ClockBlocker (profiled AMD power-management control)

Don't worry. It's just a minor annoyance.

 

I have a R9 380 4GB and i was a Clockblocker heavy user since the begin.

But months ago I noticed that they're preventing my overclock. So everytime that I run this program, my clocks go to default and I can't overclock anymore. Even on the program. I have tried some options, but my PC had some instabilty issues (blinking screen and black screen after reset), so I give up for now.

So is possible to overclock and use this problem at the same time on lasts Crimson. If it is possible, what I have to do?

 

While my ability to test is unfortunately limited without owning more AMD hardware, and while I've not seen any instability (with the Fury-X), the resetting of clocks on startup might be a misdetection of a broken or stuck clock-state. If ClockBlocker detects such a state, it'll issue a reset. That would of course revert the settings at least once per start and potentially each time that the ADL module is started thereafter (if there's any ADL rules or if methods #3 or #4 were checked). No matter what (regardless settings), later versions of ClockBlocker do this test on startup.

If you could, email or message me a log dump on your machine after starting ClockBlocker (this might shed some light on what's happening).

--You might also have some luck simply deleting "ClockBlocker_HelperModule_ADL.exe" in the program folder. This will completely eliminate any chance of CB changing wattman settings. The default compute method isn't effected by this and would continue working.

The log will show some warnings / complaints of the missing file, yet these can be ignored and won't cause any harm.

 

-- I have a reproduction of ClockBlocker "improperly" restoring default settings on the latest Crimson drivers with one of my machines. (improper jammed state detection)

As such I've also tested removing the ADL helper module, which is what I suggest doing as the temporary workaround for now [if you're effected, if not no need to do anything -- the bug is harmless and just wipes OC settings]. This will be investigated, no ETA on the fix yet.

 

Time to start filling in the blanks on YAP as it draws closer to its first beta release. (aiming for in the next two days, though there will be some bugs in this release)


^ Please note that the client releases most of its allocations (for the GUI) when in the tray, and its memory usage goes up or down depending what's currently open and visible.


YAP is a fully client-server, network driven, task manager, integrated with ClockBlocker and trying to address each complaint that people had with ClockBlocker's interface and functionality.

The client can be run locally and it can of course also be run on just a single machine with the server, however, the client can also be on a second computer that monitors the first. This includes for watching processes, setting priorities, killing processes, editing rules, enabling ClockBlocker, etc.


Most (though not all) colors in the client can be customized. The plot can be customized, the list of processes, the log viewer, etc.







As you may notice, it's also very easy to restore settings to their defaults this time. Either by a single tab of settings, or for a single setting.



YAP has a built in firewall for gamers who have Windows firewall disabled. Later versions will include auto blacklisting of addresses that have incorrect login credentials (anti bruteforce), a connected client viewer (not just via the log), etc. As this tool is networked, able to kill processes remotely, and will (probably as a premium feature) have the ability to launch processes on the remote system, it obviously "might" be dangerous if not secured on a public network.

YAP uses encryption (twofish), routine key re-exchanges, compression, and differential snapshots of processes. It uses on-average less than 15kbit a client @ 5 task-updates a second. (per 100 processes)

On a Threadripper and Xeon E5, the CPU use is under 1% of the system. The client's GUI is honestly the most demanding portion of this, yet the majority of the GUI goes dormant when minimized to the system tray.




The rule system may look a bit different this time. Due to popular demand, this is not an automatically sorted rule system like ClockBlocker. YAP's rules are ordered, and offers two evaluation modes: parallel and topdown. topdown being similar to the classic ClockBlocker, and parallel being more suited for process priority adjustment.



Although not visible here, the rule editor is scrollable and there will be even more added in here in the future (to the point that it can't all fit on the screen). Each rule can hold many processes that it's associated with too. For instance, we can have a single rule that covers every single game on our computer if wanted.

The rule engine has been drastically improved and is suitable this time for inclusion rules rather than exclusion. All criteria evaluations are cached and only run once per process, even in the case of continual enforcement of process priority, affinity, etc. So yes, if you want to just have a single rule with hundreds of games in there, you can do this by all means.



While the rule editor is open, the main way to add applications this time is via the navigation (browser) window.


I'll sum this progress update up with this for now...



There will be registration unlockable features in YAP, unlike in ClockBlocker, and I will also consider taking donations this time around as the project grows. However I would like to stress that everything that's in the BETA release will be available in the FREE release for home-users.

Most of these features will be targeting business use of YAP and things that in my mind are less interesting to "gamers". I cannot stress enough that this is not going to be a shareware in the normal sense, nor a freemium. There will never be registration "nags" in YAP, ever, bundling of other software, ads, nor will any gating of YAP's features remove or hinder the ability to use it completely free.


It's also worth mentioning that despite how grumpy I usually am about Logitech's Software and its problems with ClockBlocker, these are completely eliminated in YAP. All windows messaging has been removed during the rewrite, as the project moves more and more platform agnostic.

--In the future, a Linux client and Raspberry PI client will be coming, but I cannot say how soon.

YAP is service-based, designed to run in session-0 coupled with a bridge task for window polling in the user's session. The client doesn't need admin rights and doesn't even have to be running for rules to go in to effect. All issues with Windows 10 auto-start in 1709 and startup UAC prompts [reported by some] should be taken care of.

I fully intend to backport the recent changes to YAP's Compute Helper back to the ClockBlocker helper, yet the first priority is to get YAP out as a beta. The two softwares are not intended to be used concurrently though, so it's my hope that most people will find YAP a worthy successor to ClockBlocker.


Hopefully I'll manage to get a quick how-to guide done before the first beta release, but no promises.
Cheers all!

 

Impressive stuff, mate! Keep it up!

 

I'm pleased to say that I've managed to crush a good deal of bugs (of those I'm aware of), to get in some heavy self-testing, and to get YAP to a "fairly useable" state by xmas. Sadly, this became the primary focus and I still have not found time to write up a suitable documentation on usage yet. Fortunately, YAP is a similar design to ClockBlocker and probably will be intuitive to anyone who's used it before.

Most of the Interface is also heavily labeled with tooltips, so for now mouse over everything for descriptions.

Code:

SHA256                                                                    Size  Name
---------------------------------------------------------------- -------------  ------------
B06B15B77EA65935D1FCF425DF9A41077523B495C7EF1DD1E706C87FB9CEDE43       4323404  YAP_BETA_Setup.exe

Check out YAP's first beta release if you're interested in taking it for a test-drive.
--Let me know what you think especially on the concept, the interface (what's confusing provided just the tooltips), and of course if you find something seriously wrong with the behavior of the software.


Notes and known issues going in to the first beta:
(fixed) -On installing over an existing installation the client and server will be automatically terminated by the installer. However, graceful close of the client often takes longer than the installer waits.
This causes a 'retry' prompt (from file in use) during writing the client executable. You can simply click retry after the client finishes closing, but this will eventually be fixed.

-There is no auto-updater yet. (simply haven't gotten around to it)

-While "technically" nothing bad should happen running ClockBlocker and YAP at the same time, please (at least temporarily while testing) remove ClockBlocker from auto-start.

(fixed) -The process listbox (main browser dialog of the client) does not paint the active-selection [from mouse-hover] in its inline-widgets. (the executable column)

(fixed) -If a dialog such as the settings dialog is resized such that its edge is over the listbox's inline-widgets, then their mouse-capture can interfere with grabbing the dialog's edge. (eg, this may make resizing difficult)

(seems fixed) -The server's memory usage grows over time, as it keeps growing circular buffers in memory. Its memory manager will also hold on to heap allocations and reuse them where possible. These are intended to eventually plateau and hit a limit -- so memory raising over the initial consumption is expected and does not necessarily imply a leak. There may still be memory leaks, I'm still watching the behavior of this closely.

-The server process has a high pagefault count. These are soft-faults and do not negatively impact its performance.

(more options added) -I'm aware that some people use 27" 4KHD monitors at native resolution. Currently there is not much option for adjusting font sizes beyond the graph widget. This will be changed of course, and more choices for font size will be added.

-Order matters, EVEN WITH parallel rules. Parallel rules are executed as the name implies in parallel for each process, however the first rule to match for a process will stop execution.

For instance if you have two rules that would match for a process, and the first rule is "foreground" scope, then this will stop evaluation of a rule with "both" set for its scope. For readability I still suggest trying to avoid having scope conflict, and using foreground and background rules explicitly. (eg, instead of relying on evaluation order to accomplish this)

-Parallel rules and topdown rules can fight with eachother, as by design they're executed in tandem. The idea is that parallel rules are most appropriate for general task manager stuff (priority, affinity adjustment, etc), whereas topdown rules probably make the most sense for defining 'exceptions' for Clock Blocking.

It's probably a good idea to stick to one or the other unless you truly need both, or to just use topdown rules as a supplement for Clock Block.

-Rules by default inherit colors from their parent (higher up rules order-wise). If your color settings for a rule don't appear to be working, make sure that inheriting color has been unchecked.
**Dividers are a special case of rule where they do not respect their own color setting -- the purpose of a divider is to label and set the color for rules underneath**

-When editing rules: [be that adding a rule, adding a divider, copying a rule] These actions will insert under the currently selected item.
** If you need to move rules to the top after creation, you can use the up and down arrows**

(fixed) ^ There is a bug in that on no-selection items are not inserted at the top of the list. (they're inserted under the first element)

-On first start of the server (with no config stored), the server will report that there was "an error" loading parallel and topdown rules. This is working as intended and should go away after rules have been saved at least once.

(fixed) -3D process detection may take a few seconds if an application doesn't have a 3D API immediately loaded on launch. Detection is cached, and thereafter alt-tabbing out and in should be instantaneous. (per rule evaluation)
^ HINT: If you want a "special rule" that acts the same as in ClockBlocker, just set "*" for your match criteria of the rule and then check the checkbox. (you would also want it in this case to be a topdown rule)

-Please read the text thoroughly on the download page (the section on the password & networking). There's a reason I didn't just link the download directly.
^ If by random chance you have another program or service on TCP port 1021, you'd need to move it for now. (in the beta)


Overall my hope is that YAP is already usable as it is now on the first beta release (though this may be wishful thinking). Even if it is, it's probably not polished enough for 24/7/365 use (continuous use). It's on my own machines, but as the project has constantly been recompiled and I haven't gotten time to just-use-it, it has limited extended testing.

 

Thank you so much for this first beta release. I'll try to find time to test it over the holidays.
Merry Christmas to you and everyone here at Guru3D

 

Code:

SHA1                                              Size  Name
---------------------------------------- -------------  ------------
5F7B00B60231C337DA7FF702EFE265908B234318       4323791  YAP_BETA_Setup.exe

1.0.1 Server Changes:
-Fixed a significant bug in the caching for 3D process detection. This caused some games to not be detected and it's strongly recommended to grab the new version if you're using that feature.
-Server Service now reports a version string on startup to the server-log.
-Rule system evaluator improvements. (process detection speedup)


Client and helper modules are not touched. Simply install over your current installation to upgrade, and click retry as needed until the client is closed (see aforementioned known installer issue).

Installer can be attained like before from the downloads page. This installer carries the same filename as the prior, you can check if the new version has been installed by checking the server-log.

 

Thanks so much, mate and happy New Year!!!

 

Time for another round of fixes and tweaks.

1.0.2 BETA changes:
Bugs:

Client:
-Better handling of inline widget mouse-capture. Capture should no longer interfere with grabbing the edge of an overlapping dialog (such as resizing the settings dialog over the executable column).
-Inline widgets now correctly paint a hover-action linked with the rest of the list.
-Fixed a minor memory leak with repeated connections.
-Fixed an issue where the executable column (process list) on occasion does not properly auto-fit on the first connection.

Server:
-Fixed an issue where the server would on occasion (very rare) not close a process handle.

Additions or Misc:
-Added support for custom font-size in process list and log views.
-Added "List > Auto-Fit" menu action. (This can be used to manually fit all data for columns if auto fit columns is disabled)
-Log filter feature is now byte-code compiled and much faster on large log runs (of 10k+ entries)
-Out of bounds client settings for a field are now reverted to their default if the settings page is closed. The prior behavior was to just color bad values red.
-Minor GUI and font rendering tinkering.
-Updated build environment (GCC 7.2)

Code:

SHA1                                              Size  Name
---------------------------------------- -------------  ------------
5EFAB3AC1EDEAB46E181E6A82D9E998D70C1CE5F       4330927  YAP_BETA_Setup.exe

The newest installer can be attained from the downloads page here.

Once out of the initial betas and once there's an auto updater in there, I'll start keeping each individual version online as was done before with ClockBlocker.

 

Thank you for ClockBlocker. I have a question about a specific problem that I am trying to solve. My problem is that my GPU fluctuates from core clock to boost clock (950MHz to 1072MHz). I noticed that applying the ADL method and allowing it to fire once (i.e. opening a 3D app), then closing the ClockBlocker completely solves my problem (the clock speed under load remains at 1072MHz, never dropping down to the "core clock" of 950MHz, and still downclocking when idle). Though leaving ClockBlocker open with ADL mode does work, the reason why I'm asking is because alt-tab to the desktop during load with ClockBlocker enabled causes the core speed to drop and causes a brief lag before the clock speed recovering and I would prefer the least amount of overhead possible, understandably. I wonder if you could elaborate on what you are doing with the ADL, and maybe if you could create a mode that simply applies this ADL function once, as this would solve my throttling problem completely. Many thanks for your hard work!

 

Okay, maybe I have the problem with UVD bug. I am updated to the most recent v18 drivers with my gigabyte 280 windforce, I wonder why amd doesn't fix this for me (it's been happening since fcking 2014). I might have to flash a custom bios for my card with increased UVD clock.

 

Hey there,

The ADL helper works similar to RadeonPro. Since this approach is generally attributed to its author's effort, I figure I'll give him credit in quoting the post on how RadeonPro's blocking works earlier in the thread.

While ClockBlocker's helper can use ADL7 / next, works on newer video-drivers, and also does a few more things (such as looking for jammed clock-states caused by WattMan's clock-retention between boots and driver-crashes), the idea is the same and the changes are just adaptations to the newer-drivers.

In jist the idea is that lower power states are eliminated from selection. Unfortunately from driver 16.12.1 onward things got a bit more complicated as the lowest power-state became locked from adjustment (at least on the Fury series [Fiji], and shortly followed by more hardware). For the older generation cards (200 and 300 series) it took a bit longer, yet ultimately most AMD hardware was migrated to the newer power-management scheme [WattMan], and thus now behave similarly.

In terms of ADL usage, if you disable the watchdog polling in ClockBlocker's settings then the ADL helper is a like an on / off switch. eg, it doesn't actively do anything until the controlling app (the main ClockBlocker process), makes a decision to disable blocking mode.

**With alt-tabbing the game process, this probably is causing the ADL helper to be closed and the original clock-states restored. (should be indicated by the tray-icon of ClockBlocker, reverting to "yellow" instead of "red")
-- You could workaround that with ClockBlocker's profile system by manually adding a rule for the game process that's always-active while it's running. The 3D-process and full-screen detections are special rules that only apply to a foreground task.


It should be pointed out that with new-drivers that the ADL method probably is not sufficient (on its own) to block clock change anymore due the first / low-power state. The Compute Method on the otherhand doesn't touch the power-states and doesn't have this problem, it can also be used in conjunction with the ADL approach to try to keep a card from entering state-0. [although the compute approach does have a small overhead and is always active]

Depending on the game you may find you get better frame-rates with the compute method despite its overhead (just from clocks being kept consistent). Similarly in something such as Crysis 3 (which already would keep the clocks pegged), it may slightly hurt performance.

 

is the r9 380x supported?

 

Both ADL and Compute implementations are intended to be generic and "should" work on virtually all AMD drivers and hardware. Key word is should there though as I don't personally own an R9 300 series card that I can test with.

Compute method is generally the safest (as it doesn't touch any hardware settings, and doesn't remove lower-clocks) -- it also works on NVidia graphics cards, Intel IGP's, etc [almost everything]. This is also the default approach used within ClockBlocker, and the one that's currently integrated with "YAP" beta releases. People have used ClockBlocker on a very wide range of hardware, so I expect it'd work on yours.

 

I tried clockblocker on my two Vega's but it only works in menu's (low load) is there something I mis, because ingame the core clocks drops to normal levels, ie 1550-1600 instead of 1750 which is the max for my Vega Liquids? Note I not reaching max fps with vsync.

Edit, should have read first, coming back after that

 

Yap, it's time for another update. (of YAP)

This version's focus is purely polish. That means getting memory consumption of YAP (server & client processes) completely under control, crushing all memory leaks that I find, and fixing bugs in general. So, this version hopefully is a large-step towards 24/7 suitable code-stability.

While YAP BETA 1.0.2 was already pretty stable for gaming-rigs, it does have memory leaks that might've been a problem for systems left on for days straight [or that launched / terminated many tasks]. These leaks should now be gone, so please report any suspicious memory usage! (especially on the server service)


I am leaving documentation on my TODO list for later on, as I am hoping to get some more feedback on whether or not that GUI is intuitive / how it can be improved.

Cheers all!


YAP 1.0.3 BETA changes summary:

-Netcode improvements (reduce bandwidth and CPU-cost)
-Polling improvements (reduce memory and CPU-cost)
-Circular-log huffman coding, tighter packing (reduce memory)
-Reduced circular log limit (prior limit could cause too long client-connect / timeout)
-icon compression (shaves a few megabytes from the client and server)
-Revised more aggressive heap-manager
-Fixed a severe handle leak when a session-0 account is not available (problem spawning bridge process)
-Fixed dozens of slow memory leaks (on log addition, process addition, etc)
-Fixed the insertion case in editing rules when there is no selection
-Fixed improper window restore position when clicking the tray icon too quick (faster than fade-out / in)
-Fixed the occasional restore delay
-Fixed the InstallShield "file in use" (install-over) issue.

Code:

SHA1                                              Size  Name
---------------------------------------- -------------  ------------
BA000329ACF12386B22BC720D2AE4A794BC2F1DC       4142234  YAP_BETA_Setup.exe

The newest installer can be attained from the downloads page here.

Next features to be added will probably be a Session-Manager (for the client), a connected users list (with ability to easily add them to firewall "BLOCK"), and rule junctions (to avoid redundancy).

 

YAP version up!

Code:

SHA1                                              Size  Name
---------------------------------------- -------------  ------------
08E665D3C37C93516F05DFC24DBFA965D7FA9D15       4201998  YAP_BETA_Setup.exe

YAP Doc / Download


YAP 1.0.4 changes summary:

-Added (File menu) site-manager. [Ability to save and swap between managed machines by profile] --Loading of a site is done from the File > Connect dialog.
-Added (Rules > Server Connections) client-list (monitor recent connection attempts), with easy-blacklisting of IPs
-Added (File > Settings > Misc) auto connect option for clearing prior connect settings on client-start
-Added ability to change server server passphrase and port. ** See Notes
-Added CPU averages, rather than just instantanious (short interval) monitoring. Smoothed large time interval monitoring provides a stable way to sort the process-list by CPU utilization.
-Added rule-editor "is divider" checkbox -- to add more clarity on what are dividers and to convert rules "to & from" a divider.
-Added "status bar"
-Fixed a cause of erroneous CPU time readings
-Fixed overflow of prompts in performing actions on more than 10-items. (these will now display as "... and more ..." after the first 10)
-Eliminated the flicker with the log-view widget when resized 'exactly' to the pixel that scrollbars would show or hide.
-Eliminated the flicker during resize of all settings dialogs when textbox widgets are selected.
-Rule editor will always open scrolled to the top when changing rules
-Explicitly blacklisted IP's are now suppressed in the log. (ending with "block => *.*.*.*" in the firewall is sufficient to suppress all rejected connections)
-Bumped network protocol version (past YAP clients are now incompatible and vice-versa, with new client connecting to old servers)
-Various spelling corrections
-Various other widget tweaks, rendering tweaks, data-formatting changes, and bug-fixes

Notes:
** Changing the server service port & passphrase:

-Create a file named "credentials.txt" (ASCII, utf8, or utf16_le encoding accepted) with the following two lines of content:
//=================================================================
Port = 9001 // Sets the server's port number
passphrase = MyPassphrase // Sets the server's encryption key (please don't actually use "MyPassphrase")
//=================================================================
-Move the created credentials.txt file to "C:\Program Files (x86)\ComroeStudios\YAPM\"
(This should require administrative rights to access Program Files)

-Stop and restart the "YAP Server" service. This can be done either from a console or the services.msc GUI.
from a console:
sc stop "YAP Server"
sc start "YAP Server"

On the next startup of the YAP service, credentials.txt will be removed from the directory. Your new server credentials will be read and merged with the server's configuration.
Remember to update your client's site-settings accordingly! (These will not be changed automatically, even for a YAP Client on the same system)

It's strongly recommended to verify that credentials.txt was actually removed, as this confirms that the file was in a readable format.

Documentation:
-added page on basic firewall use
-added page on changing port and passphrase
-altered download page, I take back what I said on not allowing changing the passphrase and port during the beta -- Let's do some network testing!

As 1.0.3 is very stable (as a beta), I am not removing "1.0.3 BETA" and have started archiving versions as with ClockBlocker incase anyone has issues.
--All beta versions onward will be available in the archive section here. Beta versions will probably be removed once we're at a stable release, especially once there's an auto-updater inserted.

NOTE: I am using "stable" loosely and synonymously with non-beta. Currently YAP I believe to be fairly close to 24/7 stable / fully usable.


Rule junctions did not make it in this release:
--Next version's focus will be Rule-Junctions and Rule-Conditionals.


Known issues:
-Process Count is one less than displayed by Task-Manager.
**The difference is that YAP-Server doesn't count the "System Interrupts" entry. (This is working as intended)

-There's an extra-bracket visible in the screenshots from the firewall documentation. "before server submission)" <-- will be fixed in the next release, though not rolling out a hotfix for a typo.

(workaround attempted -- 1.0.5) -YAP client crashes on some Windows 7 installs immediately on startup. [right after the splash-screen] Other YAP processes may randomly crash. The cause is "DEP".

Seems to be related to LDR updates and people who have pushed their OS install further than Windows Update would normally take it. (with third party tools such as KUC updater) DEP behavior on these machines is odd ... certain OS installs apparently consider YAP (as well as other some other 64bit softwares) to violate DEP. Meanwhile other machines have no problem.

The code that's causing the DEP crashes seemingly has to do with conversion of system BSTR's to local strings, and specifically how this code is generated with modern MinGW / GCC. While there's also nothing wrong with the code and it works on 100% of Windows 8.1 and 10 systems, as well as many Windows 7 systems.. Removing these routines, building them in static libraries with an OLDER compiler (GCC 6.3), and then linking against said libraries in building the rest of the project [with GCC 7.3] works-fine. [on the Win7 installs that are erroring]

Some parts of the project are using C++17 features, unfortunately. Moving back to an older compiler is not an option... Annoying, but this will do as a workaround to build the YAP project in pieces with different compiler versions, then link them all together.

 

Code:

SHA1                                              Size  Name
---------------------------------------- -------------  ------------
5B94102CC65FD9EBCA35C2B4AF2EB239FA4A58B5       4107861  YAP_BETA_Setup.exe

YAP 1.0.5 BETA (hotfix):
-Attempt to workaround Windows 7 DEP crashes


Description provided in edit to prior post known-issues. This one is going to be a PITA to address, yet is being worked on and now being tested on a misbehaving Win7 install.