Post Your GTX 680M Overclocks and Temperatures

There are more and more people using Saltius vbios so it shouldn`t be a problem getting one.
If you want a copy, I can send it to you. Just msg me

Great score man. 7134 GPU points is pretty good. And 66C is amazing for a GPU with this performance.

 

That's right, even the HD7970M starts off 1200MHz (2400MHz at nvidia's standard).

 

Didn`t you test your 680M with different clocks, hackness? Did you find out anything about the stock memory clock vs overclocked GDDR5 without overclocking the GPU? What did you find out?

 

I can run a handful of tests with just OC'd vRAM fixing GPU and fixing vRAM and OC GPU to check performance difference.

 

That would be cool

 

Just about done with the other runs, will start that when complete with this.

 

I found its still core/rop bound.

 

the difference between the clevo 7970m crippled by enduro and the 680m is astonishing... i picked up 10-25 fps in every single game that i have playing. this includes skyrim, the secret world, Batman AA + AC, Deus Ex, Dead Island , and the witcher two. when you take into account that now my overclock will actually do something instead of causing utilization to go down further, I'm seeing increases even higher... especially in the secret world. my FPS DOUBLED! went from average of 32.5 - 58.4... and i increased the AA qaulity and added motion blur so the 680m was running it at a higher quality too! sorry AMD i was rooting for you and really hope you get enduro sorted out but you really should have waited to release enduro to the public until it didnt suck. My major issue with enduro is that at this time the two major companies involved, Clevo and AMD, do NOT ACKNOWLEDGE THE ISSUE. step your game up ppl!

EDIT: all my screen tearing issues in skyrim, Dead Island (almost unplayable I had so much artifacting and tearing when things were moving fast), and Deus EX are now gone as well are almost completely gone. This is something not many others have mentioned but is as nice to see as the increased fps

 

welcome to the green side

 

Well, Sager and AMD have acknowledged it: http://forum.notebookreview.com/sag...0m-underutilization-issue-logging-thread.html

But personally, that's why I opted for nVidia this time around, I didn't want to wait, I wanted it to just work. Granted when I ordered there was no assurance it WOULD work but considering Optimus has been tuned for a couple years now and Enduro is new, I felt a lot higher confidence in opting for nVidia.

 

Now that you mention it that's where I didn't test before RMAing it

This is all the data I have collected before RMA, I'm just gonna copy it from the .txt I recorded with and paste it here, feel free to add more if you guys want to fill the blank:

3DMark11 stock vbios

Stock [Bandwidth 115.2GB/s] 69C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P6134 3DMarks
853/1800 [Bandwidth 115.2GB/s] 72C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P6932 3DMarks
853/1900 [Bandwidth 121.6GB/s] 72C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P6991 3DMarks
853/2000 [Bandwidth 128.0GB/s] 72C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P7040 3DMarks
853/2100 [Bandwidth 134.4GB/s] 73C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P7077 3DMarks
853/2200 [Bandwidth 140.8GB/s] 73C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P7098 3DMarks
853/2300 [Bandwidth 147.2GB/s] 73C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P7140 3DMarks
853/2400 [Bandwidth 153.6GB/s] 74C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P7153 3DMarks
853/2500 [Bandwidth 160.0GB/s] 74C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P7176 3DMarks

3DMark 11 with svl7 vbios v4 stock voltage

876/2300 [Bandwidth 147.2GB/s] 75C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P7230 3DMarks
935/1800 [Bandwidth 115.2GB/s] 70C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P7074 3DMarks
915/2500 [Bandwidth 160.0GB/s] 73C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P7276 3DMarks

3DMark11 svl7 vbios v4 stock voltage + throttleStop5.00 beta 3

915/2500 [Bandwidth 160.0GB/s] 74C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P7373 3DMarks



3DMark Vantage svl7 vbios v4 stock voltage + throttleStop5.00 beta 3

915/2000 [Bandwidth 128.0GB/s] 73C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P23130 3DMarks
915/2500 [Bandwidth 160.0GB/s] 76C: NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: P23620 3DMarks

 

amazing this is good amount of data also will provide alot for other people wishing to overclock

 

Thanks hackness. Now what is this telling us?

vRAM 1800 to 2500 gives us ~ 2% improvement in performance
GPU has negligible effect on improvement.

I am running sets of benchmarks on about a dozen games (including 3dm11) with the following clocks:

GPU/vRAM:
719/1800 (stock BIOS/clocks)
719/2400 (stock BIOS/I can't do more than 2450 vRAM without horrible artifacting)
1000/1800 (Saltius 1.037v)
1000/2400 (Saltius 1.037v)

That should give us a good comparison of GPU clock effect (~39% OC) and vRAM clock effect (~33% OC)

 

That's with core clock at 853, for stock core clock, need someone to test it

 

are you guys using inspector or AB to OC?

 

I'm using inspector, but it doesn't much matter really. AB has profiles and hotkeys, but I like the simplicity of inspector and shortcuts work great for me.

 

Both can OC, but for vbios that doesn't have the +135MHz limit unlocked you'll have to use modded vbios to reach higher clock/voltage.

@HTWingNut,

By the way I remember when I was testing in Final Fantasy XIV outside town (driver crashes every test but still got something out of it), with 16xQ CSAA + Depth of field + Ambient Occlusion. The FPS was around 39-40 @720/1800 when there's weather effect on the screen, and bumps up to 44-45 when I pull it up to 720/2500 while at the same place, same view, same weather. GPU usage was @99% during the bump.

So I guess it'll sometimes give around 12.5% increase in performance @stock core clock if you are running a game that relies on the GPU heavily with AO and high level antialiasing enabled, those rampage the memory usage as well. But too bad I should have checked the Memory controller load.

 

Thanks for the information hackness

Looks to me that you gain some, but not much with higher memory clock in 3DMark11. Going from 1800 to 2500 gives you 300 more GPU points.
But there are some problems with this testing. With "Performance" preset, you will only load the GPU with 720p and low settings, so you won`t stress the memory enough to see changes you made with memory bandwith. I`m positive that the reason you saw 10%+ more FPS in Final fantasy with going from 1800-2500 is because you are taxing the memory with 1080p and a lot more details, so it will have to "work" more, hence hitting the memory bottleneck much more efficient.

So yeah, I think we either need to test the various memory clocks in "Extreme" preset in 3DMark 11 or use games to find out.

 

I guess this needs to be tested by who holds the registered version

 

1037/1800:

NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: X2477 3DMarks

1037/2574:

NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: X2824 3DMarks

 

That's 14.16% increase in GPU performance

 

Thanks Meaker.

Is it possible to run the lowest GPU clocks possible with your vbios (not lower than 720MHz though), and then only do a stock memory vs overclock memory, to see if we see the same % changes also with less GPU clocks than 1037MHz? I`m very interested to see if we NEED faster memory than the 1800MHz with 1080p

(I`ve given away too much rep the last 24h so will rep you when I can)

 

Yep definately need to see if 1800MHz is the bottleneck with 720MHz in core

 

Looking forward to recieve my GTX 680m.
Does it perform differently without Optimus like AMD Enduro does on the HM series?

 

Performance is the same as the GTX680M in HM(except PCI-E runs at 2.0 x 16 in HM but not a big deal) roughly under 1920 x 1080 x 188~226 pixels per sec at 32 bit colour when output to the monitor. But that's only limited to the already processed frames, the actual GPU power is the same so does the benchmark etc.

 

Thats good to hear.
Will be my first Nvidia product in years so hopefully it can live up to the fanboy hype.

Sent from my GT-I9300 using Tapatalk 2

 

560M models shipped with 150W PSUs, 570M models shipped with 180W.

 

We need someone with a 3920XM, it should show the maximum score achievable (somewhat..)

 

Feel free to send me one

NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: X1953 3DMarks

710/1800

NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor,CLEVO P15xEMx score: X2087 3DMarks

710/2575

Yes i moved the slider a tad too far

But its pretty close

 

Would that change the GPU score? Or just the overall score? Or is it just for the sake of scores...

 

Over a Ghz gfx core in games it would help in some situations.

 

Thank you very much. Excellent. You gained 6.5% better GPU score with overclocking the memory from 1800 til 2575MHz. While you gained 16% with the same memory boost on 1000MHz GPU core.

Can we assume that the 680M is not memory limited at stock settings?

 

Only 6.5% increase. I guess there's an optimal core clock to find for the 2500MHz and 2575MHz Mem clock, think I'll do that when I get mine back.

 

It is a little, but it was likely a balance between core and mem clocks to hit their power target and the core still had a greater gain per watt.

 

Just asking, how far does the 680M 4G from clevo clock at stock voltages?

And how are you guys able to flash in a clevo? I thought it's a locked card vBIOS/mobo BIOS?

 

Only in DOS mode with newer nvflash like 5.105, have gotta force flash it: nvflash -i0 -4 -5 -6 insert vbios name here.rom

Edit: Also the limit at stock voltage seems to be 915/2600, if you go pass it the GPU will start throttling because there isn't enough voltage to feed the clock.

 

Right with the cooling pad I hit 87C max at 1053/2575 in crysis 2 (DX11, everything max).

This is at 1.062V.

My problem now is sorting out all the different bioses lol. I need a better naming scheme ^-^.

 

Thanks, for the info.

I'll be looking forward to a 680M around oct-nov and hopefully will be joining you guys

 

Either you guys are lucky or I just have crap vRAM, but I can't exceed 2450MHz for my vRAM.

 

It's very possible that the VRAM on the Clevo GTX680M runs at 1.35V, and we probably got lucky on the clocks at 1.35V. If it runs at 1.5V the standard will be 2500MHz.

 

I do seem to have been lucky myself. Could be undervolted.... Do samsung rate the chips for 1.35v operation?

 

I bet this is true. Nvidia spent most of the tdp budget on the GPU.

 

It was probably rated by Samsung, that model has 2 types, both are coded 04. 1.35V for 900MHz(NVIDIA standard 1800MHz) and 1.5V for 1250MHz(NVIDIA standard 2500MHz).

Here's the link for the 2GB version, 4GB isn't shown on Samsung site yet but the model code is the same K4G20325FD-FC 04 : Samsung Semiconductor Product Selection Tool

 

so max core oc is 1037 on the clock?

 

Depends on the bios lol, I went quite a bit higher (and hotter).

 

Added some ramsinks to the GPU contact plate.

 

the vRam Voltage should be 1.35 since the engineers at Samsung most likely did testing and already applied their own factor of safety and thus Clevo/Sager doesn't need to worry about applying their own factor of safety so it might not be the 1.5 V. however they do have this on the bottom of the page:

03:0.3ns (3000MHz)
04:0.4ns (2500MHz)
05:0.5ns (2000MHz)
5C:0.555 (1800MHz); 08: 0.83ns (1200MHz)
1A:1ns (1000MHz GDDR3)
1A:1ns (1066MHz gDDR3)
11:1.1ns (933MHz)
12:1.25ns (800MHz); 14:1.429ns (700MHz)
20:2.0ns (500MHz)
25:2.5ns (400MHz)

they correspond with the ending digit after the FC/HC/BC

 

Chips can be rated for more than one speed or voltage and all chips capable of one will do the other.

 

when you design a circuit board, you have make sure the voltage going into that chip 1.35V or 1.5V , chip itself maybe able to handle it but the circuit board will designed for one or the other so thus it will influence the speed. but It might rated at 2500 mhz hence the bolded term.

 

im with wingnut. the second i scoot past 2450 i start getting yellow blocks on my screen and then a display driver crash. Im not too concerned tho. the fact that i can play the witcher 2 at stock clock makes me a happy camper