New LG Gram 17

Great work, folks!

On my unit the fan is virtually always on when I am actively using the device, but it is virtually silent otherwise.

I did end up flattening the OS and reinstalling everything a few weeks ago. Since then, I've never seen 390MHz speeds - knock on wood!

It makes me think if my earlier install just lost it for some reason earlier, and whether it was a software issue after all - or, an outright infection of some kind of malware like a crypto miner or similar which, while invisible to the Task Manager, was consuming full CPU otherwise and nuking the performance of the device.

I have not been served any BIOS updates, but I did get the audio driver update referred to above. I have not messed with any manual fan control or under-volting.

Right now the fact that the unit runs fine without down-clocking to 390MHz is miracle enough for me!

 

Interested to see how your heat pipe mod works out. Do you just put one on top of the other with some TIM in between (LM?).

 

The heatpipe is attached using thermal glue around it with kryonaut tim between the two pipes. I yet have to figure out the max size that will fit and order one.

My bios is from 01/29/2019 and it is version W1ZD1180 X64.

The ZD in that version probably means this bios is for the ZD models which are the ones that come without OS.

 

I found that using the Throttlestop FIVR function "Disable and Lock Turbo Power Limits" after doing the "Install" flow next to it does work on the LG Gram17 and makes it possible to raise the power limit on the TPL dialog. However the stock cooling then just thermal throttles at around 15.8W.

I'm now using multiple runs of Cinebench R20 to check performance. I wonder how it compares to the i7 version. I'd guess performance really is the same as both are limited by the 15W barrier. My numbers thus far:

Stock: 1293
Undervolt 100mV: 1423
Raising Power Limit to 20W with Throttlestop: 1450

 

My settings:
Going to tweak these a bit and make another post with the numbers...

I also played around with the TPL and mine still throttles down to ~15W.

I've also noticed that when I'm playing a game Throttlestop reports that the max temp is in the mid-upper 90s. I never see it hit that during Cinebench, probably has something to do with CPU+GPU load.

I'm eventually going to buy a Sonnet 650W eGPU Breakaway Box and something like an RTX 2070, which will probably stop that from happening during games.

I'm actually pleasantly surprised at the games I'm able to play on this thing with a meager HD 620. I mean obviously it's not ideal by ANY stretch of the imagination, but it's better than I thought it would be. I'm even able to acceptably play something that has a little weight to it like Endless Legend. It's a little sluggish, but by no means unplayable at a decent resolution.

 

@Pflugshaupt Yours is doing much better than mine. The only difference between the two CPUs is the 8565 has 2M more cache (8 vs 6) and just differences in clock configs, pretty much everything else is identical. So if you can somehow keep higher clocks it should be the difference maker between these, the cache size difference is probably going to only mean something under certain workloads. It's also possible that with a smaller cache it might be easier to luck into getting a better undervolt as you have less of a chance to hit less tolerant memory.

I think this is all down to how good is the thermal solution in pulling and dissipating this heat. It could be that even with a top shelf TIM like Kryonaut your specific unit has better contact with the dies than mine, it could be these heat pipe assemblies are not very consistent quality and have a bit of a high variance in tolerances as well.

I'm pretty much convinced that if there was a better thermal solution in this thing it would definitely hold better boost clocks. Mine was actually throttling to LESS THAN 15W, according to ThrottleStop's pkg power monitoring anyway.

Results:

CPU core undervolt: -125mV
CPU cache undervolt: -50.8mV (it was unstable at -60mV so this is about as low as I can go)
turbo long power limit: 20W, 1536s
Cinebench R20 Score: 1300

I was able to pull a 1337 with a -100mV undervolt on the CPU cache but the laptop locked up right after finishing. I still got a lockup at -60mV, at -50mV it's been stable for around an hour now.

So, yeah to reiterate, I also think the whole problem here is the meager thermal solution, it's kind of obvious it's saturated, and perhaps some of the heatpipes are better at wicking the heat than others, or perhaps the fan tolerances are also different. It's also possible that your new BIOS has increased the fan RPM slightly?

I tried SpeedFan but it doesn't seem to recognize the management platform on these, though it's been some time since it was updated.

At any rate, feel happy about your current performance, it's outperforming mine.

 

Interesting to see you are using different underclocks on CPU core and cache. I always thought they mostly are the same. I am running -100 mV on both as -110 also proved to be unstable. Maybe this means I can go lower on the core. I wish there was a tool like the NVidia scanner for CPUs to automatically determine the stable offsets.

I do have an egpu box I tried with the gram17. It's an Akitio Node Thunderbolt3. Thunderbolt works great on the machine, but I quickly ran into CPU bottlenecks and heavy framerate fluctuations. Maybe tweaking speedswitch with Throttlestop can help a bit as that allows more stable CPU clocks.

My next step is trying how much difference using conductonaut LM TIM makes.

On my machine I was able to totally get rid of the power limit using Throttlestop as described here: https://bradshacks.com/matebook-x-pro-throttling - in the "Mod Guide". With low temps, the wattage then hit 40+ watts for a few seconds before running into thermal throttling, but this seems dangerous without more cooling. Brad on his site claims the heatpipes give 8W additional headroom on the huawei matebook x pro. If things would work similarly on the Gram17 I would be very happy. As it is now, every watt counts. Unfortunately I might have to wait for a few week to get the correctly sized heatpipe.

 

I tried Conductonaut now. After removing the heatsink I saw my unit had a case of too much TIM. I removed what they used and was prepared to isolate parts of the CPU die, but it looks like the CPU is almost made for LM, there are no exposed contacts and there's even a metal frame that might prevent spilling. Anyway, the Conductonaut was applied and I ran Cinebench R20 three times now. It looks like I gained about 2W of thermal headroom and my best Cinebench score is now 1517 cb.

 

I ran games now and am still running into thermal throttling. On CPU heavy games, the CPU quickly reaches 91 degrees and then frame rates start to become almost random.

 

Nice find on that Matebook Pro guide. That's the second guide where I've seen the Core/Cache voltages should be identically set and iterated. I tried that and failed miserably and the results were not as good as moving them independently. Is there a good reason why the voltages should be offset identically for core & cache?

It's good that LM helped a little! I have some conductonaut, but I'm loathe to use it on this unit as I am planning on carrying it around a lot. I put some on my EVOC in all three places: GPU, CPU lid, CPU die, so I'm no stranger to it, just don't want to do it to this one. It's pretty obvious that even with the Kryo this heat management system is fully saturated.


After undervolting everything, as I had not undervolted the iGPU yet, I settled on this as the best I can do for stability.
core: -125mV
cache: -50.8mV
iGPU: -50.8mV
iGPU unslice: -50.8mV

Still turning 1300, and over time the pkg power will continue to slowly decline due to the saturation of the heat management causing a slowly increasing throttle frequency.

One REALLY weird thing happened. I also unplugged in order to just test on battery to make sure that everything was still stable there as well. Since I'm not overriding SpeedShift in ThrottleStop i have the power plan revert to "better battery" unplugged. When I plug back in, the FAN won't step back up to full speed, even if the CPU is at 90, it just continues to slowly throttle little by little, I could hear that the fan wasn't spinning up even though the CPU was at 90. I had to power cycle to get the fan to be able to spin back up.

LG really needs to afford us some more direct control over the fan.

In some interesting news, I decided to just blow a fan I have right at the back of the laptop where the two vents are. Pulled a 1420. I noticed that some of the CPU cores were hitting 90 easily while others were hovering at 84-85. It seems that perhaps those cores are not getting good contact with the plate attached to the heat pipe. I may lap the plate just a smidge and reapply Kryo again to see if that makes any difference. Falling short of that the only thing I can think of doing is using a shim.

EDIT: I found this die shot of a Whiskey-Lake U and highlighted the 4 cores. I did some more investigation and In throttlestop the hyperthreaded cores that are always warmer are 0,1 & 4,5. So those were probably either two across the top row or two in a column. It's pretty obvious based on that the plate is not sitting flat against the die and two of the cores are not getting good contact.


I went back and looked at the shot I made of the heat pipe contact plate, it does look like it has some slight imperfections, could use a lapping. I'll order some lapping material and see what I can do. I think if I can solve this I might get a better result. I suppose it's also possible the plate is not fully attached to the pipe and there's an imperfection there perhaps right where the cores are. Probably not though.

 

Great findings! Bad news about the fan.

I went back to Kryonaut now as I also don't like LM in my laptops. Cinebench went down to 1473, but that's not too bad. It's sad to see how easily LG could have made this machine much more powerful by just adding a larger fan or at least bigger fins on the fan.

I found a register in the EC that controls the fan using RWEverything. Offset 0xC2 goes from 0-5 depending on fan speed and if the machine is idle (fan off) it can be manipulated. The absolute max speed seems to be 7, but the system never writes anything higher than 5. Once temps are high the register gets constantly overwritten, but at least it's something. Unfortunately even at 7 there is not much air going through that fan. I tried creating a Notebook Fan Control config with this, but so far no luck as NFC doesn't want to launch for me Maybe later today. In any case this won't help too much as the max speed for this fan is just too low, all it can help with is running into thermal problems a few seconds later than now.

 

I made an xml config file for Notebook Fan Control. It sort of does work as it allows fan control. However sometimes there's a fight between LGs software and NFC. Most of the time NFC wins, but not always .

Hmm.. maybe if I set the fan to silent on the LG control center..? Nope, that totally disables the fan..however it makes me think that maybe there is a way to make the fan spin faster. It appears to be limited to level 5 of 7 levels currently.

 

Wow, great work!

I ordered a new heat pipe assembly from eBay. Funny it's the same exact part number in the 13" model. It seems like they are using this same one across all their Gram models. What I'll do is see if this new one contacts the die better and use that one if so. I was thinking about then using the extra heat pipe to mod with. It's a 5v fan so I could probably tap into USB power to just switch it on and off. At the very least the extra heat pipe would probably afford some extra sinking if I can't get it to safely fit in the chassis. Imagine a dual pipe fan in this thing perhaps I could attach it and use a different sightly smaller fan in that empty area to the left of the existing fan. I found quite a few fans online that are close to the same power requirement of the existing one.

This is probably going to be a little destructive modding. But it looks like at least attaching the pipe and fan wouldn't require any permanent changes. If it does fit, creating a vent would.

 

Incredible work, people!

Ironically at my end after my last post I've started seeing CPU slowdowns again.

Fortunately the slowest I have as yet is 890MHz, which - believe me - is a huge improvement over 390Mhz.

I am all too happy to admit that at this time I have sort of started reliving the glory days of the LG LW70.

I am excited to hear more about your fan mod and maybe combine it with the Notebook Fan Control XML profile!

Please keep up the great work.

Thank you!

 

msintle, are you using Throttlestop? It should be able to tell you what's holding your CPU back on the limits window and maybe allow you to override the issue. Did you already do a repaste? Mine had ridiculous amounts of TIM.

I ordered my extra heat pipe and it should arrive in a few days. I'll keep you posted about any progress. I do have a fully modded Matebook X Pro as described in the guide I linked to and I have a few ideas to try once my order arrives. My matebook x pro is able to sustain 36W under full load before it runs into thermal throttling, the LG Gram 17 only is able to do 17W. I hope to reach 25 W somehow - which would improve performance quite a bit.

hfm, I tried using the heatpipe from an extra assembly on a macbook pro but found it impossible to detach the heatpipe itself from the other bits of the assembly. Therefore the heatpipe I'll use is a straight model. 150mm x 12mm x 1mm and able to transport 25W under ideal circumstances. I'll have to modify the fan a tiny bit, but I have a feeling in the long run I will try to install a different larger fan anyway (if I can find a suitable candidate). Before doing that I'll try to measure the fan rpm and whether it running at full spec. It feels like it could run faster, but is limited by the software. I wouldn't mind more noise under heavy load, but unfortunately, LG does not give us that option.

 

No, I haven't done a repaste, or installed ThrottleStop. Please keep us posted with your progress. I am looking forward to hearing all about it!

 

Ugh...any time I seem to think the machine has stabilized, a new issue seems to be cropping up.

New issue of the day: Connect an HDMI cable to a TV. Device immediately blue screens with an Intel driver file referenced. Driver stock. OS installation stock.

Has anyone else run into this before? Do you know of any updated drivers that might work? I found a thread from 2017 where users were experiencing a nearly identical issue. Unfortunately the legacy driver they referenced to fix the issue wouldn't install on the Gram 17.

Do people not even test this stuff anymore? Its utterly ridiculous to have an HDMI connection blue screen the entire OS the moment you plug it in.

Don't know what else to say, really...

 

I just tried and am now typing on my TV. No issues for me. My intel GPU driver is: 25.20.100.6325 from 9/21/2018

 

My version is 25.20.100.6471 from December 2018.

@Pflugshaupt - You had the Korean model, right?

 

Just tried to install the latest driver from the Intel site. Setup bails saying it's not been validated for this computer. There's a workaround to force an installation, but can anyone with the US model let me know if they're seeing the same issue first?

 

Yes I have the korean model, but I doubt there is any internal difference besides the i5. I updated my igpu driver using the Windows device manager and now I'm at 25.20.100.6519 from Jan 9th 2019 and still no issues with the hdmi port (at least not with my TV).

 

Sure would be nice if they shipped the i5 Korean model here with US keyboard. They'd probably sell a lot more.

 

First I put an arctic thermal pad (2mm) on top of the SSD to get rid of SSD throttling.


The I ran the AIDA64 system stability test for SSD only and got this:


So that does take care of the SSD although it leads to a hot spot on the bottom of the case in case there is heavy SSD writing.

 

Then I went crazy and put a graphene thermal spreader sheet I had left from the Matebook X Pro project and glued it to the bottom and added 2mm thermal pads to the CPU heatpipe. This makes the bottom case into a giant heat sink.



Now the performance improved a lot! However once the laptop reached thermal throttling, the bottom was almost too hot too touch, so it's not a great idea unless the fan also improves. Using this temporary mod I was able to Cinebench R20 with very little thermal throttling, the i5 CPU stayed at 3.7 GHz all-core turbo for the most part and power draw was around 30W. At the end the bottom of the case probably reached almost 90 degree celsius.. not good for the lap. However a laptop cooling stand would blow air at just the right spots, I just don't have one right now.


The highlighted run is a 25W run I did later, the 30W is the one above and the score is a whopping 1809 cb. These CPUs are beasts if they are not held back by power or thermal limits, so it is worth modifying these laptops. One additional thing I noted was that the factory fan curve worked a lot better with more heat capacity. It's almost as if it is controlled by case air temp and not cpu temp.. So once everything got quite hot, it started blowing full tilt, but that was a long time before reaching the cpu thermal limit.

Next things I'll do is trying to add some more heat fins to the exhaust to improve the fan performance and then hopefully my extra heatpipe will arrive and I'll put that in. I also ordered more graphene and might just cover most of the bottom case with it. Unfortunately it is costly, but I think the magnesium alloy used in the gram is not a good heat conductor, so it does help spreading the heat around.

 

Wow that's amazing! Very impressive work.

At my end I already have that blue thermal pad on my SSD, which probably has ensured that I wouldn't be reporting any issues in that regard.

Unfortunately my HDMI is still blue screening the laptop. Is anyone with the US model able to test for this issue? I did update to the latest driver using the Driver Manager update method described above...but the same issue remains. Moreover, that post I'd found from 2017 indicated that the issue was fixed by using an older driver, not a newer one (so probably I'm out of luck there).

 

I force installed the latest driver last week. Once you do that, you can install the Intel provided drivers from that point forward.

 

I bet the fan would be able to cool things a little better if there were a vent hole below it so it could pull in cool air, the problem would be that the laptop also doesn't have feet to speak of lifting it up far enough so that air can be pulled in... I wonder if creating a duct from the fan intake to the other vent would help. I have a sneaky feeling it's just mostly pulling hot air from the case and blowing it across the heatsink. There is a little but of ducting for the exhaust itself if you notice half of it existing on the bottom cover exhaust vent. I suppose it's supposed to pull in air from the other vent since that's the only intake opening that exists.

I'm waiting with baited breath to see what adding another heat pipe does. It looks like thankfully the fan housing is metal so at least that surface area can be used.

Also, I ordered some of these Coollaboratory Liquid MetalPads for Notebooks which come with shims you are supposed to use with them. One layer of MetalPad, Shim, then another layer of MetalPad placed on the die, then reinstall the heatsink. I bet these will fix my poorly performing heatpipe connectivity. It definitely will not have any issues finding the temp to cure it! I read some reviews of these and they are all positive, supposedly the only thing that beats them is real liquid metal.

Those numbers you got with the pads to the bottom cover are insane. Unfortunately there's no way I'm going to do that if the bottom of the laptop actually reaches 90c. Whew boy! It looks like the best thing we can probably do is look for a better fan that would be able to fit where the old one is. I'm also a little suspect of the VRM stages in this thing, do we think we can sustain 25-30W to this CPU for extended periods? It definitely doesn't seem like it was designed to do that based on the heat management and software. Maybe they were just super conservative with this thing so they could say it wieghs less than 3lb and it stays quiet. Tradeoffs...

Or... LG could release an app so we can manage it!

 

I think it would be a bad idea to add vent holes. In my opinion the air has to move in at the spot you described so the other components like vrms and the ssd get some cooling. I believe the air needs to move a little in the chassis. However improving the air channeling would probably help and I did a first step. I think my second heatpipe will also help with the airflow as it will block the air from going to the fan directly. In order to cool the whole motherboard the air should go around the CPU and the heatpipe, then enter the fan and pickup as much CPU heat as possible from the fins so it doesn't just heat up the air inside the case.

I ordered some additional heat fins from china, but no one knows when they will arrive, so I tinkered with some SMD coolers and made some ugly extra fins I then attached with the last graphene tape I had left.



I added some vinyl tape to make more air go through the fins. All this could be done much better if the fan and heat sink were removed, but I save this for the time I'll add the 2nd heat pipe as that will require disassembly anyways.

However there is an improvement already with this nasty looking mod. I am now able to sustain 20W during cinebench and the machine keeps running at 3.0 ghz. leading to a 1636 cb score.

Having a larger fan would probably be amazing and there is so much room for one in the chassis. The tricky part is finding a flat one, the places that sell notebook parts I have looked at so far unfortunately do not provide fan height measurements.

 

Just an information for others. I just upgraded my Gram 17 to 40GB RAM (using one 32GB DIMM) and to 3TB NVMe SSD total (Samsung 970 PRO 1TB in x4 slot and Samsung 970 Evo 2TB in x2 slot). I'm also dual booting to Windows 10 Pro and Fedora 29, everything works fine

 

I pondered upgrading mine to 24 GB, but decided against it as using 2x8 = 16GB RAM is the only way to have it run in dual-channel mode for maximum performance.

 

Yeah, unfortunately LG did not just provide two DIMM slots for this thing. Not having dual-channel RAM is a deal breaker for me as well. I think the 17" model is suffering from them using the same exact basic PCB design for all their Gram models. There's so much extra room to make some slight improvements that really wouldn't add any extra weight. It doesn't have to be much, just a little tinkering on their part would put it over the edge. I do understand though, they are saving quite a bit of money re-using all these parts. But at $1699 US without even having an NVMe drive default it's slightly overpriced. I think they are taking advantage of the fact that there really isn't any other laptop that can compete with the Gram 17 right now for what it is and does and how light it is.

 

During the upgrade of all the parts (ssd, ram), when I saw how bad is the thermal management of this little light baby (single tiny fan), I don't care if I have only single channel memory, only x2 pcie etc. It still great machine overal for my purpose (coding on the go). I also like the plain looks of the machine and its name "gram". I also like how its case flexes , during the upgrade I dropped some screws on the case and the case just flexes and spring screws back to air, no damage You just need to know you are buying a machine full of compromises and to know what are its strong and weak points

Sent from my Pixel 3 XL using Tapatalk

 

regarding the single/dual channel ram access, I'm not expert in this area but I think if you add additional 32GB DIMM to the soldered 8GB RAM, 16GB (2x8GB) will operate in dual channel mode, the rest (24GB) in single channel mode. It's still better than 16GB only, isn't it? (If you need more than 16GB of course)

 

regarding thermal throttling, I'm using Dell XPS 13 (9370) and this machine also throttles as hell, even during just booting to fedora linux. If you need a machine which doesn't throttle like this you need to buy something substantially heavier...

 

Yeah I'm not really complaining I love the notebook, I just see where it could be improved slightly. The XPS 13 throttling I would expect since there's literally zero room in there to do anything about it unless you want fans that run at 52dBA. It really would take a minimum of changes to make this unit throttle less. The room is there. But they are saving a lot of money by just reusing the same heat management system as all the other smaller SKU in the Gram line.

 

Regarding ram: That would only apply if the ram was running in flex mode and I don't know if the motherboard/uefi support that. I guess HWInfo64 would display "flex" in that case?

You could also measure ram speed in aida64. dual channel leads to around 30 gb/sec.

Regarding light/throttling: There is no reason a light machine shouldn't be able to get rid of 40W of heat. However I guess intel limits these CPUs on purpose to 15W so they don't perform like the higher end CPUs. With improved thermals and bypassed power limits it's possible to get a machine that can perform like a mobile workstation and still save battery like a champ if needed. It's been demonstrated with the huawei matebook x pro by brad from brad's hacks and I think the LG Gram17 can do it as well, especially with all the extra room available for all kinds of mods.

 

Hwinfo64 shows my 40GB RAM is running in dual channel mode. I interpret this as that first 16GB is running in dual channel mode 2x1200 and the rest (24GB) in single mode 1???MHz

Sent from my Pixel 3 XL using Tapatalk

 

good for you! I guess that means 16 gb of dual channel speed + 24gb of single channel speed and in that case it is using flex mode. Flex mode has a bad reputation from the past, maybe it has been fixed with newer CPU generations. Can you try a memory speed test like the one in AIDA64?

 

for sake of clarity, I'm using Samsung 32GB DDR4 SODIMM 2666 MHz Laptop Memory Module/Stick

 

I get similar numbers for memory with my 2x8 config. However my cache numbers are like 1/3 better throughout which seems kinda weird. But anyway good to know that I could upgrade to more memory in the future without suffering from single channel speeds.

 

I ran another memory test with the tdp totally unlocked and I guess that makes a big difference for the cache tests. Now cache numbers are even higher.

It's interesting to see you are using the same BIOS I have.. so it does exist for the US model.

 

Have you been able to verify that HDMI works when attaching to a TV direct from the HDMI port in the laptop?

 

How do you `totally unlock` the TDP?

Where would one get that latest BIOS for the US model, then?

 

That's almost the identical configuration I am running.

I do have 40 GB RAM, which is vital for my work.

I was unable to get the SSDs to work though (trying 2 GB in the 4x slot and 1 GB in the 2x slot) because my SSDs were one (or two) generations apart, and the UEFI/BIOS auto speed negotiation (presumably) failed, resulting in three successive reboots and then a total shutdown, without the device ever booting into Windows.

As I wrote earlier, if one of the SSDs was totally uninitialized, you could boot into Windows, and then even initialize and use the SSD; only to have the UEFI/BIOS lock you out on your next reboot - how frustrating!

If anyone knows where I might get the latest BIOS, that'd be great.

 

To fully unlock the TDP you need to use throttlestop and install an extra driver on its FIVR page and then configure things on the TPL page. However as long as the thermal situation is not improved, the unlocked TDP won't make a big difference as the machine overheats immediately anyway. I have it unlocked so I can see if my mods make a difference.

 

You might have an issue where the drives are sharing a gptid.. I've seen Windows lock up like this after cloning drives or other tasks. Can you boot from a recovery USB and completely wipe the secondary drive out with diskpart?

If you don't have one you can create one using the free version of macrium reflect.

 

Thanks for that but the lockup is in the UEFI BIOS, and not Windows; the latter of which works perfectly well. Its only the buggy former that reboots thrice (without ever passing control to Windows) and then shuts down.

 

Oh, nevermind then.

Maybe we need to find that new BIOS, and I'd love to know why none of my utilities are exposing that an update even exists.

 

Yeah, same for the Intel GPU update. I've downloaded the latest GPU driver and will forcibly install it, hoping that it may solve the HDMI issue. We'll see. Did you ever manage to test the HDMI issue at your end?