NVIDIA Kepler VBIOS mods - Overclocking Editions, modified clocks, voltage tweaks

[doctr_nick]

Hey guys new to the forum here. Been reading multiple threads about overclocking/bios mod for the gtx 680m and am curious as to whether anyone has tried any vbios mod in the MSI GT60 as I see most of you are running Clevo's. I've used Afterburner to overclock, but get the same throttling issues that seem to be well described if i clock the core around 850. This is with leaving the memory at stock. If I try to do both, even the "turbo" function that the laptop comes with (which brings the core to 771) induces throttling.

Do we know what seems to be the culprit for the throttling? Is it that the GPU doesn't get enough juice at higher clocks, or is it that the PSU's just aren't strong enough (the GT60 is a 180W)?

[lolklkl]

@0xsergy - set 3DMark11 to run test 1 only. Set your core at 850 and leave the memory stock. See if the throttle goes away. If it does, set it for 1000Mhz and test again. If you go too far with core or memory you are going to throttle. What you see others running can be deceptive. If they are using a locked core vBIOS, what it seems like the core is running will not be accurate.

I have the free version of 3dmark 11 unfortunately so I cannot do that. I can run the test a few times and just quit out after the first test.

EDIT: I get it, you meant setting it to 1000 instead of 900. No idea why the Memory is sometimes stated as 1800 and sometimes as 900...

That's because DDR5 ram is derived from DDR3 ram which when run in dual channel mode, such as in your graphics card, is actually at 900MHZ each(I think I got that right, someone correct me if I'm wrong). Many programs just double the value to keep it user friendly, etc.

Hmm, I don't remember if I had memory OC or not when I gave it 20 to the core. I'll try again now with memory at 1800 and core up to see how it performs.

EDIT: Memory at stock it managed to get up to 50 on the core before throttling. I'm starting to wonder if this is perhaps just an issue with all the newer cards..?

[doctr_nick]

EDIT: Memory at stock it managed to get up to 50 on the core before throttling. I'm starting to wonder if this is perhaps just an issue with all the newer cards..?

[lolklkl]

I can tell you that my system behaves the exact same way. I have to keep the memory completely stock to get any sort of OC, and then when I do, all I'm able to get is between 800-850 before it throttles. Temps are fine @ 88C under full load.

I mean, it's not that terrible I suppose. It's only on these artificial tests that it throttles on. I just tried with a 135/400 OC in BC2(don't have BF3 as I don't like it very much) with everything maxed out and AAx32 just to put extra load on it and it never throttled once(although it did put out a nice framerate, constant 90+ FPS)(temps weren't bad either, without FN+1 they maxed out at 84C and with FN+1 they didn't go past 77). Although I have another problem on the GPU that worries me and I'm currently discussing it with the manufacturer. When I have a program running to take temperature readings(such as Afterburner) the GPU makes a very small/quiet click noise every time a temperature reading is taken. I haven't seen this problem mentioned anywhere else so it worries me and just for this reason alone I may RMA the card depending on what the manufacturer says might be the cause.

[MrValle]

It is a mystery why these cards throttle when benching. My card throttles with stock bios and all stock clocks. I tried to underclock it but it still throttles so its weird.

[lolklkl]

It is a mystery why these cards throttle when benching. My card throttles with stock bios and all stock clocks. I tried to underclock it but it still throttles so its weird.

What manufacturer is your card from? Bios version, memory?

[doctr_nick]

I mean, it's not that terrible I suppose. It's only on these artificial tests that it throttles on. I just tried with a 135/400 OC in BC2(don't have BF3 as I don't like it very much) with everything maxed out and AAx32 just to put extra load on it and it never throttled once(although it did put out a nice framerate, constant 90+ FPS)(temps weren't bad either, without FN+1 they maxed out at 84C and with FN+1 they didn't go past 77). Although I have another problem on the GPU that worries me and I'm currently discussing it with the manufacturer. When I have a program running to take temperature readings(such as Afterburner) the GPU makes a very small/quiet click noise every time a temperature reading is taken. I haven't seen this problem mentioned anywhere else so it worries me and just for this reason alone I may RMA the card depending on what the manufacturer says might be the cause.

It's good that yours doesn't throttle while gaming. If i clock mine at 850 with stock memory, mine doesn't throttle in games either, but for some reason the framerate just isn;t what I'd expect. In Crysis 2 for example I'd get maybe 2fps more in a given scene, but then all of a sudden I get dips of like 5-10fps even though afterburner doesn't indicate any throttling of the clocks. If I leave the core at stock and OC the memory to say 2300, I get the same 2fps increase, but there's no dips, and performance overall is better. It's weird.

[lolklkl]

It's good that yours doesn't throttle while gaming. If i clock mine at 850 with stock memory, mine doesn't throttle in games either, but for some reason the framerate just isn;t what I'd expect. In Crysis 2 for example I'd get maybe 2fps more in a given scene, but then all of a sudden I get dips of like 5-10fps even though afterburner doesn't indicate any throttling of the clocks. If I leave the core at stock and OC the memory to say 2300, I get the same 2fps increase, but there's no dips, and performance overall is better. It's weird.

I haven't tried it in Crysis 2 however so the comparison isn't exactly accurate. You might want to decrease the polling time in Afterburner to, say, 500 ms. It might be throttling just so quickly that AB doesn't see it happen.

[Wei Lu]

Hi guys,

Can someone tell me which oc version bios of DELL 680M is best to be used?

[doctr_nick]

I haven't tried it in Crysis 2 however so the comparison isn't exactly accurate. You might want to decrease the polling time in Afterburner to, say, 500 ms. It might be throttling just so quickly that AB doesn't see it happen.

So I found out that it was the CPU throttling that caused the problem. Throttlestop sovles that. So it has to be a general power thing where once the components draw so much, both the CPU and GPU have throttling algorithms that kick in. It makes sense as some people have indicated that running at the same clock, using a vbios that increases voltage over stock induces more throttling. Ironically however, using Throttlestop for the CPU didn't cause my GPU to throttle any more frequent even at 850Mhz.

[Guest]

There may be a correlation in overall system power draw that can be influenced by the CPU, and that might turn out to be a really important discovery. I'm glad you posted it, as it could lead to some very favorable conclusions. But, merely having ThrottleStop running does not seem to improve or worsen the situation, at least it does not on my system.

I am assuming you may not be overclocking the CPU all that much. What settings are you running on your CPU for multipliers, flex and pri plane? The reason I ask is that I run ThrottleStop at system startup and use it 24/7. I typically have my M18x R1 running at 45x4 all the time, with c-states enabled in ThrottleStop in the boot profile. My performance profile for gaming and benching disables c-states, which should increase CPU power application. This might be sucking some power away from the PCIe bus, which may affect the behavior of Link State Power Management. Have you tried disabling LPM in your BIOS or Windows Power Profile? (I have it disabled in both places and I believe I may be experiencing less throttling than it seems others are reporting.)

When I have time I will test your theory of CPU influence by running my CPU at a lower power state with less aggressive OC. If I see something that seems to correlate with GPU throttling like you observed, I will post the result. Are you running M18x R1 or R2? I ask because there may be a difference with Ivy Bridge versus Sandy Bridge. I think the complexity of resolving this issue is the variety of chipsets and system configurations involved, Clevo versus Dell cards, etc. There must be a common link among all of them, but we need to figure out what that is, and whether it is isolated to the GPU architecture or influenced by multiple factors.

The 680M will throttle if any of the following conditions are present:

• Too much core OC
  
• Too much memory OC
  
• Wrong combination of core and memory OC
  
• Too much voltage
  

The behavior varies by application. What throttles in one benchmark test or game might not throttle at all running other software. One example is 3DMark11 Test #1. A moderate overclock will cause throttling (or a TDR/driver crash) in Test #1, but a more aggressive overclock does not throttle in Tests 2, 3, or 4. The Combined Test (last one) will often experience a TDR/driver crash with an OC that is more aggressive, but works well in Tests 2, 3, and 4. You can also push the GPU harder in Vantage than 3DMark11. There seems to be a dotted line connection with DX11, but I am not 100% certain about that yet. What I have seen from my own testing with 680M and 7970M is that any benchmark or game that uses DX11 does not handle overclocking as well with 7970M or 680M. That may have something to do with increased use of memory, shaders and more tessellation, which may somehow trigger throttling or driver instability with an aggressive overclock.

[sirana]

Does Throttlestop make any sense with my i7 980? Running it on stock 6x 3,33 Ghz and never had an issue with it.

[lolklkl]

So I found out that it was the CPU throttling that caused the problem. Throttlestop sovles that. So it has to be a general power thing where once the components draw so much, both the CPU and GPU have throttling algorithms that kick in. It makes sense as some people have indicated that running at the same clock, using a vbios that increases voltage over stock induces more throttling. Ironically however, using Throttlestop for the CPU didn't cause my GPU to throttle any more frequent even at 850Mhz.

I don't experience CPU throttling. AFAIK the P170EM only throttles the CPU under very heavy loads so I don't see how throttlestop would help my gpu scores but I'll try it in a bit anyway, just to rule out any possible what-ifs.

[Guest]

I don't experience CPU throttling. AFAIK the P170EM only throttles the CPU under very heavy loads so I don't see how throttlestop would help my gpu scores but I'll try it in a bit anyway, just to rule out any possible what-ifs.

I am not sure what system doct_nick is running. If he has an Alienware with an unlocked system BIOS, he may be able to overclock his CPU more aggressively, which might have an effect that we cannot replicate between different systems.

[doctr_nick]

I am assuming you may not be overclocking the CPU all that much. What setting are you running on your CPU? The reason I ask is that I run ThrottleStop at system startup and use it 24/7. I typically have my M18x R1 running at 45x4 all the time, with c-states enabled in ThrottleStop in the boot profile. My performance profile for gaming and benching disables c-states, which should increase power implementation.

When I have time I will test your theory of CPU influence by running my CPU at a lower power state with less aggressive OC.

Im running an MSI GT60 with a 3610QM. I don't overclock the CPU at all. I just have throttlestop set to run it at the turbo multiplier (which according to the log file is 31) when gaming. I'm not so sure the CPU can actually influence the GPU as it is the other way around. Realistically the components are all connected to the motherboard which has to get power from a common source (PSU). Its how that power gets distributed that may affect things. Are the CPU and GPU intricately tied together? When clocking up the GPU, i tended to get poorer performance on occassion, and this was due to the CPU multiplier being lowered, and not throttling of the GPU. Running throttlestop for the CPU has solved this, but has not changed the behaviour of my GPU (i haven't had to lower clocks because the GPU has started throttling due to increased power consumption of the CPU). This suggests that the two may not be related and what i saw before using throttlestop could have just been a coincidence.

What we know:

1) vbios with too high voltage induces throttling

2) heavy OC on stock voltage induces throttling

3) keeping stock mem speed can increase max core OC before throttling

4) Running max mem overclocks however does not do the reverse (ie. doesn't induce throttling at stock core clocks) -interestingly enough

All three of the above have in common a resultant increase in power consumption. So this leaves a number of questions:

1) Is it driver/vbios coded where if the chip consumes beyond it's pre-defined levels it throttles?

2) Is it designed that way at a hardware level?

3) Is it somehow a system bios limit to prevent the graphics card from sucking too much power through the motherboard?

4) Is it that our PSU's can't supply the power demands (unlikely as it would make most sense for the system to shut down as others have described)

And why are we so limited with stable overclocks (and by stable, I mean without throttling or locking)? I think it has to do with the Kepler arcitecture. Keep in mind the 680m still has a whopping load of transistors squished together with the 28nm process. Any increase in clock would have a significant jump in power consumption which is probably why nvidia targeted the clocks of the 680m so low in the first place. It may just be that we have to trade throttling for frying our chips and or the power circuitry that supplies them

[lolklkl]

Any why are we so limited with stable overclocks (and by stable, I mean withouth throttling or locking)? I think it has to do with the Kepler arcitecture. Keep in mind the 680m still has a whopping load of transistors squished together with the 28nm process. Any increase in clock would have a significant jump in power consumption which is probably why nvidia targeted the clocks of the 680m so low in the first place. It may just be that we have to trade throttling for frying our chips and or the power circuitry that supplies them

But what about people like Meeker that are able to achieve 1035/2400(iirc) clocks without throttling? I think it may have a bit to do with the production differences between cards. Not all cards are produced identical so this probably plays a big role in all this.

[sirana]

Meaker uses the vbios of the other thread, which is slightly overvolted. Sadly it does not seem to work correctly for every Clevo card...

[Guest]

That's what I was indicating about variance between systems/platforms. We have Alienware, MSI and Clevo systems running 680M and some with 680M made by different OEMs. The power supply (AC adapter) capacity, as well as how power is managed, is different on each platform. Some are overclocking CPU with an unlocked BIOS, which may draw power away from the PCIe bus. There are many variables in play, and an immature understanding of Kepler affect our ability to know what corrective measure(s) will result in the killer solution. More to follow... this is still a very new product.

Remember how things were with the Dell GTX 580M to begin with, especially in the M18x R1. Nobody wanted it and many were quick to declare that a failure and 6990M CF as the victor. Well, that was absolutely true at first. Throttling was the driver of that situation as well. Once talented people got involved (svl7 being one of them) and we figured out the secret sauce, the M18x R1 with GTX 580M SLI romped all over everything. It was a bit of a massacre once it got sorted. Then the winds shifted and a bunch of folks were clamoring to get GTX 580M SLI.

Let's keep on plugging away at it as a team and we will get it figured out before long. All good feedback and ideas being shared here. Even without any overclocking, the 680M is a real beast. So, I don't really see a major down side or negative aspect of having such a wicked GPU installed in any laptop that is able to support it properly.

[ajbutch123]

So, no working overvolt for the clevo cards? I am looking to flash the oc edition tonight, but would like to try an overvolt also to see if it will allow me to push the clocks even further

[Guest]

Hey guys. Here's an example of what I was talking about earlier today regarding throttling in 3DMark11 Test #1. Test #1 was run at 895/1013. Look at my screen shot and you can see what the rest of the benchmark was run at... all at stock voltage.

NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-2920XM Processor,Alienware M18xR1 score: P12907 3DMarks

[doctr_nick]

So I did some experimenting this morning with excellent results.

I used the modified MSI ES bios from the thread "Using your GTX 680m to safest and full potential" with clocks at 915/2250 and 1.0v overvolt and can say that things are very stable. Completed multiple runs of 3Dmark11 @ ~7200 points and ran Crysis 2 for 20 mins maxed out without a single instance of throttling or crashing. I specifically even looked at 3DMark 11 Test 1 with Afterburner and had no throttling at all. Locked stable at 915. Interestingly, after applying the bios, I couldn't install any video drivers as it said there wasn't a recognized adapter. I went over to laptopvideotogo and grabbed the newest 306.97 with a modified INF and installed just fine.

So it seems, at least with my MSI GT60 that power consumption is not an issue, and perhaps the throttling is actually driver based. Forcing the driver install with a modified INF may prevent the driver from applying some of its throttling routines. Has everyone who's used this bios had to use modified drivers?

Im going to test more extensively to ensure things are stable, but at the moment it's looking good. Temps are surprisingly awesome as well, hovering around 84 max load.

[sirana]

Hmmm I think I will give the MSI vbios another try... just because I'm desperate and svl7 has not updated the Clevo rom yet that is

Speaking of which, @svl7 do you know when/if you are releasing your next vbios?

[doctr_nick]

Tested successfully at 1030/2350. 3DMark11 7854.

I found however just how important it is with clocking up the GPU to use throttlestop and uncheck the BD PROCHOT box! I forgot to uncheck that box at first and found that while this didn't impact 3DMark in any significant way, I would continually get fluctuating performance (dips into the high 20s and thirties) in Crysis 2. Reviewing my Afterburner log showed the GPU clocks holding rock stable, but the GPU usage was jumping all over the map from 98% down into the 60s and 70s at times. So you have to run throttlestop and uncheck the BD PROCHOT box. Doing this now has my GPU usage back at 98-99% constant and running Crysis 2 Dx11 Ultra with hi-res textures 50-60fps vs 35-45fps before bios mod and OC.

Only issue now is heat. Running those clocks with throttlestop brings me up to 93C even with turbo fan enabled. It's too bad this particular laptop only has a single cooling fan.

[Tonrac]

@doctr_nick : try 1005/2400

[MrValle]

i also got the "no recognized adapter" when trying to install drivers on stock bios and modified bios. The .inf file from laptopvideo2go didnt work either so i had to mod the .inf file myself with the hardware id of my card to get it to install. And the throttling is there whichever bios or overclock and underclock i run. And throttlestop is also a no go.