or BF4 at max settings.

Effective memory bandwidth is doubled in SLI. That's why despite having equal or less shading/texturing/geometry throughput 750M SLI performs significantly faster than 765M and on-par with 770M.

 

Heaven use less memory and bandwidth than Valley

Why are you talking about higher res than 1080p. That is used by how many? 5%? Of course I agree that you need a 256bit bus there.

 

Well now that I think about it, you're probably right. Turning up tessellation to Extreme probably hits the shaders harder than anything else. Tessellation probably saves a lot of memory and bandwidth in the first place by reducing poly counts. And of course the rendered environments in Heaven and Valley are a lot different. But overall Heaven is still the more strenuous test. I definitely get higher FPS in Valley than in Heaven.

I would assume that at least some owners of those huge SLI notebooks use them as desktop replacements plugged into their big-screen 2.5K monitors every now and then.

And supersampling has made a comeback in the last few years due to more games using deferred and HDR rendering and thus having issues with traditional forms of antialiasing such as MSAA. It either doesn't work at all (e.g. Witcher 2), causes visual anomalies when forced on (e.g. UE3 games), or the performance hit is astronomical (e.g. BF3/4). Post-process techniques such as FXAA, MLAA, SMAA, etc. were developed and implemented in these games out of necessity, but because of side effects such as blurriness and temporal aliasing, eye candy Nazis will have none of it. That's why supersampling is becoming popular again. Sure, it's a brute force method that is guaranteed to kill performance on all but the most powerful of rigs, but it works on pretty much anything and the image quality is unmatched. Many developers have started natively implementing it in games such as Witcher 2, Sleeping Dogs, BF4, and PS2, something which I didn't see five years ago.

And here's the kicker: Since supersampling is essentially rendering the game at several times the resolution of your monitor and then scaling it down to your native resolution, it consumes VRAM and memory bandwidth like no other. You want to see how your 1080p notebook can handle 4K right now? Fire up BF4, increase Render Quality to 2 for 4x supersampling, and watch it blow up!

So I have to disagree with you somewhat about not needing more memory bandwidth even at 1080p, to say nothing of the resolution of future displays.

 

Or try Crysis 3 at 1080p with all settings to max including some antialiasing. That ate 1950mb of ram even with it running at 185GB/s! 192 bit has to be OCed to be enough in modern titles.

 

With the 780M SLI I see FPS performance increases all the way up to 1650mhz (6600mhz effective).

 

3DMark 11 Performance benchmark:
499MB, 45% memory controller load @ 1200MHz

3DMark 11 Xtreme benchmark:
793MB, 55% memory controller load @ 1200MHz

Unigine Heaven Maxed out:
1174MB, 64% memory controller load @ 1200MHz

Unigine Valley Maxed out:
1239MB, 68% memory controller load @ 1200MHz.



Out of those 4 benchmarks, Valley is the greatest one to test your graphic cards. It use way more resources than 3DMark Xtreme preset, it doesn`t skew the result if you bench it with a faster CPU. It is very much GPU bound and the result is in FPS, not in score, which makes it a lot easier to conclude.

The problem with comparing results with Crysis 3, is that it its impacted by CPU speed. Meaning a 3930XM with 780M will do better than 3820QM with 780M. But for your own testing, its absolutely better to test the system than the 4 benchmarks I tested. No doubt.

octiceps: I have to admit, I have never heard about supersampling before. Please dont shoot me lol. I have to go and read a bit about it. Sounds interesting

 

So you don't get confused, it is sometimes also called downsampling and, more archaically, FSAA (full-scene anti-aliasing). They're different names for the same thing.

The story of supersampling in computer games is pretty interesting. Years ago it used to be the only anti-aliasing available on most video cards, then it died away when "optimizations" such as MSAA become popular. Now that MSAA no longer works in many games and the post-process AA that replaced it is visually lacking for many people, SSAA is making a comeback.

 

Now that I'm back with Nvidia, I've started experimenting with SGSSAA (Sparse Grid Supersampling AA) via Nvidia Inspector. The results really are spectacular.

 

The naming is getting out of hand.

 

You think that's a mouthful? We haven't even scratched the surface of anti-aliasing techniques and their naming conventions LOL.