Acer Swift X (2021) - Owner's Thread

As someone who travels for a living, I love this laptop. At 3.06 lbs it is the lightest gaming laptop you can buy. It's not marketed as such, and it only has a 60Hz screen but if you need the lightest laptop possible that can still play games this is it. The best part is that with 100w PD I don't even need to bring the power brick and can have one charger for all of my devices.


Specs:

• CPU: AMD Ryzen 5800U - Zen 3, 4.4 GHz boost, 37W sustained, 48W boost
• GPU: Nvidia 3050ti - 40W TDP
• Memory: 16GB, 4366 Mhz, LPDDR4X
• Wifi: Mediatek 7920. I replaced it with an Intel AX200 with no problems.
• Screen: 1080p, 60hz, 350 nits (mine was not overclockable)
• Storage: two NVME gen 3 slots. Comes with 512GB SK Hynix drive.
• Power: USB-C direct passthrough power delivery up to 100w (tested with watt meter). It might go higher but my PD charger only goes to 100W. It's limited to 90-93W when the battery is charged but it can go higher when charging the battery.

I’ve spent a lot of time tweaking and modding this laptop to get the most of it. Here is what i have done and figured out so far.

Spoiler: Fan control

NotebookFanControl does not work with zen3 processors. The service just crashes and restarts continuously. ECprobe, which comes with NBFC, does work though and after playing with it for a long time I was able to determine some interesting registers.

• 020: fan speed - read only
• 016: corresponds to acer’s “Quick access - System Usage Mode”
  • 000 = normal
  • 002 = silent
  • 003 = performance

Writing to register 016 can be used as a roundabout way of controlling the fan curve but even performance mode tops out at 13 (according to register 020).

Then I remembered that when I did a bios update the fan sounded like a jet engine and figured there must be a register that controls it. I wrote a script to cycle through every register and write 0-255 to it before moving on to the next one and I found it.

• 034: on/off switch for max fans when you write value 008. Register 020 shows a value of 20. Almost twice as fast as performance mode and allows the laptop to maintain 90W combined CPU and GPU without thermal throttling.

I have included a script to toggle max fan mode.

Spoiler: Thermal Pads

One of the reasons this laptop is so light is because it only has one fan and heatsink. It does a pretty good job under normal load, and while gaming on battery, but when using all 90w it gets a little toasty. I added some thermal pads to make contact with the magnesium chassis to spread the heat. It seems to help and the bottom doesn't really get hot enough to burn you, but it would be uncomfortable on your lap.

I added 3 mm directly on top of the CPU (you will have to remove the plastic “screen” glued to the chassis) and 2mm on the heatpipe above the fan. I think if I did it again I would add a bigger strip above the CPU.

Spoiler: CPU Control

Intel laptops are lightyears ahead of AMD when it comes to cpu control with XTU and throttlestop. Ryzen master only works with desktop CPU’s but luckily there is a program called ryzenadj that lets you adjust all sorts of things.

• STAPM LIMIT - total package power limit (W)
• PPT LIMIT FAST - boost power (W)
• PPT LIMIT SLOW - sustained power (W)
• THM LIMIT CORE - throttle temperature
• STT LIMIT APU and STT LIMIT DGPU - I’m not quite sure what these values correspond to but I know they are related to temperature. At the default value of 40 the CPU will throttle pretty aggressively starting around 65 degrees. Changing them to 50 allowed the CPU to reach 95 degrees but that can be controlled by setting THM LIMIT CORE to whatever max temp you want.

The process is very different from tuning intel laptops, choosing temperatures and powers vs clock speed and voltages, but it still works. Unfortunately, it's all command line and not very user friendly so I wrote some scripts to set different values depending on workload.

• AC gaming - sets 25w sustained, 30w boost, 85 degree throttle temp, STT limits to 50. Changes the fan profile to performance.
• Battery Gaming - same as AC gaming except 15w / 20w.
• Max Battery - sets total package power to 5W, sustained power to 3W, and fan curve to “silent”. With the 3050 ti off i was still able to watch 4k video without problems.
• Default CPU & fans - resets all default values. Useful if you need full CPU power.
• RyzenInfo - Shows all ryzenadj values on a 1 sec loop. Useful for seeing what you computer is doing.

Spoiler: Battery Life

The default scheduler for this CPU is pretty bad for battery life. At idle it will sit between 15-20w. But if you use ryzenadj to set the max package power to 5w, the entire computer will draw about 10w max giving you at least 5 hours on battery.

That assumes that your 3050 ti is sleeping like it should. I don't know if it's all the tinkering I did or its a problem with the chipset / driver / etc.. but my 3050 ti either wouldn't sleep at all or would alternate between idle and sleep constantly. Even with no applications using the dgpu according to the nvidia usage icon in the taskbar. Idle power for the dgpu is about 12w which will destroy your battery life. I included scripts to disable / enable the dGPU in case anyone else had the same problem. Just disable when on battery and enable when you need it.

Note: MSI afterburner will ping the GPU constantly keeping it idle. Exit the program or pause monitoring when on battery.

Another problem I had is power consumption during sleep. Again, I don't know if because of my tinkering or the computer in general but the laptop doesn't really sleep. The screen turns off and media will stop playing, but it still draws 20-30w from the wall. Playing around with the unlocked bios I noticed some settings for “modern sleep” selected with “S3” as the other option. I will play with these and see if it solves the problem. In the meantime, setting the power options to hibernate when you close the lid seems to do the trick.

Spoiler: Shunt Mod

I know, it sounds ridiculous to shunt mod an ultrabook but I noticed I was only pulling around 80w in games so I figured there is at least 10W of headroom for some extra GPU power.

If your interested these are the resistors you need to shunt. The two R005’s just below the GPU VRMs (R22).


All you need to remove is the wifi card to access them. Also disconnect the battery any time you plan on touching the motherboard. To add 10W you need to add two 1206 R20 (.02 Ohm) 1/2W resistors on top of the existing R005 resistors. I used these: https://www.digikey.com/en/products/detail/delta-electronics-cyntec/VSRP1206SD-R020F/9762072

Materials needed:

• Soldering gun with fine tip
• .015 solder
• No clean flux
• Fine tip swabs
• Desoldering braid just incase

Instructions:

• Clean everything you plan to solder with alcohol.
• Attach some tape, sticky side up, to a table to hold the R020 in place while you tin the sides.
• Apply some flux to the soldered ends of the R005 resistors.
• Place the R020 with tweezers on top of the R005 and apply some pressure to the top to hold them in place.
• Touch your solder tip to both ends to flow the solder and create the bond. You might need to add a little more solder depending on how much you tinned them.
• Repeat for the other resistor.

Results - wattage read from the wall with a watt meter.
Overclock for all tests: 250 Core, 550 Memory

Unigine superposition:

• 40W: 2434, 57W
• 50W: 2873, 76W

3D mark Timespy:

• 40W: 4060 graphics, 59W
• 50W: 4498 graphics, 77W

3D mark Firestrike:

• 40W: 10823 graphics, 3900 combined, 59W
• 50W: 11589 graphics, 4519 combined, 77W

​

After the shunt mod my GPU temps still never went above 75. Over all, I am extremely happy with the results.

Spoiler: Bios Mod

I worked with Dudu over at bios-mods.com to unlock the bios to access AMD PBS and CBS settings. This allowed me to adjust all sorts of things including the infinity fabric frequency which is currently set to ¼ ratio (1067 Mhz). My laptop couldn't quite handle ½ ratio at 2133 Mhz stable but I was able to get it to 2100 Mhz and lower the memory clock to 2100 MHz (4200 DDR) to get ½ ratio. My latency went from 110 ns to 89 ns and my read speeds almost doubled in AIDA64 memory benchmark. Should provide an extra 5-10% boost in games.

There should also be a way to undervolt the CPU in the bios, I'll look into that next.



If you want to mod your bios don't update to 1.04. I did and had an extremely difficult time getting it to flash compared to 1.02. Could have just been a coincidence but I would stay on 1.02 to be safe. Send dudu an email at [email protected]. He will ask you to dump your bios using software and send you a modified one to flash. It should be a quick process since he already figured it out with mine. Please tip him what you can. He will not demand it but he is one of the only people who offers this service.

You will need an SPI programmer to flash the bios. If anybody is interested I will post more detailed instructions but its not for the faint of heart. You can check out bios-mods.com to get an idea of what is required.

I feel compelled to post some essential tips just in case anybody wants to try it out.

• AMD laptops use 1.8v for their bios chips. I tried two ch341a programmers and two 1.8v adapters and the only way i could get it to flash without errors is by disregarding the 1.8v adapter (using straight 3.3v) and connecting the laptop's power barrel plug with the main battery and CMOS battery unplugged. I have no idea why this worked but i suspect the laptop's charger was providing 1.8v which was averaging with the programmers 3.3v down to a lower value that the bios was happy with. ???
• Read the bios at least twice and make sure the CRC's match before flashing. This way you know you have a good connection.

 

Thank you for that post - very interesting and comprehensive.

 

Awesome post! Thank you for all the information!! I heard the jet-like fans when updating the BIOS as well haha.

Quick question, when I try to run the max fans script, I get the message "could not load EC plugin. Try to run ec-probe with elevated privileges." I've set all the .exe to run as administrator, and I ran the scripts as an administrator to. Do you have any idea what may cause this? Thank you!!!

Edit: Nevermind, I figured it out! For future users, I just simply typed out the command "ec-probe write 034 008" into PowerShell running on admin privileges. Not sure why the scripts weren't working.

 

Thanks!

I did a fresh windows install and noticed the same thing when I ran the script. I tried to strip down the NotebookFanControl files to only those essential for ecprobe but i think i missed a few files. I will upload a new version that works shortly.

EDIT: I figured it out. I didn't miss any files. Windows just blocks files that came from another computer by default. If you right click on the zip file and check "unblock" before unzipping it then all the scripts will work. I have reuploaded the zip compressed with 7zip instead of windows built in compression which might solve the issue entirely.

 

I just received my Swift X on Sunday and am really liking it so far. I came from a Dell XPS 13. How picky is the SwiftX with USB C chargers? Will a 65W PD charger work or must it be a higher wattage charger? My Dell had Thunderbolt so it seemed like basically any power delivery charger worked but it did not have discrete graphics so it didn't require much juice. Thank you.

 

Personally, I haven't tried using a 65W charger. I know that the most I've ever seen it draw when just charging (laptop off) is 55W so if your lucky it will charge the battery when doing stuff like web browsing and drain the battery when gaming (that's pretty bad for your battery though). There is a chance that it just wont work at all and will constantly disconnect. I guess just buy one from somewhere with a good return policy to try out.

 

Thanks for the reply. I will order one on Amazon and see how it goes. There are a few compact 100W ones that are ~$40-50. Will post results on here.

 

Any 20v charger will work, and possibly some 15v as well. I dont own any 100w chargers, but my two 65w 20v/3.25a charger work fine.

 

Not sure if you guys have figured this out yet, but the dell xps 15 9510 40w+5w vbios will work for the Swift X.
https://www.techpowerup.com/vgabios/235236/dell-rtx3050ti-4096-210427

Time Spy: https://www.3dmark.com/compare/spy/23644338/spy/23685051
Firestrike: https://www.3dmark.com/compare/fs/26463623/fs/26474536

Now the question is... what will the shunt mod do on top of the vbois?

Everything looks very stable, all the usb ports are working correctly, cpu seems to still get 37w. I played Overwatch for a couple hours, no issues. I did download the scripts, but i didnt try any of them other than max battery and disable gpu. I kinda wish those two could be combined as for my case use, when im on battery, im definitely not intending to do gaming.

 

Your 3dmark results are locked, can you type the scores?

That was the first thing I tried and the power reporting wasnt right. Thats why i resorted to shunt modding. It would report 6 watts in afterburner / hwinfo no matter the load on the GPU was and drew over 75 wats from the wall when running timespy (compared to 55 stock). Then whenever I would unplug the power cable it would get stuck at 700 watts (obviously wrong) and 200MHZ clock. only a reboot would fix it

I think the XPS uses different power registers and since its reading 6w it just keeps adding power, going way over the 45 watts. Then again, maybe I used a different bios and the one you used works fine. Can you tell me your power draw when running benchmarks and while plugging / unplugging the power?

Also how did you flash it? I couldn't get nvflash to see the 3050 and just ended up using my SPI programmer.

 

I use them combined, but I didn't know if others would forget that the GPU is disabled and get mad when it wasnt working right.

just copy "pnputil /disable-device "PCI\VEN_10DE&DEV_25A0&SUBSYS_15361025&REV_A1\4&C337064&0&0009" from the disable script and place it above "START:" in the max battery script.

 

sorry, i updated the 3dmark links to compares 35w vs 40w, should work now, sorry about that.

i used nvflash 5.692..0. the regular procedure, --protectoff, save the bois, then -6 bios.rom. hwinfo64 was reporting the wattage correctly for me. i will look at gpuz to see what it says. i dont have a watt meter unfortunately. though i do suspect the system was pulling cpu power and adding it to the gpu.

I noticed the power up link has a bios state of early september. Let me look at my BIOS revision because I believe I found multiple date codes and went with the latest one. Dell did offer up a bios update on those because of GPU performance issues.

 

and what about adding "enable gpu" to one or more of the other scripts?

you are right that keeping them separate is probably a good idea. but if its just as easy as customizing them ourselves, that is cool too.

 

i dont know why 3dmark is showing a blank score page. i think they are having issues. anyways....

Time Spy: graphics 5048 cpu 8296

Firestrike: graphics 13104 cpu 23407

 

Well, I got nvflash to work. I have no idea what my problem was before. I flashed the XPS bios and got the same results as before. GPU wattage stuck at 6.1W spiking up to 750W when i plug/unplug the power.

Idle:


Running a benchmark:


Plugging in power:


Unless we have different hardware i suspect you are getting the same results. It would explain your 5000 timespy score as well since I am only getting 4800 after shunt modding up to 50 watts. At 45w, you wouldn't be getting a score that high. From my testing its probably pulling closer to 60w on the GPU. As long as temps seem ok I dont think it will hurt to run it that high but then again I dont know if the VRMs are spec'd that high.

I did notice that it isnt getting stuck at 750W anymore locking my core clock at 200. It will shoot up to 750W when plugging/unplugging the power and then fall back to 6.1W. Maybe a driver update fixed it. Theoretically, as long as it reads 6w it will feed power to the GPU until the laptop runs out of supply power (90W)

All in all, i wouldn't recommend it since we dont know how much power the VRMs can handle.

 

Very interesting. When i have time i can do some 30 minute test and see what the core clock settles to. Nvidia has presets for each wattage range so we can probably deduce from the peak/average core speed what the actual power is.

 

Hey guys,
My Swift X should arrive this week, fingers crossed.
I got a replacement SSD for it - an SK Hynix P31 Gold (1TB).
An Intel AX210 wireless card to replace the Mediatek, a 100w USB-C charger and all that's left to do is to take care of the cooling.
I'm basically looking for recommendations to keep that thing as cool as possible, I'll be repasting it with TG Kryonaut, but I'm also considering adding pads because the cooling system seems pretty minimal.
I saw OP's recommendation of using a 3mm and a 2mm pads to transfer the heat from the heatsink to the case, looks sweet. But is there anything else besides that to consider? I tried looking at motherboard pics but it's not that easy to tell how much clearance there is, and where the important components are (not even YT's teardown channels showed what's under the heatsink).
BTW, any chance to upgrade the panel? I'd love to have touch support (I use the touchscreen a lot on my Precision 7510) \ a brighter screen \ both? - but it seems we're kinda limited with that 30pin connector.
What are my options exactly?

 

i highly doubt there is any vrm differences between a 35w and an 80w(+15w) 3050 ti mobile. im sure nvidia has some binned chips for low, mid, and high when it comes to the efficiency range, but they should all be using the same vrms. you are also probably right about being well above 50w draw on the gpu. following other 3050ti benchmarks, a 5000 graphics score is nowhere close the linear performance curve that it should be seeing. my out the box score was 3700, although i think i was in balanced mode, and 4000 seems to be the out the box performance mode score for this wattage range. an extra 5w should only be lifting the score about 10-12%.... not 20%+. again, i will take it, even if it means im losing a little bit of cpu score. i only game plugged in desktop mode anyways. the 5800u still has a massive 20mb cache, and in most games you would never notice the performance difference. but, i am definitely interested in some more tests on the cpu side to see whats going on, but, as long as it continues to provide adequate power boosting in cpu only tasks, i dont see a problem.

 

I went from FHD 30 to UHD 40 panel on my Clevo so, it's possible to probably do a 40P Touchscreen w/o much fuss.
http://forum.notebookreview.com/threads/nh58raq-np6855-screen-upgrade-experience.833976/

First thing you need is HWINFO to find the screen model info and then head to panelook.com and get the outline size to find a touch that's the same size physically.

Converting to touch from non-touch might be a bit more tricky since you're introducing a digitizer for the screen and potentially other HW to make it work. As far as your CPU/GPU supporting it there's really no doubt at this level. It's just a matter of fitting it with all the pieces you need and taking the chance that it might not work. My laptop never came with a 4K option but, after researching the CPUGPU schematics / specs showed everything was capable and ended up using the 30>>40cable provided in the service manual to get the job done.

Being this model is so new though it might not have a cable listed if it didn't come with a panel at 40P.

 

The VRM's themselves will be the same on a 90w 3050ti but I can guarantee there is more of them. The swift has 2 VRMs (you can see them in the picture i posted of the shunt resistors labeled R22) where a 90 w model will have at least 4. That's probably why your only pulling ~60W instead of 70 - 80 which is what would be left over from the total 90w budget when running a low cpu workload like 3dmark. Your basically maxing out the current those two VRM's can handle. Probably not great in the long run but if your comfortable with it more power to you.

Essentially what you are achieving is the old school shuntmod where you short the resistors with solder and other components are the only limiting factor. Not recommended in general but at these low power levels it might not be an issue.

 

Before I got the swift I tried out an ASUS Zenbook 14. With the two next to each other you could tell that the screen size and bezel size was identical so Im sure there are multiple panels that will fit.

The only thing i would worry about with a touchscreen is the digitizer connection. You can see on this ebay listing that there is an additional connection for the digitizer. without openeing up the laptop im not sure if there is a place to plug it in on the swift x.
https://www.ebay.com/itm/402517421401?epid=27036218062&hash=item5db7e86d59:g:jfIAAOSwbYJfmFKh

 

Thanks for the info! When I'll do the repaste, I'll also take a peek inside \ maybe even take a couple of pics because I couldn't find anything online.
I looked at other panels of well known 14" laptops, such as the Zephyrus G14 (at panelook.com), and they all seem to be quite different in their "Outline Dimension" part of the table. All have bigger stats than the stock screen, which makes me wonder if they do it on purpose just to discourage people from replacing their screens?
Is there something that can be done if a panel has an outline that's 8-10mm bigger than the stock panel, or should I forget about it?
I mean, if touch is out of the question, then it's either a higher res panel for Netflix \ high refresh rate for.. eSports games (3050ti) + having fun moving my mouse cursor around,
I couldn't find too many threads regarding panel swaps on the Swift (wasn't much of a problem on other models), mostly unanswered questions.
Do you guys happen to know better panels than stock that would fit nicely?

EDIT:
https://www.panelook.com/LM140LF1F02_Panda_14.0_LCM_overview_48911.html
That's the panel I mentioned earlier, there's a high res 120hz version as well

 

I tried to give you info on this but, you didn't read it. You need to provide the current panel info to figure out which panels will fit physically and then narrow down the features you want it to be able to do.

 

My bad!
I forgot to add that I was looking at the Zenbook's panels but I couldn't really find the really good ones for sale (like the 90hz, high resolution and color accuracy OLED panel from Samsung - the SDC4154, bummer).
I'll keep on looking manually, I guess.
My reference panel was taken from a random Swift X review - they're all the same, at least dimension-wise, right?
I'm not going to order anything before I get the laptop and double\triple check that the numbers I used were correct, but still.
I browsed through tens of panels, skipped the "boring" ones, but the good ones are either using weird dimensions or aren't for sale.
I did read your comment (maybe something lost in my translation, since English isn't my first language and as of now I'm awake for more than 48h straight, not the sharpest atm) and I really appreciate the will to help and the dedication of you guys to both log your experiences and then share them with such details so people wouldn't have to go through the entire process and waste their time.
You guys are awesome and sorry if I got you triggered. I'll be better in 10h or so, once I get home and have my 8h of sleep.

 

They're all different. You need the specific model of your laptop for the dimensions and pin count/pitch to start looking them up. Spending on how tight the fit is you might have a couple of mm's of tolerance. If the reference panel was from another identical model it might be a starting point for looking for something else. It's best though to double check your specific machine to make sure you have the correct model info before hunting down a panel. It's easy to find them on panelook by dimension rather than compatible models lised for the same specs.

It should be N140HCE-EN2 based on https://www.ultrabookreview.com/49973-acer-swift-x-review/

https://www.panelook.com/modeldetail.php?id=30961


https://www.panelook.com/sizmodlist.php?st=E20&pl=&so_attr=&sizes[]=1400&st=E20&by=desc&page=8 --- this is where your size will start

There's a few touch options / a few 120-144hz options but, mostly 60hz for this sizing.

Going touch will move you to a 40 pin connector which will either need a different cable or an adapter to make it work.

https://www.panelook.com/modelcompare.php?ids=30961,45278,48352,48911 -- here's 1 touch (three's more ~10 options) / 1 120 / 1 144 --- all will be 40 pins which isn't a big deal if you can find a cable

 

just a heads up on the repaste. there is red thermal goop on all the vrms and memory chips. you probably wont be able to reuse it and you wont be able to use thermal pads either, it will completely screw up the die contact. I would recommend buying a large tube of CMQ2 and using that to replace it.


I'm kind of tempted myself to try the 120 or 144hz panel that tech junky posted. I dont know much about replacing panels but if he says a 30 pin adapter will work then it should work. Kind of curious what those extra pins are for though. gsync definitely wont work fyi

EDIT: if your going to do it you might want to get the 120Hz option.
https://linustechtips.com/topic/1340853-30pin-edp-2-lane-cable-to-a-40pin-edp-4-lane-cable-adapter/

 

@ Pedro69 did something along these lines and has a good write up @ http://forum.notebookreview.com/threads/msi-gt75-8rg-screen-upgrade.834742/page-9

Different model but, same philosophy applies. I used a cable that went from 30P GPU end to 40P panel side w/o dealing with the converters. http://forum.notebookreview.com/threads/nh58raq-np6855-screen-upgrade-experience.833976/

The same cable I used might be an option for this sort of thing. The basic theory I've come up with is even through the GPU is 30P it still supports higher options to connect to different panels. Not all of the pins are being used on the panel side unless you're maxing out the full bandwidth of the bus on the GPU. A full 4-lane connection (8 pins +/-) leaves 22 to be used for other signaling on a 30P GPU.




We can see from these that:
Lane - 3/4 pins
Lane - 6/7 pins
Lane - 9/10 pins
Lane - 12/13 pins

The rest in the 30P are shuffled to hit the key points for power / ground / signaling.
12 pins in use on the 30P
20 pins in use on the 40P

On the touch cable there's a pigtail that splices out another connector that presumably is the USB controller function on the panel or it could be the digitizer function.

 

That's great to know, because I actually considered to apply only the paste for now, since pads aren't available in the stores near my area.
Can you give a rough estimate on the sizes of pads I'll need to get everything properly covered (including Acer's cost-savings, if there are any)?
BTW,
Does the modded BIOS let you adjust the timings as well? Because you didn't cover that part, and boy those are some high numbers.
What do you do in case of a failed boot? Would disconnecting the battery and pressing down the power button for a few seconds make it return to stock settings? I'm used to overclock on desktops, but never actually tried it on a laptop before.
But it should be worth it, at least with an AMD chip, considering how sensitive they are to memory tweaking.
I never bothered "critiquing" laptops' timings because I assumed I wouldn't be able to tweak them, so why even bother (it's the first time I'm hearing about BIOS unlocking), so I pretty much stuck to "real world" tests instead of the usual synthetic tests I run on my desktops.
Honestly, it was for the better, because it was shocking to read that you ram didn't run in (1:1) sync with the infinity fabric.
I know 2133Mhz (4266/2) is quite a big number for FCLK, but I wanted to believe they could make it work with it being soldered directly to the board / possibly better binning because of it being a low wattage chip and the superior 7nm manufacturing process laptops CPUs have / it being Zen 3 - my 'Zen 2' 3600 is sitting comfortably at 1900 FCLK, next gen 'Zen 3' 5600 can in many cases reach 2000, and the 5600G with the integrated graphics can reach even 10% higher FCLK than the non-G variant due to it's single die design.
Considering this CPU is in fact Zen 3 (5600\5800 are Zen 3, the rest - like the 5700u, are Zen 2), and has integrated graphics, it should be in theory be able to handle 2133MHz fclk.
But maybe if it 2133 is a stretch, which also makes me curious why it's the case - why didn't Acer give their system the most basic tuning, by taking the speeds down a notch and making sure they're coupled? And even if they can't be bothered by it, to cut down their costs - why not enable the option and let us do it ourselves then?
I honestly suspected they did it for the marketing - to be able to pump sexy phrases such as "4266MHz LPDDR4X" instead of taking it down because not all of their chips are able to reach that in 1:1, but surprisingly, their site doesn't actually specify the ram speed.
Anyway, I'll get me that kit you mentioned so I can flash the bios as well.

One last question, let's assume Acer's releases a new optimized bios that you'd like to have, are you going to have to flash the older bios first and then update? What will happen if you try to use their installers without going back? It will probably see something it doesn't like and abort, right?
And then what? We will have to dump the newer package and have the guy unlock the lost features once more?
Got material on what's needed in order to modify that thing myself? I can understand an IC specs sheet and I'm familiar with programming (even asm), but I don't feel like reinventing the wheel, would just like to get the general idea of what's needed to unlock the thing myself without having to rely on others to do it for me.

 

No, what I was saying is not to use pads. you will have to use paste. something like CMQ2 to replace the red goop.
unless your talking about the pads i put on top of the heatpipe. i would get a .5mm sheet and stack them to get the height you want.

yeah it does. I haven't messed with it too much because ive been busy but you could probably tighten them up. getting 1:1 working was a major pain. I had to add some voltage to the SOC using hex offsets in one of the menus and i didnt do extensive testing on 2100:2100 but it does boot and run aida64 memory benchmark.

you have to remove the cmos battery (its just a 2 pin connector that you unplug) and the laptop battery connector for about 30 seconds.

it will probably work just fine. i would bet it just looks at the bios version and if its lower it will work. if it doesnt you will always have your initial backup bios (pre-mod) that you can reflash.

I worked on that for a while but gave up. You can use UEFItool to extract modules from the bios and try to figure out which hex values to modify. It was too much work and i decided paying someone was a more efficient use of my time.



I have to stress how temperamental these bios voltages are. I still cant figure out why i couldn't get it to write reliably at 1.8v. If you do try it first make sure your CH341a is outputting 3.3v on the 3.3v pin, some of them have a problem with that.


The only way I could get it to work was by having the laptop power plugged in (not usb c) with the programmer supplying 3.3v at the same time (battery and cmos unplugged). it wrote reliably multiple times like this but any other configuration would brick the bios.

Ohh, and my motherboard voltage was not stable. I had to use an external usb 2.0 hub with its own power supply for the ch341a.

 

What's going on with that programmer? This vid was posted in June 2020, but it's still being sold with the same defect and users are expected to modify it themselves? 5v would fry some chips, gee.
So you ended up with a 3.3v output on the programmer, had the laptop turned on, and only then it allowed you flash?
I'm guessing only the chip itself requires 1.8v, and you probably chose the easier \ more logical approach of using a clip instead of desoldering it.
I can come up with 2 guesses, one is not having clean enough power via USB, which you solved by using an external power, which is great.
And the second thing is the chip not being isolated, so the voltage dropped probably on other components on the motherboard as well.
Did you measure the voltages? If that's really a 1.8v, I wouldn't feel comfortable using 3.3v + connecting the battery, without knowing the layout of the motherboard.
But I'm glad it worked for you, would be interesting to see the bios menu as well.
From what I understand, it's a licensed bios and the options are there, and each manufacturer can choose which options to grant access to the user (or hide from us)?
So the guy (Dudu?) alters properties of certain menus which were hidden by Acer, to now be visible to the user (like the .ini config files in games), and then gives us the modified bios back?
Nothing is added to the bios itself, just a matter of privileges?

 

I bought two ch341a's and neither one of them had that problem for what its worth.

I did use a clip. i think your right about the voltages dropping across the other components. I will say that that method (clip + external power) worked reliably on two swift x machines over multiple flashes but that could be true only for my specific combination of hardware so it might not be safe in general. It sounds like you know what your talking about so im sure you would be able to flash it successfully.

yes. insydeH2O provides the bios and acer locks down the options that they dont want the user to see. There are videos of people doing key combinations at boot to display all of the unlocked menus but I couldn't find one that worked for this laptop.

I dont think its as easy as changing a hex value from off to on. I worked with dudu for almost a week with him sending me modified bios to try before we got everything unlocked. I did a hex compare on my original vs the modified and their were a lot of changes.

 

It's not that easy because you're modifying an existing build instead of making it from scratch, considering the size of bios chips nowadays, there's a lot of code and changing \ deleting something can cause a butterfly effect, especially in languages like asm.
I'm assuming there's a standard of some sort, because programs in asm are usually portioned into "chunks" - like a space in memory that holds all the static information, like your motherboard's serial number and what not, a place that holds the functions - like the drivers needed for the motherboard to function, and I wouldn't be surprised if there are tests that check to if the original code was altered by someone, which isn't a hard thing to do.
So I'd imagine the guys doesn't need to go through megabytes of code - that's a lot of code, even with the padding of empty spaces between the code chunks, take a look at your motherboard - even at the box, and see that it's made out of an insane amount of sub-systems from different manufacturers crammed into a single PCB and need to behave like one.
So there has to be some sort of a template that he uses, there has to be a "generic" flow chart of the insydeH20.
And then you have to debug if things don't work as intended, as in having to search if the thing you altered was referenced somewhere else in a way that's causing a conflict \ you failed some kind of a security check in case they check for integrity. I'm sure he has some kind of a debugger that would show him the dependecies, and I'd love to know what he uses, because when I compiled code on an 8052 mcu I've had only several kbytes of code space so it was manageable without it, I imagine that a program in 2021 would probably guide him with arrows and what not, transform those megabytes of code into a flow chart.
That's basically what you get when you mess with someone else's code, you have to play a bit of detective, and even if it seems to be working fine, you can't really know for sure that you didn't forget something that might bite your butt later.
While the guys who burned your chip probably have an easy to follow gui, and they pretty much tick the right boxes according to the motherboard's hardware spec and manufacturer demands.
And the program fetches all the drivers it needs from it's library of drivers, fetches the agreed upon splash screen splash screen for the boot up process according to the maker of the order, makes sure you can't mess with anything and then adds an anti tempering measure. All there's left to do is to stamp a unique serial in both the box and chip when they're made (0 effort because it should be in the same place).
I think it's worth the hassle, though, I'd love to be able to tweak my memory and voltages.
I mean, I don't get why I have to go through all this considering it's my laptop, I paid for it - and I should be allowed to act stupid with it if I wish to.

 

Hi thanks for all this trick to get maximum performance out of this laptop but I just have a couple questions about the whole thing since I am quite new to all of this. I understand the scripts you have included changes watts and fan speed etc but when I try use Max battery it works for the first few seconds on the Ryzen Info script you have included but after it just goes back to 25W TDP and 23 W sustained. Is there any way to fix this?
Also is there any way to delete Ryzenadj as I was trying to start this whole process from the beginning but it wouldn't allow be due to some dlls being used by Powershell. Thanks for any help

 

It sounds like you have the battery gaming script running at the same time as the max battery script. I wrote the scripts to loop because amd's scheduler likes to reset the power limits occasionally.
you can only use one power script at a time. ryzeninfo and max fans dont modify the power settings so they can be run along side the others.

 

Stock VS Dell BIOS flash - it gets rather toasty with the BIOS flash, tops at nearly 90c for the CPU and 85c~ for the GPU
Haven't had the time to repaste yet, but I'm pretty sure the heat would be manageable once I'll get it done. Reverted the flash for the meanwhile.

 

the asus vivobook pro 14x and 16x have 5800h models and list their 3050ti's as 35w+15w dymanic. ive been scouring the web to see if i can get ahold of that vbios to give it a try and see what the results are. would be nice to get consistent reports with hwinfo/gpuz on power usage.

 

got to admit, 90hz oled sounds awesome. no usb-c charging makes it a dud for me though. https://www.ultrabookreview.com/49266-asus-vivobook-pro-14x-review/

I'm not sure that bios would work on this machine though. Presumably, it bases its 15w boost on CPU load. I wonder if it requires a setting in the motherboard bios or a additional chip on the gpu to do that. If it was just a vbios setting I wonder why all manufactures don't enable it. Thermals maybe?

As for getting the vbios, ive had luck asking people to make a gpuz backups on here if anyone makes a post about it in the asus forum.

 

Only way to know is to try it. Ive bern asking around but those midels are rare at the moment. Also, some benchmarks definitely show "stuck" in 35w mode while others show scores we would assume from a 50w power budget. Jarrodstech recently did a review and I don't think you picked up on the fact that he was getting 35w scores while other reviews show significantly higher. My assumption is exactly as you said that it depends on CPU load and where the system decides to deliver peak power.

 

Would there be any consequences for running the max fans script for long periods of times during a gaming session? Or is it fine for long term use?

 

should be fine. i do it occasionally and havent had a problem yet. worst case the fan is cheap and easy to replace.

I'm going to add it to the main post but you want make sure your nvidia settings are configured correctly for good thermals. there's no point in making more heat than necessary with a 60 Hz screen.

Max FPS: 65
Low Latency: Ultra
vsync: Fast

took me a while to figure out but this gets rid of screen tearing without gsync. If you see stutters use "borderless windowed" mode in the game.

 

Looks like Acer is adding a 16-inch model with Intel ARC graphics for 2022.

 

nice.

the 14" looks like a dud compared to the 2021 model. the only difference is the switch to intel but the 5800u is plenty for the 3050ti, at least for gaming. it usually only sits at 20% load even when the 3050ti is maxed out. good chance to pick up a discounted 2021 model when it comes out.
EDIT: looks like the screen is taller at 16:10. might be worth it if you need thunderbolt.

the 16" looks really good though depending on how arc performs. 3.7 lbs, 1440p. if it supports passthough PD charging like the 2021 model it will be fantastic. doubt you'll be able to shunt mod it though, lol.

 

https://www.techpowerup.com/vgabios/241526/241526

So it looks like this is the Asus Vivobook Pro 14x/16x 35w-50w vbios. Im not near my SwiftX at the moment to test it out. Im not exactly sure if it is the Intel or the Amd model though. Im hoping the 5800h model would have monitoring sensor data that matches our laptops.

I flashed back to the stock 35w-40w bios since the dell 40w-45w vbios is definitely does not work correct for the swiftx

 

great find!
looks like the M7600QE-L2014T that it came from is the ryzen 5900H model. I am very tempted to try it but I wont have access to my spi programmer in case something goes wrong for another week or so. if nobody else is willing to try it I will as soon a I can.

 

@Nikoli707 I added our bios to techpowerup ( https://www.techpowerup.com/vgabios/241569/241569) and the more I compare the two the more confidant I am that it will work. The two bios are identical except for the subsystem id and power limit.

 

ahh i didnt even notice to look at the bottom of the bios page it shows the M7600Q model. looking forward to seeing how it does.

 

it has 3 vrms compared to the swift's 2 vrms but i personally know the swift can handle 50w since ive been running it for months.
https://www.notebookcheck.net/Asus-...durance-and-mighty-OLED-display.576989.0.html

 

well I flashed it. No problems with the flash but it has the same problem as the XPS bios. Wattage always reports 6W so it just adds power until it reaches its physical power limit (around 60w). Plugging / unplugging the power connector registers 750W and if you do it with a load on the GPU (gaming) it will get stuck and lock the core clock at 200mhz until the load is removed. then it resets to 6w.

ohh well. would have been cool if it worked like its supposed to. I wish there was a way to tell which component is reaching its power limit when running this way to know if its going to fry the gpu long term.

Timespy score: 5651, graphics: 5332, CPU: 8554

 

Thank you all for all your work with regards to the Swift X. I tried the Asus vbios and observe the same behavior that Axalt reports. I have flashed back to the Acer vbios - back to normal.

 

I wanted to get some extra performance in games so I went back into the bios to play around with the memory timings and i managed to get the latency down from 114ns to 77ns. I was running stock settings since i had to reset the bios a while back and was too lazy to play around with it but the improvement is pretty drastic.

Stock:

• Mem clock: 2133 Mhz
• FCLK: 1067
• UCLK: 1067
• SOC: 0.8v

Modded:

• Mem clock: 2000 Mhz
• FCLK: 2000
• UCLK: 2000
• SOC: 1.125v
• Power Disable: Off

There are options to change the timings in the modded bios but unfortunately they have no effect so im stuck with the defaults. Still a great improvement though. I cant believe how unoptimized the memory is by default in this thing.