My Browser Builds (Part 5)

[roytam1]

New NewMoon 27 Build!

32bit https://o.rthost.win/palemoon/palemoon-27.10.0.win32-git-20240629-f69c4dc576-xpmod.7z
32bit SSE https://o.rthost.win/palemoon/palemoon-27.10.0.win32-git-20240629-f69c4dc576-xpmod-sse.7z
32bit noSSE https://o.rthost.win/palemoon/palemoon-27.10.0.win32-git-20240629-f69c4dc576-xpmod-ia32.7z

64bit https://o.rthost.win/palemoon/palemoon-27.10.0.win64-git-20240629-f69c4dc576-xpmod.7z

source repo: https://github.com/roytam1/palemoon27

repo changes since my last build:
- import changes from `dev' branch of rmottola/Arctic-Fox:
 - Bug 1237842 - Unlock mMutex before calling CloseActive. r=cpearce (d0677f1981)
 - Bug 1230857 - Ensure GMPService has sufficient file permissions to delete GMPs. r=gerald (e7f0c4b2b6)
 - Bug 1236380 - GMPStorage::mShutdown=true until Init() succeeds - r=cpearce (fde2025f4f)
 - Bug 1254311: [mp4] Ignore empty raw sample. r=cpearce (216a9417c3)
 - Bug 1215115 - part1: Replace the vorbis by opus in MediaEncoder and also reomve the VorbisTrackEncoder files. r=rillian (760c559e3c)
 - Bug 1215115 - part2: Mux opus into webm, remove bitdepth. r=rillian (1c996f0aee)
 - Bug 1215115 - part3: Fix gtest. Remove TestVorbisTrackEncoder.cpp. r=rillian (5a68915a4a)
 - Bug 1215115 - part4: Enable MOZ_WEBM_ENCODER by default. r=ted (6638b7fffb)
 - Bug 1257318: Pass TRACK_EVENT_ENDED events through to the TrackEncoders r=padenot (b92b2dcc94)
 - Bug 1261007 - Part 3 - Remove the same/redundant code of checking the unique image. r=jolin (608e6477bc)
 - Bug 1243611 - When EOS, call vpx_codec_encode correctly. r=rillian (83887c89c8)
 - Bug 1260353 - Remove unnecessary method AnimValuesStyleRule::AddPropertiesToSet() r=hiro (36f5e7fcc9)
 - Bug 1213775: VP8 automatic resizing breaks ffmpeg-based players; turn it off in VP8TrackEncoder r=jya (23c2a27371)
 - Bug 1185171 - Modify gmp-test-output-protection.h to prevent failure on machines without a physical monitor attached. r=bobowen (8375c5075d)
 - Bug 1185171: Add 0xc02625e5 as a valid failure code for GMPOutputProtection test. r=cpearce (1d10a75aeb)
 - Bug 1151746 - Origin tuples in should include schemes. r=edwin (32610b0cfa)
 - Bug 1180101 - Test 0 length atom inside moov; r=jya (3fae8aee45)
 - Bug 1244523: [mp4] P4. Add gtest. r=kentuckyfriedtakahe (3f71b5060a)
 - Bug 1255626: [gtest] Properly shutdown task queue should error occurs. r=gerald (4ec1bf360e)
 - Bug 1224363 - Added vp8/ivf test case - r=rillian Bug 1224369 - p1: Test cases given as list - r=rillian Bug 1224369 - p2: Added vp8/ivf test case - r=rillian Bug 1224361 - Added vp8/ivf test case - r=rillian (595ebe09be)
 - Bug 1231075. Respect the timestamp of video frames and don't pop frames as fast as we can in real-time mode. r=roc. (b72329c0fa)
 - Bug 1237160: Do not count frames not composited as dropped. r=cpearce (e7e18d0700)
 - Bug 1233648 - Fix some insufficient includes. r=kinetik. (e36cdd3e05)
 - Bug 1216460 - [1.1] Refactor data types, fix logs and prevent harmful type promotions in SourceBuffer eviction handling. r=jya (047a7ca64f)
 - Bug 1259916: [MSE] P1. Fix eviction. r=gerald (13195f392b)
 - Bug 1216460 - [2.2] Refactor SourceBuffer frame eviction and threshold defaults. r=jya (105962c942)
 - Bug 1259274: [MSE] P1. Remove unnecessary abstraction layer. r=gerald (e7b7603f30)
 - Bug 1259274: [MSE] P2. Remove unused code path. r=gerald (dce9fa447c)
 - Bug 1259274: [MSE] P3. Refactor handling of tasks so they only ever run concurrently. r=gerald (9c3f40d9b8)
 - Bug 1259274: [MSE] P4. Add AutoTaskQueue convenience class. r=gerald Just like TaskQueue, but doesn't require to be shutdown. (0310ff2b7f)
 - Bug 1259274: [MSE] P5. Use new AutoTaskQueue with MSE objects. r=gerald (3f72558eb2)
 - Bug 1259916: [MSE] P2. Bump audio source buffer eviction threshold to 30MB. r=gerald (2ffe148c1a)
 - Bug 1259916: [MSE] P3. Simplify eviction calculation logic. r=gerald (11250c02bc)
 - Bug 1199879: [MSE] Use latest demux end time to detect discontinuities. r=gerald (f89bdd763f)
 - Bug 1239983 - Diags around TrackBuffersMgr promises - r=jya (57f3e58636)
 - Bug 1258410: [MSE] P1. Abort if mInputDemuxer has been reset. r=gerald (07ca58adb0)
 - Bug 1258410: [MSE] P2. Disconnect init promise if any pending. r=gerald (0627c5a174)
 - Bug 1259985 - Add missing return after null-check - r=jya (b6ee457b89)
 - bit of Bwqug 1259274: [MSE] P3 (200d743676)
 - Bug 1216560 - [3.1] Make eviction thresholds const. r=jya (b44c78f999)
 - Bug 1259473 - per comment 14, move actions involving |this| to Init() from the constructor. r=jya. (30c402aacb)
 - Bug 1258562: MSE] Abort if MediaSource has been shutdown. r=gerald (6fce6bc9db)
 - Bug 1246358: [MSE] Take pre-roll time into consideration when seeking. r=gerald (dacbcd7f36)
 - spaces (abbb56d413)
 - Bug 657791 - Update WebM demuxer to clamp cueless seeks instead of failing. r=kinetik (785ae83126)
 - Bug 1219178 - [9.1] Make SeekPosition available with tests disabled. a=me for fixing build problems (cd1bdef203)
 - minor format (4a718e47f2)
 - Bug 1265399 - Replace 0.7071 with sqrt(0.5) in downmixing equations; r=padenot (2243d331c5)
 - Bug 1265794: P1. Ensure we can always fit a complete audio frame in an audio buffer. r=rillian (37f575184c)
 - Bug 1256626. Workaround Microsoft macro silliness. r=me (18930fbccd)
 - Bug 1264898 - Remove unnecessary |FinishAddTracks| call in |DOMHwMediaStream::Init|. r=jesup, r=pehrsons (1b610cdb4f)
 - Bug 848994 - p5. Check Silverlight presence - r=cpearce (98b4521ae3)
 - Bug 848994 - p6. Analyze Windows issues - r=cpearce (9de769166a)
 - Bug 848994 - p7. Filter front-end notifications - r=cpearce (e3aab89a95)
 - Bug 1256533 - Use std::deque<int32_t> instead of nsDeque - r=cpearce (e21c02fcab) (fef8b08889)
- import changes from `dev' branch of rmottola/Arctic-Fox:
 - Bug 1257063 - Don't destruct the runnable inside the lock when TaskQueue::Dispatch fails. r=bobbyholley. (1f6b254bb0)
 - Bug 1202148 - Move current in only one location in Intervals.h, r=jya (2c98d86b10)
 - Bug 1258673. Part 1 - cache mStreamOffset so we won't read at the wrong position when Other Read() interrupt the current Read(). r=jya. (87ab65cc30)
 - Bug 1258673. Part 2 - since mStreamOffset is not updated until the end of MediaCacheStream::Read(), we have to pass stream offset to MediaCache::NoteBlockUsage explicitly to avoid hitting the assertion. r=jya. (f02806ea1c) (9c8d43b8a2)
- import changes from `dev' branch of rmottola/Arctic-Fox:
 - bug 1191889 skip Close() when not initialized r=roc (9bbb78c01e)
 - Bug 1175319 - Remove outdated MDSM documenation. r=jya DONTBUILD (775d7b1cb4)
 - Bug 1265978. Part 1 - add mozDumpDebugInfo() to HTMLMediaElement.webidl. r=bz. (6518ad6a08)
 - Bug 1265978. Part 2 - add methods to MediaDecoder and MDSM to dump debugging info. r=jya. (1b8ad138c8)
 - Bug 1265978. Part 2.5 - also dump reader data. r=jya. (d866537269)
 - Bug 1265978. Part 3 - invoke mozDumpDebugInfo() from JS. r=jya. (7a5a7b5837) (f69c4dc576)

[roytam1]

New regular/weekly KM-Goanna release:
https://o.rthost.win/kmeleon/KM76.5.4-Goanna-20240629.7z

Changelog:

Out-of-tree changes:
* update Goanna3 to git 2828386673...f69c4dc576:
- import changes from `dev' branch of rmottola/Arctic-Fox:
 - Bug 1237842 - Unlock mMutex before calling CloseActive. r=cpearce (d0677f1981)
 - Bug 1230857 - Ensure GMPService has sufficient file permissions to delete GMPs. r=gerald (e7f0c4b2b6)
 - Bug 1236380 - GMPStorage::mShutdown=true until Init() succeeds - r=cpearce (fde2025f4f)
 - Bug 1254311: [mp4] Ignore empty raw sample. r=cpearce (216a9417c3)
 - Bug 1215115 - part1: Replace the vorbis by opus in MediaEncoder and also reomve the VorbisTrackEncoder files. r=rillian (760c559e3c)
 - Bug 1215115 - part2: Mux opus into webm, remove bitdepth. r=rillian (1c996f0aee)
 - Bug 1215115 - part3: Fix gtest. Remove TestVorbisTrackEncoder.cpp. r=rillian (5a68915a4a)
 - Bug 1215115 - part4: Enable MOZ_WEBM_ENCODER by default. r=ted (6638b7fffb)
 - Bug 1257318: Pass TRACK_EVENT_ENDED events through to the TrackEncoders r=padenot (b92b2dcc94)
 - Bug 1261007 - Part 3 - Remove the same/redundant code of checking the unique image. r=jolin (608e6477bc)
 - Bug 1243611 - When EOS, call vpx_codec_encode correctly. r=rillian (83887c89c8)
 - Bug 1260353 - Remove unnecessary method AnimValuesStyleRule::AddPropertiesToSet() r=hiro (36f5e7fcc9)
 - Bug 1213775: VP8 automatic resizing breaks ffmpeg-based players; turn it off in VP8TrackEncoder r=jya (23c2a27371)
 - Bug 1185171 - Modify gmp-test-output-protection.h to prevent failure on machines without a physical monitor attached. r=bobowen (8375c5075d)
 - Bug 1185171: Add 0xc02625e5 as a valid failure code for GMPOutputProtection test. r=cpearce (1d10a75aeb)
 - Bug 1151746 - Origin tuples in should include schemes. r=edwin (32610b0cfa)
 - Bug 1180101 - Test 0 length atom inside moov; r=jya (3fae8aee45)
 - Bug 1244523: [mp4] P4. Add gtest. r=kentuckyfriedtakahe (3f71b5060a)
 - Bug 1255626: [gtest] Properly shutdown task queue should error occurs. r=gerald (4ec1bf360e)
 - Bug 1224363 - Added vp8/ivf test case - r=rillian Bug 1224369 - p1: Test cases given as list - r=rillian Bug 1224369 - p2: Added vp8/ivf test case - r=rillian Bug 1224361 - Added vp8/ivf test case - r=rillian (595ebe09be)
 - Bug 1231075. Respect the timestamp of video frames and don't pop frames as fast as we can in real-time mode. r=roc. (b72329c0fa)
 - Bug 1237160: Do not count frames not composited as dropped. r=cpearce (e7e18d0700)
 - Bug 1233648 - Fix some insufficient includes. r=kinetik. (e36cdd3e05)
 - Bug 1216460 - [1.1] Refactor data types, fix logs and prevent harmful type promotions in SourceBuffer eviction handling. r=jya (047a7ca64f)
 - Bug 1259916: [MSE] P1. Fix eviction. r=gerald (13195f392b)
 - Bug 1216460 - [2.2] Refactor SourceBuffer frame eviction and threshold defaults. r=jya (105962c942)
 - Bug 1259274: [MSE] P1. Remove unnecessary abstraction layer. r=gerald (e7b7603f30)
 - Bug 1259274: [MSE] P2. Remove unused code path. r=gerald (dce9fa447c)
 - Bug 1259274: [MSE] P3. Refactor handling of tasks so they only ever run concurrently. r=gerald (9c3f40d9b8)
 - Bug 1259274: [MSE] P4. Add AutoTaskQueue convenience class. r=gerald Just like TaskQueue, but doesn't require to be shutdown. (0310ff2b7f)
 - Bug 1259274: [MSE] P5. Use new AutoTaskQueue with MSE objects. r=gerald (3f72558eb2)
 - Bug 1259916: [MSE] P2. Bump audio source buffer eviction threshold to 30MB. r=gerald (2ffe148c1a)
 - Bug 1259916: [MSE] P3. Simplify eviction calculation logic. r=gerald (11250c02bc)
 - Bug 1199879: [MSE] Use latest demux end time to detect discontinuities. r=gerald (f89bdd763f)
 - Bug 1239983 - Diags around TrackBuffersMgr promises - r=jya (57f3e58636)
 - Bug 1258410: [MSE] P1. Abort if mInputDemuxer has been reset. r=gerald (07ca58adb0)
 - Bug 1258410: [MSE] P2. Disconnect init promise if any pending. r=gerald (0627c5a174)
 - Bug 1259985 - Add missing return after null-check - r=jya (b6ee457b89)
 - bit of Bwqug 1259274: [MSE] P3 (200d743676)
 - Bug 1216560 - [3.1] Make eviction thresholds const. r=jya (b44c78f999)
 - Bug 1259473 - per comment 14, move actions involving |this| to Init() from the constructor. r=jya. (30c402aacb)
 - Bug 1258562: MSE] Abort if MediaSource has been shutdown. r=gerald (6fce6bc9db)
 - Bug 1246358: [MSE] Take pre-roll time into consideration when seeking. r=gerald (dacbcd7f36)
 - spaces (abbb56d413)
 - Bug 657791 - Update WebM demuxer to clamp cueless seeks instead of failing. r=kinetik (785ae83126)
 - Bug 1219178 - [9.1] Make SeekPosition available with tests disabled. a=me for fixing build problems (cd1bdef203)
 - minor format (4a718e47f2)
 - Bug 1265399 - Replace 0.7071 with sqrt(0.5) in downmixing equations; r=padenot (2243d331c5)
 - Bug 1265794: P1. Ensure we can always fit a complete audio frame in an audio buffer. r=rillian (37f575184c)
 - Bug 1256626. Workaround Microsoft macro silliness. r=me (18930fbccd)
 - Bug 1264898 - Remove unnecessary |FinishAddTracks| call in |DOMHwMediaStream::Init|. r=jesup, r=pehrsons (1b610cdb4f)
 - Bug 848994 - p5. Check Silverlight presence - r=cpearce (98b4521ae3)
 - Bug 848994 - p6. Analyze Windows issues - r=cpearce (9de769166a)
 - Bug 848994 - p7. Filter front-end notifications - r=cpearce (e3aab89a95)
 - Bug 1256533 - Use std::deque<int32_t> instead of nsDeque - r=cpearce (e21c02fcab) (fef8b08889)
- import changes from `dev' branch of rmottola/Arctic-Fox:
 - Bug 1257063 - Don't destruct the runnable inside the lock when TaskQueue::Dispatch fails. r=bobbyholley. (1f6b254bb0)
 - Bug 1202148 - Move current in only one location in Intervals.h, r=jya (2c98d86b10)
 - Bug 1258673. Part 1 - cache mStreamOffset so we won't read at the wrong position when Other Read() interrupt the current Read(). r=jya. (87ab65cc30)
 - Bug 1258673. Part 2 - since mStreamOffset is not updated until the end of MediaCacheStream::Read(), we have to pass stream offset to MediaCache::NoteBlockUsage explicitly to avoid hitting the assertion. r=jya. (f02806ea1c) (9c8d43b8a2)
- import changes from `dev' branch of rmottola/Arctic-Fox:
 - bug 1191889 skip Close() when not initialized r=roc (9bbb78c01e)
 - Bug 1175319 - Remove outdated MDSM documenation. r=jya DONTBUILD (775d7b1cb4)
 - Bug 1265978. Part 1 - add mozDumpDebugInfo() to HTMLMediaElement.webidl. r=bz. (6518ad6a08)
 - Bug 1265978. Part 2 - add methods to MediaDecoder and MDSM to dump debugging info. r=jya. (1b8ad138c8)
 - Bug 1265978. Part 2.5 - also dump reader data. r=jya. (d866537269)
 - Bug 1265978. Part 3 - invoke mozDumpDebugInfo() from JS. r=jya. (7a5a7b5837) (f69c4dc576)

* Notice: the changelog above may not always applicable to XULRunner code which K-Meleon uses.

A goanna3 source tree that has kmeleon adaption patch applied is available here: https://github.com/roytam1/palemoon27/tree/kmeleon76

[RamonUn]

7 hours ago, roytam1 said:

Test binary:
IceDove-UXP(mail) https://o.rthost.win/hbl-uxp/icedove.win32-20240629-id-656ea98-uxp-f35b2f4d3d-xpmod.7z
IceApe-UXP(suite) https://o.rthost.win/hbl-uxp/iceape.win32-20240629-id-656ea98-ia-93af9a0-uxp-f35b2f4d3d-xpmod.7z

Those are 404 for me, I can't see them listed in https://o.rthost.win/hbl-uxp/ either.

[RamonUn]

Double post mistake, delete please

Edited June 29 by RamonUn

[UCyborg]

Just as I embraced browsing with a 64-bit Pale Moon on this trusty old desktop, the aging hardware has started to fail. Two days ago, I scheduled a shutdown and went to sleep, when I woke up it was still running, completely unresponsive, screens in standby. After I reset it, I noticed Windows bugchecked with DRIVER_POWER_STATE_FAILURE in amdppm.sys. Not much is written about this driver, the file description just says Processor Device Driver, internet suggests it does power management for the CPU among other things.

A day later I have a browser open and a video playing in PotPlayer and all of the sudden, video starts to lag and computer gradually become unresponsive, to the point nothing responds anymore, almost totally frozen picture, just the mouse pointer still moves.

Another reset, run MemTest86+ 5.01 and it hangs 5 minutes into the test, just the plus in MemTest86+ text is still blinking, but everything else, including Time, stops. I noticed it stopped in 4-6 GB range, where the cheap new-old Kingston 2 GB I got half-year ago or so is.

Remove the Kingston and run it again, this time I get error spam few minutes into testing. Go to BIOS and reset CPU Overclocking from Overclocking Profile, where 7% was set below, to Auto (it was providing slight frequencies' boost, 0,2 GHz for the CPU and slightly higher other clocks; RAM, FSB...), run the test again, it seems stable.

Then I reinstall the Kingston and again, hangs 5 minutes into test. So I'm back from 6 GB of RAM to 4 GB.

And I suppose it's only a matter of time before factory frequencies become unstable. The 7% overclock configuration has been stable for 15 years!

Edited June 29 by UCyborg

[modnar]

@roytam1 BOC/UXP (Mailnews) link also doesn't work (404). Please fix it.

[roytam1]

3 hours ago, RamonUn said:

Those are 404 for me, I can't see them listed in https://o.rthost.win/hbl-uxp/ either.

 

3 hours ago, modnar said:

@roytam1 BOC/UXP (Mailnews) link also doesn't work (404). Please fix it.

forgot to upload before sleep, and they're uploaded now.

[roytam1]

3 hours ago, UCyborg said:

Just as I embraced browsing with a 64-bit Pale Moon on this trusty old desktop, the aging hardware has started to fail. Two days ago, I scheduled a shutdown and went to sleep, when I woke up it was still running, completely unresponsive, screens in standby. After I reset it, I noticed Windows bugchecked with DRIVER_POWER_STATE_FAILURE in amdppm.sys. Not much is written about this driver, the file description just says Processor Device Driver, internet suggests it does power management for the CPU among other things.

A day later I have a browser open and a video playing in PotPlayer and all of the sudden, video starts to lag and computer gradually become unresponsive, to the point nothing responds anymore, almost totally frozen picture, just the mouse pointer still moves.

Another reset, run MemTest86+ 5.01 and it hangs 5 minutes into the test, just the plus in MemTest86+ text is still blinking, but everything else, including Time, stops. I noticed it stopped in 4-6 GB range, where the cheap new-old Kingston 2 GB I got half-year ago or so is.

Remove the Kingston and run it again, this time I get error spam few minutes into testing. Go to BIOS and reset CPU Overclocking from Overclocking Profile, where 7% was set below, to Auto (it was providing slight frequencies' boost, 0,2 GHz for the CPU and slightly higher other clocks; RAM, FSB...), run the test again, it seems stable.

Then I reinstall the Kingston and again, hangs 5 minutes into test. So I'm back from 6 GB of RAM to 4 GB.

And I suppose it's only a matter of time before factory frequencies become unstable. The 7% overclock configuration has been stable for 15 years!

you may open up the chassis and inspect the motherboard, maybe some capacitors are failing (most noticeable symptom is popping on the top)

[UCyborg]

Yeah, I suspect it's capacitors too, they look fine, but as far as I know, as they age, capacitance goes down and equivalent series resistance goes up.

I'm shocked by the sheer number of them, there's more than 26 of them on the board! 26 with familiar top with a T shape, then at least 6 with a red stripe on top and possibly few much smaller ones (assuming I didn't confuse them with something else).

With my poor (de)soldering skills and tools, I'm not confident about touching this board right now.

I guess the heat could be a smaller factor, it's pretty hot these days and the room it's in is about 27 °C. Blowing the dust off harder accessible spots is still on my TODO.

[roytam1]

1 hour ago, UCyborg said:

Yeah, I suspect it's capacitors too, they look fine, but as far as I know, as they age, capacitance goes down and equivalent series resistance goes up.

I'm shocked by the sheer number of them, there's more than 26 of them on the board! 26 with familiar top with a T shape, then at least 6 with a red stripe on top and possibly few much smaller ones (assuming I didn't confuse them with something else).

With my poor (de)soldering skills and tools, I'm not confident about touching this board right now.

I guess the heat could be a smaller factor, it's pretty hot these days and the room it's in is about 27 °C. Blowing the dust off harder accessible spots is still on my TODO.

they may not leaking any juice, just bulging a little bit but it does hurt.

[modnar]

Thank you for the new releases, roytam1.

By the way I have to add - that mozjs.dll rare crash with videocardz.com - some days ago I have noticed just how important in Windows the Service "Tag" values are and that they are properly registered in "GroupOrderList". One really needs to go through all "Group"'s mentioned in CurrentControlSet/Control/GroupOrderList and search in "Services" if their "Tag" value is mentioned in "GroupOrderList". This goes for any service, even with the "Start" value of 4 (stopped), if it has a "Tag", it has to be registered in CurrentControlSet/Control/GroupOrderList. So I was fixing M$ half a$$ed work and for service "i2omgmt" of "SCSI Class" with "Tag"="2d" I couldn't be bothered to type in all the values from 01 to 2d, so I just punched in 03 01 02 2d (with the usual 3 pairs of zeroes for each), but that while making the whole driver stack more fluen, also wasn't optimal and mozjs crashes are an indicator - I know that now. So yesterday I added in all values from 01 to 2d and all is well with 06-22 and 06-29 versions of Serpent. Unbelievable that even in Windows 7 there are some left out (to mind comes some HP service and its tag of 103 while in SCSI Miniport only 100 and 101 are metioned). Haven't checked Windoze 10 yet. Here are my registry files for such problems 3 groups appear to have them and it really helps to have these in order - even Diskeeper12 (has very low level drivers) works much better (renamed to .txt - for perusal - the second, FSFilterInfrastructure is shortened to 03 because there aren't any filters with Tag=4 anymore).

GroupOrderList_Filter_0B_not_06.txt GroupOrderList_Infrastructure_03.txt GroupOrderList_SCSI_Class_01_to_2d.txt

[UCyborg]

6 hours ago, roytam1 said:

they may not leaking any juice, just bulging a little bit but it does hurt.

Yeah, can't say I noticed any. BTW, I missed some when counting for first time, there's 31 T-on-top-shaped ones and 11 ones with s red stripe, and possibly 6 smaller ones and 8 even smaller ones.

I didn't get around to blowing dust out for at least 2 years now, so finally did it, seems stable now, even with 7% overclock, CPU heats to 50 °C now rather than 54 °C when under heavy load. There was a lot of dust in a quite massive heatsink of Akasa AK-968, the aftermarket CPU cooler I have.

That Kingston's RAM stick still doesn't work correctly, one of the tests just make everything freeze, so stuck with 4 GB of RAM again. If it was at least DDR3...oh well, I'm used to it, but those extra 2 GB could really make a big difference in general.

[we3fan]

20 hours ago, UCyborg said:

With my poor (de)soldering skills and tools, I'm not confident about touching this board right now.

If you ever decide to try this yourself, there are (at least there used to be) cheap Soldering Kits from Aliexpress, good for practice.

For example, 5-10$, there is 10x10cm board inside with 10 LED diodes and instructions, you solder the 10 LED diodes on the board, if completed successfully - all LED diodes light up in a certain pattern.   All in the name of practice, before you attempt the real deal.

[AstroSkipper]

13 hours ago, UCyborg said:

That Kingston's RAM stick still doesn't work correctly, one of the tests just make everything freeze, so stuck with 4 GB of RAM again. If it was at least DDR3...oh well, I'm used to it, but those extra 2 GB could really make a big difference in general.

Mixing RAM modules from different manufacturers with different timings is always problematic. More like a game of chance. It is best to replace them all with same modules from one production series. That's what I did with my lady's Fujitsu notebook, built in 2010. I bought the modules second-hand a few years ago as new ones were no longer available. I completely replaced two 1 GB modules with two 4 GB ones. It worked perfectly and still works in these days. 

Edited June 30 by AstroSkipper

[Mathwiz]

9 hours ago, AstroSkipper said:

It is best to replace them all with same modules from one production series.

Yes, that is best, but it's not really practical! Two of anything, even from the same manufacturer purchased at the same store at the same time, won't always be from the same production run.

Just try to get the ratings the same; that's often the best you can do.