GPU Battle: GTX 670M vs GTX 570M vs GTX 660M

Kepler is really a jump from the previous generation Fermi architecture. I can’t say I’m excited or something like that, but the benchmarks and tests so far show good improvements: a midrange GPU like the GT 650M with a 128 bit memory bus width is almost as fast as the high end GTX 560M and sometimes even surpasses it. Plus lower power consumption plus the fact that you can find the GT 650M in a 800$-900$ gaming laptops where before you wou;d have to spend 1100-1200$ to get an 1080p 15.6″ gaming laptop with the same power and overall quality. That’s a difference.

Now we’re gonna talk about the lower high end section of mobile GPUs. The new line of Nvidia includes the GTX 660M, GTX 670M, GTX 675M. The GTX 670M and GTX 675M are actually the Fermi based GTX 570M and GTX 580 renamed and manufactured in a 40nm process. The GTX 670M is OC’ed version of the GTX 570M with additional ~20MHZ to core and high increase in memory speed (addiotional 350MHZ) which translates to ~10% averagly max gain in performance (see games benchmarks here).

The only exception here is the GTX 660M (and ofcourse, the yet to be announced king GTX 680M). The GTX 660M is the only Kepler based GPU in the gang, manufactured in a 28nm process. It has the same core as in the GT 650M only it has a GDDR5 memory (GT 650M has a DDR3 version too) and it’s core and memory clocks are considerably higher altogether (additional 100MHZ for the GDDR5 version). Compared to it’s bigger brothers the GTX 670M, it has more shader cores (but clocked as high as half of the 5xxM shader cores) and it’s more power efficient. It’s raw processing power is between the GTX 570M and it’s OC’ed version – the GTX 670M.

Model Core, shaders (vertex / geometry / pixel) memory bandwidth (GB/s) Core speed processing power (GFLOPS)
GTX 660M Kepler (GK107) @ 835MHZ,  384 unified : 32 : 16 @ 835MHZ, 28nm process 64 GB/s : GDDR5@2000MHZ, 128bit 835MHZ 791.3
GTX 570M Fermi (GF114) @ 575MHZ, 336 unified : 56 : 24 @ 1070MHZ, 40nm process 72 GB/s : GDDR5@1150MHZ, 192bit 575MHZ 772.8
GTX 670M Fermi (GF114) @ 598MHZ, 336 unified : 56 : 24 @ 1196MHZ, 40nm process 72 GB/s : GDDR5@1500MHZ, 192bit 598MHZ 803.6
GTX 675M Fermi (GF114) @ 620MHZ, 384 unified : 64 : 32 @ 1240MHZ, 40nm process 96 GB/s : GDDR5@1500MHZ, 256bit 620MHZ 952.3


In real world scenarious the GTX 660M is averagly a bit slower than the GTX 570M and ofcourse from the GTX 670M too. Here are some results in games benchmark (taken from here)

Settings GTX 660M,I7-3610QM GTX 570M,I7-2630QM GTX 670M,I7-3610QM
Battlefield 3 1920×1080 ultra 4x MSAA 16xAF 18 18 20
1366×768 high -AA 16xAF 40 41 47
Starcraft II 1920×1080 ultra 42 42 48
Metro 2033 1920×1080 DirectX 11 / very high AAA 4xAF 12 14 14
1600×900 DirectX 10 / high AAA 4xAF 35 37 40
Dirt Showdown 1920×1080 Ultra Preset 4xAA 24 29
1366×768 High Preset 2xAA 70 73
Diablo 3 1920×1080 High onAA 47 60
CoD: Modern Warfare 3 1920×1080 Extra, all on, Image Quality: Native 4xAA -AF 51 55
The Elder Scrolls V: Skyrim 1920×1080 Ultra Preset 8xAA 16xAF 28 26 33

The difference between the GTX 660M and GTX 670M is around 15%-20% averagly. You can also see that in some games, the GTX 660M performs even better than the GTX 570M, like in the Skyrim test. Also, in some games, the 660M is very close to the GTX 670M, like in “Dirt Showdown” and “CoD: Modern Warfare 3” and more. Taking into account that the 660M is much more energy efficient, we can say it’s a real improvement.

However, one cannot wonder what’s the deal with the GTX 670 – it is only slightly faster than the GT 660M (as we said, about 15%) and in most cases it won’t be the GPU that will take your game from stuttering to the smooth way of things and with the fact that it is more energy hungry I can’t wonder what’s the point and why Nvidia did this. Seems kinda wrong.

The GTX 675M is a different story as it is the previous generation GTX 580M and it is a good 15-20% faster than the GTX 670M averagly (check it here)


So, GTX 670M vs GTX 660M?

In my opinion, if you can save 100-200$ and buy a 660M instead of 670M than do it or even if you prefer almost to have more battery runtime and almost the same performance. But it comes down to the price, again.


And GTX 675M vs GTX 670M?

Well, if you can find a GTX 675M for only a small price premium than you probably better take it, if you need that kind of performance. Otherwise, You’ll consider the GTX 670M which we know has a good competition from the GTX 660M.

Anyway, we haven’t seen the whole line of Radeon counter parts yet, but only the Radeon 7970M which is the fastest mobile GPU currently (excluding alien tech). The strange thing is that some companies offer the 7970M upgrade for less than the GTX 675M upgrade. It might be just aggressive pricing thing and it might be due to some problem of some kind (the AMD ‘Enduro’? bad batch of 7970M?)