Just how quick is a 780m anyway?!

[Xaser04]

This is a question that has bugged me for a while yet I have never been able to get conclusive proof; Just how fast is a GTX780m(880m) in relation to desktop cards?

Various reviews and forum posts state the performance is anywhere from GTX670 levels (makes more sense) right the way down to the 750Ti.... (Just no). Actual analysis compared to desktop cards is hard to come by.

Well not any longer!

I recently got fed up with Eyefinity profiles causing all manner of trouble with my 290X. So I decided to make the move back to Nvidia Surround. I am waiting on the Maxwell cards to hit to properly upgrade, but in the meantime I bought this little beast:

This is an Asus GTX760 DCII Mini, it clocks in at 1150/6000 out of the box and is just so damn cute!

Given this card offers up slightly faster than GTX670 reference speeds, I thought it would be fun to use it as a base line comparison against the GTX780M both stock and overclocked.

The reasoning behind this is that I have always though the 780M to have similar overall performance to a GTX670 based on its specifications alone. For those who don't know this is how the cards stack up:

GTX780M - 1536 SP @ 849Mhz / 128 TMU / 32 ROPS / 5000Mhz 256bit GDDR5 160GB/s
GTX670 (Ref) - 1344 SP @ ~980Mhz / 112TMU / 32 ROPS / 6000Mhz 256bit GDDR5 192GB/s
GTX760 (DCII Mini) - 1152SP @ 1150Mhz / 96TMU / 32ROPS / 6000 Mhz 256bit GDDR5 192GB/s

All of the cards are based on the Kepler GK104 architecture with the only differences being how many SMX's are enabled and the clock speeds.

We all know that a reference GTX760 is only around 5% slower than a reference GTX670 clock for clock, thus my 760 will match up quite nicely against a mildly clocked 670 (say 1050Mhz core). Arguably at 1150Mhz on the core it is closing in on a reference GTX680... :worried:

This sets the stage for a long overdue comparison.

The Benchmarks

I am running through a few benchmarks which I have easy access to on both systems. I was going to add Titanfall to the benchmark but the multplayer nature coupled with the inability to select a map made this not really possible for a quick comparison.

Test Setup

GTX780M
i7 4700MQ @ 3.2Ghz
16GB DDR3 @ 1600MT/s
H87M Clevo platform

GTX760
i7 4770S @ 3.9Ghz
4GB DDR3 @ 1600MT/s
MSI B85i mITX

GPU clock speeds (core/memory):

GTX780M Stock - 849/5000 (undervolted by -25mV)
GTX780M OC - 950/6000 (undervolted by -25mV)
GTX780M "Max OC" - 980/6000 - This is the max overclock at stock voltages

GTX760 Stock - 1150/6000
GTX760 OC - 1202/7000


Battlefield 3

1080P Ultra Settings with 4xMSAA
5 minute FRAPS run through of "Operation Swordbreaker"

Well there you go. The 780M is marginally slower than a GTX760 at stock clocks and quicker once overclocked. The overclocked 760 is marginally quicker again but the difference is quite small. This is a pattern that will become all too familiar...
I forgot to run FRAPS with FPS over time for BF3. Having re-run the same section 100's of times for various reviews I really didn't want to go back and do it again....

Crysis 3
1080p "High" system preset with Very High Textures - SMAAx1
5 minute FRAPS run through of "Welcome to the Jungle"

Well what do you know, the performance is in line with what we saw in BF3. The delta is closer this time by a small margin (in favour of the 780M).

Sleeping Dogs

1080p "Max" settings with High AA
Built in benchmark

And again.... This time the 780M OC is much closer to the 760OC. If I could get the memory higher it would overtake it!

Thief

1080p "Max" settings with High SSAA
Built in benchmark

The two cards overclocked are basically identical in this benchmark. The 780M at stock clocks is a little slower than the stock 760 but the difference isn't massive. Interestingly the minimums are much higher on the 780M. I re-tested this a few times and each time the result was the same. This could be a case of the raw SP advantage being more important than clock speed alone (possibly due to scheduling?).

3DMark11

Standard test (P)

The 780M wins out here despite a weaker CPU. I must admit, this was the result that got me thinking about doing this comparison in the first place.

Benchmarks Summary & Conclusion
I have always though the 780M performed (or could perform) roughly the same as a GTX760/670 and here is my proof. In all of the games tested the 780M is right there side by side (some could say slightly ahead) of a pre overclocked GTX760.
Ultimately this puts its performance squarely at mildy overclocked GTX670 levels, which in a laptop isn't half bad!
Overall I am very pleased. I now can't wait for the true successor to my 780M!

 

[digibucc]

nice job. how much do you think ram made a difference?

 

[Xaser04]

nice job. how much do you think ram made a difference?

Thanks.

I think the ram made more difference for the 780M than it did for the 760. At 1080p "playable settings" the 192GB/s bandwidth seems enough for the 760 core to keep it fed, thus the increase to (7Ghz) 224GB/s yielded less benefit. At 5888x1080 this is a different story mind with the bandwidth being much more important.

I remember doing some memory scaling on the 780M in Crysis at the "Very high" preset. The move from 5Ghz to 6Ghz netted around 3FPS, whilst the core increase from 849-980 yielded around 5-6FPS. Together they totalled up to around 8-10FPS more at the same settings.

Due to the thermal constraints it is all a balancing act. The Clevo GPU heatsink design is for want of a better word crap. The memory can go further than 6Ghz but at the expense of core clocks which offer up more raw performance improvement.

EDIT: Actually do you mean system RAM quantity?