overclocking the gtx675m

all,
my machine is currently in germany at the alienware depot, after i had a dead gpu.
this got replaced with the whole motherboard and various cables at my house..
the tech dude wasted too much parts (2x mobo , 2x gpu, 2 sets of heatsink)
so,,, i had to send it ..

it will be back somehwere around next week, and still have the same config: i7-3610 / gtx675.

now.... i will upgrade the gpu, but also thinking, what if i just oc the hell out of it..
it would reach on stock clocks (620 / 1500) in stress tests about 68C as absolute maximum..(with arctic silver 5 paste)
there is still some thermal room here,, what can i expect from an overclock on the STOCK vbios..
it doesnt let me change voltages, only core and mem speeds..

how far can i push it with the stock voltage/cooling?

 

Hi, I find on mine that 700/1650 is nice and stable on the stock vBIOS. This is a new card from dell after the last one died (thank you NV for your super drivers!). The last one would run 720 without issue. This one craps out at around 710 with flashing dots appearing. It's a suck-it-and-see kinda thing. Variations in manufacturing runs will give different results.

Now, on the stock vBIOS the main problem is P1 throttling. This is triggered by some sort of algorithm that takes power draw into the trigger. When you OC you are usually only changing the P0 setting. The throttle/2D clock is known as P1. If you see this happening then you can use nvidiainspector to raise the P1 clock to match P0. Problem with this is that not only does the clock drop but the voltage also drops, from 0.87v to 0.85v so more susceptible to artifacting. This is where a modified vbios is useful. The one I had would match P0 and P1 clocks at 620 AND voltages so no power throttle.

I'm not sure why but I have noticed over time that the power throttle happens less and less - could be just good drivers/better game programming?

I use nvidiainspector to raise P1 to 620 when I have P0 set to 700 with the stock vbios. This way I get to see when it is happening and it does not destroy a gaming session. The stock P1 setting of 410mhz is unplayable

Temps never go above 68c and more often than not is a few degrees lower than that, and it makes quite a difference in gaming. Around 10% FPS increase average.

 

Hi!
Just wanted to jump in and ask you to be careful with the overclocking of the GPU because it will decrease the lifetime of the GPU and may end up harming the computer as a whole. Just make sure you don't push it to hard and monitor the temperatures at all times.
Damage caused by overclocking is not usually covered under warranty.

 

The 675M should be fairly safe at stock volts, just be careful with over volting these as it can be dangerous.

 

the lovely new nvidia drivers screwed my card up, so i have a brand new gpu..
i will change it for gtx770m, but till then, i will have a play with the old power hungry fermi beast..
if i can get it cool enough for 700 / 1650, that will increase the performance by about 10%..
i think that will do
i have this card, just for fun, since dell replaced almost everything in my laptop (gpu / mobo / cpu / heatsinks / fans)
so im starting on almost brand new laptop..

first i will leave it stresstesting overnight,, if its still running in the morning on tuesday, i will see how far i can push it (and maybe might flash a custom vbios)

 

I had my 570m up at 850mhz with volt mod. 800mhz with volt mod was normal for the 675m iirc so 700mhz should be doable on stock volts. Of course every chip is different.

 

i got 1 now that runs REAL cold.. on stock clocks it wont hit 65C..
this is repasted with arctic silver 5..
i can do 705/1650 on stock bios, BUT, there is 1 problem..
at these speeds, it will reach 67C..
then gpu is losing load...

clocks stay this high, but it just loses gpu usage...

i dont know how to fix this (i read something about nvidia inspector, but i cant change clocks there)..

is it worth flashing a custom vbios? and just get rid of the 67C throttle?
then push it to the hard limit? i think i can reach 800 / 1750,, if it was possible on this bios (im really thinking about this.. but how big is the chance to brick the card flashing vbios?, i dont care about oc limits, or burn the card)

 

There is no 76C throttle. It may happen around there but it is not temp related. It is triggered by a larger than normal power draw. Overclocking makes trhis happen more often.

You say your clock speeds are staying high but this is not normal at all! When the power throttle cuts in the main clock (P0) drops from 620 to 410 (P1). You can use nvidiainspector, advanced settings to up the P1 to 620 quite safely which will reduce the effect of the power throttle. Downside to doing it this way is that the power throttle also drops the voltage from 0.87v to 0.50v so pushing P1 over 620 can cause artifacting. I do have a vbios that matches P1 and P0 voltages and clocks (620/0.87v).

Temp is still way low at 67C. There is a throttle for overheat but it happens at around 78C. I find that a 700/1650 overclock does not trigger the power throttle except when it is supposed to - like when an in-game cut-scene plays. My last 675 at 750 would get to around 72C. This one will not get even close to that. Since I saw problems at 710 I stuck with 700 and all is good, of course YMMV.

If you want the vbios I can post it for you. There is also one there with a 0.92v (not recommended) increase which could allow you to get much higher clocks. Remember though this is not a particularly reliable card and overvolting may damage it!

Edit: You use the shader clock to adjust the core in nvidiainspector.

 

So basically, u never get a big big oc on this, without overvolting it?
Overvolting is e on these cards and will destroy it...
Hmmmmmm... i will get pn ebay tommorow, see for a replacement gpu, something kepler..
I might go 770+custom bios+ big oc (150+ core 250+ mem)
Or go big bucks 780m, lr 680, but 680 still damn expensive..

So what i really want, is a p0 and p1, matching 700core/1650mem, with 0,87volt ln both states..
But cant i just unlock the full package, and push p0 and p1 to 750/1700 on 0,90 v, i think alienware is capable of cooling it, but will the old cripple fermi core survive this stick in the butt

 

Right, The most I have seen on stock voltage (from others as well) is 750. The overvolt Vbios is needed as you cannot get to the voltages via software.

To get what you want to try you need to use the modified vbios that matches P0 and P1 clocks and voltages. from there you use nvidiainspector to set the clocks where you want to test. I do not recommend above 680 on P1, since the card is complaining I prefer to let it slow down a little. Up to you though

Attached:

675_original.rom = stock taken from my new 675m.
675_mod rom = matched P0 and P1 (620/0.87v
675m92v.rom = overvolted, but I do know for sure the clock speeds as I never tried it. I assume it's the same as the matched one just with extra voltage.

Bricking from a vbios flash is rare and I've never seen it but of course it's all at your own risk. And overriding the mismatch the nvflash program asks you for is normal (before this happens and you ask) .

Edit: forgot to add nobody has managed to get a 770m to work on an R3 or R4 so it might not be the best option

 

You could get a 670MX (which works in the R3 at least) instead of the 770M, but to get the same performance you NEED to flash a modified VBIOS and overclock by a large amount - safe though in my opinion. It's effectively the same chip as the 770M in terms of number of stream processors & memory layout, clock speeds lower though, (and it's a GK104 chip as opposed to GK106, but that's not really important).

(Can't find any on ebay anymore though, at least not to buy new - strange as I bought mine in the Summer).

 

i just flashed the modded bios (matching p1 and p0 on 0.87 volt, 620/1500)..
flashing it was super duper easy!!

u guys are GREAT

in furmark it will go up to full original clocks 620mhz core, but my oc on the memory will boost up in furmark (1750)

got it stable, at 725 / 1750, no artifacts, and under furmark it reaches a MAX of 75 C

 

I'd use Unigine Heaven benchmark rather than Furmark to test temperatures & long term stability. Furmark is not representative, NVidia drop the clocks automatically & artificially on Furmark to protect the card as otherwise Furmark places unrealistic loads on the GPU. Furmark is not testing your overclock properly if it's dropping the clocks.

 

will test it later, i have a very slow connection, and im at work now..
it seems stable, temperatures are low enough (seen it reach 77c max), but still the gpu usage isnt 100%..
i dont think my cpu is bottleneck, i7-3610 should be fast enough..
maybe its my ram? its still the 6gb original

 

RAM's good - as in your RAM is fine.