Khenglish

Yes your TOLUD is 3GB.

I don't see why not. I have the feeling that you won't get much more out of 120Hz screens though.

Nice improvement. Still pretty short of the 780m though. Only variable left that I can think of is memory controller voltage, but I feel that nvidia might just directly use the memory voltage for that. By adding a 470u cap I went from 4860 to 4937 on 1.455V (measured while running) fully stable. Will add another cap or 2 and possibly inductor swap soon, as well as raise the voltage a bit more. I had to back the voltage down since it kept drifting too high on me with pencil mods. Plan now is to piggyback a 100k resistor between .7V and GND. I could get 5100 stable at higher voltage before adding the cap, so I expect 5200 or so eventually. What clocks did your memory do before the filtering change?

It's possible, but I recommend against trying it. I assume your goal is to put in a longer cable. A longer cable combined with the extra solder locations may make the PCI-E unstable, even if you solder perfectly. If you were to put in a longer cable I suggest fully removing and replacing the entire cable. Many of the signals require very high quality solder joints where the wires are pressed firmly into the contacts. The reason PCI-E 2.0 compatibility came out as late as it did was because the mHDMI connectors did not have good enough connections.

What was your cooling to get those crazy core clocks? Are you gonna switch out the inductor and caps on that card to get the memory clocks up? And how did you disable optimus to get a high score in ice storm?

What improved process? They both use the A2 revision of the GK104 core. The markings on the core denote a change in packaging, not the core. The 780m also runs at higher voltage, higher frequency, and has the 8th SMX unit enabled. Power 780m vs 680m: (1.012V/.987V)^2(8/7)(771/720) = 28.6% more power draw than the 680m core. (1.5V/1.35V)^2(1250/900) = 71% more power draw than 680m memory. The reason both have a TDP of 100W is because the 680m was well below 100W, but top end cards are labeled as 100W parts anyway. The 580m was also labeled as a 100W part, but people switching from the 580m to the 680m also got much lower temperatures, since the 580m was actually close to 100W. My understanding is that the m17x does have good cooling, so you should be OK though even though the power draw will be higher. You should be OK with the 240W power brick.

Stability refers to a card not outputting errors. Calculation errors can cause the graphics driver to crash and visible rendering errors such as black splotches, missing lights, random triangles, etc. Sometimes they look pretty cool, but if you see them a crash is coming soon. Yes fluctuation in clocks is normal. The card will do that to save power if the extra performance is not needed. For example if whatever you are playing is limited by the CPU and the gpu has plenty of power to spare, you can expect the card to clock a bit below what you set it to. Sometimes though it does underclock the card too much if you have performance set to "adaptive" in the nvidia control panel. I have mine set to "maximum performance" which makes the card clock down less, and this does result in an fps increase in some cases. I haven't noticed any difference in battery life since the card is off most of the time anyway when on battery. There is no required ratio between core and memory clock. You can clock both independently to whatever you want. Also most people use nvidia inspector to overclock laptops since msi afterburner has a bug that makes it take forever to open on optimus systems. With the card running on default voltage you can expect to reach 890->980 on the core, and 4300->5000 on the memory, depending on how good your card is.

With the default bios you are limited to only a 135mhz overclock on the core but are unlimited on the memory. The modified bios removes the 135mhz limit, as well as bios settings that reduce frequencies when power draw is high. Without these power limiting features removed, your core clock would be running below what you set it to when overclocking. Some clock reduction will still occur when the card is under light load and the power is not needed, but you will not have clock reduction due to the card working hard and drawing a lot of power. svl7 also increased the default core clock slightly to 758 since the default 719 is incredibly low for what the card is capable of. In regards to memory overclocking, it does provide a substantial benefit to the 680m, and 680m memory can be overclocked a lot. Many programs report memory clocks oddly. The actual default frequency of the 680m memory is 900MHz, but you will often hear of it being referred to as 1800MHz, and 3600MHz. This confusion arises because of the advent of DDR (double data rate) memory around 12 years ago, and the fact that GDDR5, the memory in your card, is not DDR, but QDR (quad data rate). When DDR came out people started saying that cards ran at double their actual frequency since it provided twice the bandwidth at the same bus width and frequency as single data rate memory. When GDDR5 came out not everyone doubled how they reported frequency again, even though GDDR5 gives double the data rate as DDR at the same frequency and bus width. Overclocking programs like msi afterburner and nvidia inspector use the old and inaccurate convention of reporting double the actual frequency, even though that number corresponds to nothing on GDDR5. So in summary the default speed is 900MHz, but effectively 3600MHz. Overclocking programs report it as 1800MHz. If you overclock +400MHz on memory, you will actually be running at 1100MHz, or 4400MHz effective. Higher voltage allows a card's core to run faster. Ex. at the default voltage of .987V my card can do 940MHz stable. With the voltage raised to 1.05V it can do 1019MHz stable. Voltage increases even with no frequency increase result in a V^2 increase in power draw. Ex. running 1.05V at the same clocks as .987V results in a 13.2% increase in power draw. Running higher frequency increases power draw linearly. Ex if you double the frequency while holding the voltage constant, you doubled your power draw.

It doesn't matter. expresscard 34 and 54 have the same connector.

Throttling is when a processor slows down from its expected speed. What you describe is overheating. It sounds as if your heatsink is barely touching the processor die.

Well I added a 470u cap to the 2 existing 330u. Made a slight improvement, but nothing like what you got clyde. It looks like it's the inductor swap that made the big difference. I wonder why a 1uH inductor is used in the 680m, but only a .56u in the 780m. I feel like the lower inductance inductor would be cheaper, so why was the higher inductance ever used on the 680m?

Well I had 30 470u 30V caps arrive today. I think I'm going to go higher capacitance than you 30V is overkill, but I got 30V so I could also put caps between Vin.

Hmm I guess I'm just special. I'll pop open my psu and see if it was modded for an m17x.

I couldn't find that anywhere for a decent price so I got the m18x psu. Was $44 with shipping. I'll PM him about the shutdowns.

Works. Thanks for linking.

I did for a 150em. Problem is the system still tells the psu to shutdown at 220W or so. That may be corrected later today by modding the current sensing circuit on the motherboard. My 330W psu was listed as used, but it looked new when I got it. psu cables still had all the factory creases in them. Hopefully someone didn't swap in a 240W chip to get it to run in an m17x. I highly doubt that though.

Good to see the gpu heatpipes bumped up to 6mm each. Core will also run much cooler due to the extra 7mm of depth to the radiator. Again I fear for the gpu memory though.

Nice! Unmodded 680m has horrendous voltage ripple then with such good results. Have you thought of going even higher on the capacitance? And did you replace the core caps as well?

Ouch. That is pretty bad throttling. I never saw your temps over 76C so it looks power draw related. You can probably fix it by looking up the power monitoring chip and modding the current sensing circuit. Is 1.012V the default voltage, or did you raise it?

I don't understand making the gpu0 radiator thicker, but then not having the radiator go to the end of the fan. At least the gpu1 heatpipes are now in the middle of the radiator instead of close to the top edge. Even after redesign the gpu0 pipes are still close to one end though. I expect gpu1 to now have better cooling that gpu0, assuming die contact is equal. And clevo had big enough radiators before. The problem is their atrocious die contact and gaps for air to pass around radiators. gpu0's air gap is now even bigger, and gpu1's fan blows the wrong way (air flow will be heavier on the end closer to the ends of the heatpipes, instead of the start closer to where the heat comes in). Die contact fix remains to be seen, but I am doubtful. I like the idea of having the backplate drop down so that the vents are all in the backplate, getting rid of the seam between chassis and backplate, blocking airflow. The problem is clevo has that bar anyway and added another for some reason. Vent pics aren't here, but they were posted on nbr. Also noticed that the gpu1 memory is now cooled by the same radiator as the core. That backside memory is gonna get toasty. In my p150em the memory radiator is much cooler than the core radiator. This looks like a great laptop if you're ok with spending hours modding it fixing die pressure, restrictive vents, and air gaps, voiding the warranty the day you get it. The thing is most people don't do those things.

If anything vapor chambers would be thinner since you wouldn't need the copper plate between the processor dies and the heat pipes. All a vapor chamber is is a fat heatpipe. I'm thinking of ordering a 12mm heatpipe and directly contacting my cpu die with it. 12mm should be fat enough that when flattened it's wide enough to cover the entire cpu die. same concept as the vapor chamber idea, but the vapor chamber can be even wider. As for the look again, I never liked the previous AW look. I still don't like this one either though.

IPS option! Definitely very interesting. Hopefully the response time will be good. What would truly be awesome is if the resolution was over 1080p. Of all laptops, SLI laptops with the top end cards make the most sense for super high res, and the fact that the m18x has the biggest laptop screen of anything just makes it seem silly to only offer 1080p. I'm really not aware of any use of expresscard slots besides eGPU use, and the m18x has no need of an egpu. I suppose you can have expresscard based SSDs, but I haven't heard of people getting them. The light up touchpad looks childish to me. It's like dell decided to tone down the AW look since many people avoided them simply for looks, but they kinda got stuck in the middle. Clevo is having a new approach where the same laptop will have a plain chassis, and have a 2nd chassis option with a flashier look. I think that is a better way to go. Hopefully cooling is good. One thing I thought of is that AW layouts make it possible to not use heatpipes at all, and just have vapor chambers the directly contact processor dies and the radiators. That would result in tremendous temperature reduction. Good to hear that AWs are still aluminum. I'd so much rather have a little more weight for the tremendous durability increase. And what's up with lady gaga and other chicks in those pictures?

They all have the same default clocks. What's different is the voltage. Default is .987V, while those are 1V, 1.025V, and 1.05V.

Did your splash screen not pop up twice prior to the modded bios? I thought it did for everyone.

2 cards? Sounds like you can spare one for me to have! I hope you get lucky and they clock like your 675mx. That thing was absurd.