win32asmguy

Recently I decided to attempt a display swap for the Legion Pro 7. My intention is that the standard WQXGA panel while nice for applications that support display scaling, is not really useable for me at 100% display scaling. Even worse under Linux where using Nvidia with Xorg does not fully support GPU scaling. Some other systems do offer lower resolutions but either have poor cooling solutions, high pitched fan noise, or are not available with the 13900HX or 4090. My target panel was a BOE NV160WUM-NX3 V8.0 which is a 16 inch WUXGA, 165hz, 100% sRGB, 500nit panel which was used in the base model Asus M16 2022. Unfortunately no GSync support for this panel. It does have the same 40 pin connection, with the same connector location and pin pitch. Also electrically its backlight uses the same voltage range so the brightness steps used for the stock panel work as expected. For the swap, the initial step is to remove the plastic display bezel. The back of the bezel is completely covered in adhesive tape. I used the iFixit Halbard Spudger to work the bottom center free enough to begin slowly pulling the rest of the bezel away while continuing to use the spudger to separate the adhesive as needed. After the bezel is removed, the next part is the metal hinge cover. I used the blue iFixit Opening Tool, starting at the right edge. I was trying to use the tool to pull on the hinge cover while not pushing/leveraging against the display circuitry as it can damage the pcb or even crack the display panel. After the hinge cover is removed now the double-sided tape under the display must be pulled off. There are two small tabs at the bottom of the display that can be slowly pulled to release the tape. After the tape is removed the display can be rotated out of the a-cover and rest on the keyboard to remove the display cable connector. There is just a piece of tape and then a metal latch securing the cable to the display pcb. I still need to pick up some replacement tape to secure the new panel to the a-cover. I will use Tesa 61395, and because this tape is not an L shape I will have to put a horizontal piece of the that is not adhered to the panel so there will not be a gap between the panel and housing. The vertical strips of tape should be enough to hold it in place and be removable via the same method if I need to swap it back in the future. The bezel should be reusable as most of the adhesive is intact and its shape seems to be correct. It is a $47 replacement part from Lenovo if you want to make it look like new. Some other people in the Lenovo discord are also experimenting with WQUXGA (4k) panels and miniLED panels with some success.

The steps should be correct but maybe I missed something. Look at step 5 in this guide just in case. https://brendangreenley.com/undervolting-2020-dell-laptops-like-the-vostro-7500-and-more-tips-to-improve-thermals-battery-life-and-speed/#cpu-undervolt

Here is how I modify those settings: Get grubx64.efi from https://github.com/XDleader555/grub_setup_var/releases On a small USB thumbdrive, formatted FAT32, change its file name placing that file in a folder location of drive:\EFI\Boot\bootx64.efi In the bios, disable secure boot and enable overclocking Reboot, selecting the thumbdrive At the prompt, enter "setup_var CpuSetup 0x381" which should show a current value of 0x1 (undervolt protection enabled) Again, enter "setup_var CpuSetup 0x381 0x0" which should update the value and confirm the operation succeeded Ctrl-Alt-Delete, go back into Windows and now undervolting production should show disabled. RU.efi also works but I have found it to randomly freeze or not actually apply changes so I prefer using the modified grub bootloader instead. It is also pretty easy to clone the github repo in linux and build it yourself if you are concerned about malware or such.

You might also try the kernel params "iommu=soft" or "dis_ucode_ldr" to see either allows it to boot. I would be curious if the same problem happens with the Asus or Lenovo models using dragon range chips. I was curious about Windows 10 just because sometimes compatibility is not gauranteed and it is what I would run if I had one. Luckily the m18 Intel model has full support for Windows 10 22H2. Last year I tested the Asus Strix Scar 17 with Intel 12th Gen HX and it did not fully support Windows 10 and was missing many critical drivers, for the USB controllers for instance. It also was not able to easily boot Linux on it either - the bios had some kind of arcane grub2 memory mapping issue.

The 7740 issue started occurring after the microcode update was pulled in with routine updates to the distribution. In general you could try booting with "acpi_osi=! acpi_osi='Windows 2022'". Are you booting with the MUX set to hybrid or dedicated mode? Also curious if you have tested Windows 10 on the AMD M18 R1 variant.

I tested cold / hot booting Fedora 38 (kernel 6.3.8, nvidia 530.41.03, xorg, bios 1.8.1) with battery power under 60%, no issues. I did have to set nvidia-drm.modeset=1 to get rid of a phantom display but that issue occurred regardless of being on battery or plugged in. I do have secure boot disabled and no battery charge limiter set. I have had a similar issue with the Precision 7740 in the past (would not boot linux on battery). The issue was a bad Intel CPU microcode update bundled with Ubuntu at the time which had to be disabled via a kernel flag.

I can test fedora 38 tonight, but mine is the Intel model. I will let you know if I can replicate it. I take it you have also installed to an SSD but it still cannot boot on battery under 60%?

I am thinking about trying to mod the bottom panel of my M18 to remove the stock honeycomb slats over the fans, so it removes the high pitched fan whistle. If I run the system with the bottom panel off, there is no whistle at any RPM. I also tried switching heatsinks so I think it is not the brand of fan causing the issue. I can pretty easily order a replacement panel spare part from Dell if I mess it up. The hard part is finding some replacement mesh to use instead. Any good sources for something like that? Most of the metal mesh for car related mods (custom grills, etc) seems to be 20GA or more which is too thick for this application I think. I also thought I might be able to find an old used bottom panel from ebay and use its mesh as a donor.

Interesting results. I did not expect my 12900K to be that far behind yours, or maybe even ahead at 4.2Ghz all core. I had tested with LLC at 1.1 but in order to get 21,500 it had to be on a cold boot with ambients at 65F or lower. I heard that 14th Gen is coming to LGA1700. Shame that we cannot have compatibility with that via a bios update here. Missed opportunity. Every day I talk to tuners working with these new BGA-only models and it really isn't comparable to what we used to be able to do with modular CPUs.

Sadly dynamic undervolting is still disabled by default in the new bios version, so you would still have to manually set that flag via UEFI setup variables. This bios does have a build date of 5/29/23 but maybe they did not have enough time to integrate that change and validate it. However now I can train and boot 6400 CL38 1.35V via selection of XMP profile 1. It also appears to disable SAGV with XMP active now. Before on 1.7.0, any type of memory change would end up with SAGV re-enabled which could have been why it could not train and boot. There was a similar issue on the Legion 7 Pro that the community discovered earlier.

A new bios was released this morning, 1.8.1. Related to XMP memory support according to the notes, but maybe undervolting protection is disabled now too. Time for me to test the Kingston 6400 CL38 1.35v memory again to see if they can boot.

64gb will operate at 5200mt/s per Intel spec as it is dual rank. But if it supported XMP timings it could achieve 5600mt/s at higher voltages with adequate cooling.

Dissapointing about the performance, but not surprising. The only advantages I see from this new model compared to competitors is three 2280 M.2 drive slots, easy to replace keyboard, non gamer accented chassis, and if system76 gets their open firmware and ec working it could be more tunable with some work.

It make sense if it were from the 3500 Ada and up, as the same heatsink should be compatible 4000 Ada, 5000 Ada and even the RTX 4090 mobile option.

Yeah, it was like this from the factory. The previous m18 I had last month also had Delta fans and the same noise. I would like to try a version with the Sunon fans but the challege is getting the part number and ordering it because the reps do not usually give out this information or allow selection of a specific brand when getting a warranty replacement. Usually spare parts sales reps will sell you the part if you already know the number. Attached is a picture of the Delta version heatsink. The part number sticker is located under the rear IO cover.

Does anyone here happen to have an m18 R1 with 4080+ and the Sunon or FCN fan heatsink? I am troubleshooting some fan whistle (any RPM, louder as RPM increases). Mine has the Delta fans, part number 27GFC for the whole unit. Also its possibly an interaction between the bottom panel venting cuts as the whistle is not present with the bottom panel removed.

@iieeann I would suggest resetting to factory defaults in the current bios and then trying again. If it still does not work maybe try removing bottom panel, disconnecting the battery and then holding the power button for 30 seconds (they call it draining residual flea power). Dell does have this "force a bios update" process, but ideally the normal process would work first: https://www.dell.com/support/kbdoc/en-us/000134938/forcing-a-bios-update-without-the-ac-adapter-attached-on-a-dell-laptop

I tested both stock Windows 11 Pro OEM install (by Dell) and also a clean install Windows 10 Pro which basically test the same for GPU score. For CPU score Windows 10 Pro is indeed superior. Do you have VBS/HVCI disabled in Windows 11, and Turbo Boost 3.0 disabled (allowing GPU Dynamic Boost) in the bios, and GPU Memory ECC disabled? Those could potentially effect performance. You would probably want to run the test while recording GPU Power used, to ensure its using 135-140W under load. There is a small variance in silicon quality but I would suspect that at the same TGP and temperature an RTX 5000 Ada should always be higher performance than a RTX 4000 Ada. I will say that I would not be surprised if there is a Dynamic Boost bug as those have existed in the past models, 7770 for instance with 3080Ti could not go beyond 130W without overriding power limts using certain drivers, despite being advertised as "150W TGP" in Nvidia Control Panel.

Ok, I think power is limited somewhat in balanced / optimized, so you might want to try "Ultra Performance" and see if it improves. I would expect your system to get around 17k in Timespy Graphics at around 135-140W. Also if possible you could contact 3dmark support and ask them to whitelist the RTX 5000 Ada GPU as it is new and hidden by default. You can check if they fixed it by opening the link you posted in an incognito window.

I would say it is not improved in that regard. It uses the same cooling solution and fan tables. With the system repasted, and the performance mode set to "Cool" to force the fans on, the CPU is around 45-50C and keyboard I guess is around 30C, with 21C ambient temp.

Yeah my Timespy results are running default settings. You may want to monitor temps on yours to see if it reaches 87C on the GPU which will trigger short throttles which can lower the score. https://www.3dmark.com/spy/38088494

It has been a few days and so far it seems like CBR23 Multi has gone up slightly (best seen was 29,900) and CBR23 Single can now sustain 5.5ghz, scoring 2150. Also Timespy Graphics scores 16,600 which is up a bit as well. I also got the bios dumped, and tried to do some undervolting which was not successful. The "undervolt protection" remains enabled regardless of the value set to the hidden UEFI variable similar to the experience reported on the XPS 17. Setting voltage offsets for P-cores also did not have any effect. I also tried modifying AC/DC LL to both 1.35 which did not boost scores any.

The other 4 SODIMM systems all perform a bit worse if you are only using two SODIMMs. Basically they operate at 4000MT/s regardless of what faster speeds are supported, with the exception of the GT77 where with the advanced bios you can override the behavior (but also greatly increase boot time). With the Precision 7780 and the CAMM to SODIMM interposer it can boot 64GB Kingston Fury at DDR5-5200 CL38. The same motherboard can also take a 128GB CAMM module if capacity is needed over speed. Of course, it is probably a moot point in terms of performance uplift because this system is also limited by its small cooler, lack of undervolting support, low GPU TGP, 240W power brick, and lack of fan speed control.

Yeah more CTO options would be nice if they have the chip supply to do it. I could see them adding 13980HX or 7945HX with 4080 in the future. As far as pricing I wish I could pay them more to send the GAN 330W power supply, and also to not apply Element 31 to the high end CPU and GPU options. They should also add some bios options to disable the rainbow RGB bootup effect, disable the rear light ring, and set the keyboard to white-lighting only.

Today they added the 13980HX + 4090 option as well.