NOTICE: This forum is mostly an archive, though new posts are allowed. Registration may require manual admin activation. After registering visit https://bitsum.com/contact/ to request account activation.
Started by bertie97, August 10, 2012, 02:23:15 PM
Quote from: TfH on August 12, 2012, 11:35:56 AMI might remember wrong (waked few minutes ago, 6.30pm here) but I remember reading somewhere that disabling that option can have undesired effects to SSD/HDD performance. Don't ask me why, I don't remember.
QuoteThe C3 Auto Demotion and C1 Auto Demotion options have been shown to increase 4K Write speeds on SSD hard drives significantly. When running on battery power, leave these unchecked and set the C1 button in the picture as high as it can go so your CPU can enter the deeper C sleep states and conserve power.
QuoteSo don't take this reply as "pro" reply, being half wake and having first Java cuppa
Quote from: TfH on August 12, 2012, 03:53:43 PMThanks Ben OT: I always wake up between 5pm->7pm and go to sleep between 8am->10am
Quote from: bertie97 on August 13, 2012, 09:06:53 AMThanks guys.I have heard that Core-parking is a Win7 function that affects multi-cores - AMD & Intel. I saw a post specifically about reg hacks for CP, AMD & WoW. So I figured it must apply to my PC. I do not know definitely though.I have noticed since disabling CP in probalance that the PC is faster (illusion? Maybe PL#6 is just doing a better job), it certainly seems much quicker in ToW2 than it was. Even when I am in a <200 units battle it's smooth. (The game sims every bullet/shell fired, trajectory & collision/richochet event - so the CPU takes a serious thrashing.)
QuoteIf you look at your CPU utilization and see 15%, that is actually the percentage of time the CPU was active within a relatively large interval (usually 1 second). Actual CPU utilization most often occurs in micro-bursts where the CPU is fully consumed for a few nanoseconds or microseconds. You want those bursts to execute as fast as possible, and this is why all power saving technologies, including frequency scaling, incur *some* performance hit. This is also why a faster CPU does matter, even if you typically don't utilize 100% of available CPU time over a larger interval.
QuoteAdditionally, I've found Windows CPU Parking to be excessively aggressive by default. In one test bed, an AMD Bulldozer platform (8150), the scheduling problems were *very* apparent. Turning off CPU Parking causes an immediate boost in responsiveness, particularly for the virtual machines and multimedia applications.
Quote from: bertie97 on August 13, 2012, 01:44:37 PMAs I am running AMD I have real physical cores - no HT.This is one reason I find PL useful as I am able to pipeline processes to what would possibly be unused cores with some software.Finding out that newly written s/w isn't targeted at multi-core - such as BFCs Combat Mission - is something that I believe PL can ameliorate.I spent a while testing ToW as I had issues (see my post about Theatre of War). Being able to offer more cores than the code initially targets seems to help.
QuoteI wonder how difficult it would be to hand-off 'surplus' queued threads from a prog written for a single core to multiple cores. By which I mean have software written for a single core access multi cores as one 'big' single core.I don't know enough about header flags etc to hypothesize in any depth, but it would be cool to do it for stuff that is coded for 1-2 cores when the CPU has 4, 6, 8, 12 cores etc..Spoof core ID on the fly to the process queue perhaps?
QuoteAdditionally, parking CPUs can potentially reduce core thrashing - when a thread is swapped between cores unnecessarily, causing a performance hit
QuoteBy which I mean have software written for a single core access multi cores as one 'big' single core.
QuotePL 7 to the rescue?
QuoteMany threads are swapped between cores unnecessarily, so the core need reread/write data from/to L2/L3, and causing cache miss to other threads as the capacity is small.
Quote from: bertie97 on August 14, 2012, 08:30:48 AMIt wasn't a question of speed that concerned me, just the 'appearance' of available bandwidth across a 'single' core.i.e. if one created a virtual wideband core from multiple cores in terms of available addressing for the routines being piped to it.I don't know how you would cope with things like load-balancing, queueing & synchronicity of processes as I am not a software genius this is just a bit of lateral thinking.
Quote from: bertie97 on August 14, 2012, 02:22:31 PMBack on topic.... Having been forced to play several hours of games to test PL core parking - I can say my overall performance has improved with the PL over-ride/management.Once it is fully tweaked I am expecting an even better overall experience. I am also interested in the possibly coincidental (?) reduction in temps. Not much, but a 3-7C degree drop definitely. Is spreading the load helping to balance localized core over-heating I wonder?It's one of those settings that I would otherwise not play much with but I am quite surprised at the overall impact on the 'feel' of the PC.Don't know what I'd do without PL making it easy to alter the environment.All this talk of cake & pizza, whats next? Beer?
Quote from: bertie97 on August 14, 2012, 08:30:48 AMAs with your cake (& now I'm hungry)
QuoteIt wasn't a question of speed that concerned me, just the 'appearance' of available bandwidth across a 'single' core.i.e. if one created a virtual wideband core from multiple cores in terms of available addressing for the routines being piped to it.
QuoteMy understanding is that multi-core allows one to hand off processes to dedicated cores. (So a dev can allocate each software component to a set of physical resources within his codebase.) Then it all has to be synchronized to events.
QuoteIt would be nice if software specifically written for 1 or 2 cores could deliberately take full advantage of all available cores in an optimal way.Taking an 8 core & creating 2 virtual cores (each having access to 4 real cores) could provide extra bandwidth for the software in my (uninformed) theory.
QuoteHowever, to calculate e, you need the results of both threads. So part of the algorithm can be parallel, but another part is implicitly sequential. It depends on results from earlier calculations, so there is no way to run this calculation in parallel with other dependent calculations.
QuoteAmdahlâ€™s law deals with these limitations of parallel computing. In one sentence, it says this: QuoteThe speedup of a program using multiple processors in parallel computing is limited by the time needed for the sequential fraction of the programThe sequential parts result in situations where threads for one step in the algorithm have to wait for the threads of the previous step to signal that theyâ€™re ready. The more sequential parts there are in a program, the less benefit it will have from multiple cores. And also, the more benefit it will have from the single-threaded performance of each core.And that brings me back to the original point: people who think that the number of cores is the only factor in performance of multithreaded software.
QuoteThe speedup of a program using multiple processors in parallel computing is limited by the time needed for the sequential fraction of the program
QuoteNow wondering how to get back on topic
Quote from: Victek on August 14, 2012, 01:56:21 PMThis is a bit over my head, but I am interested in learning more about core parking to the extent that it can be managed in Process Lasso. It would be nice to know what the advantages and disadvantages are for various scenarios. It sounds like Jeremy is working on it. There hasn't been an update for, like, two days though - maybe someone needs to run out and get a pizza? LOL I'm sure we'll be hearing from him soon
Quote.....So PL7+ needs to hack Win CPU scheduler
Quote from: bertie97 on August 14, 2012, 02:22:31 PMIt's one of those settings that I would otherwise not play much with but I am quite surprised at the overall impact on the 'feel' of the PC.Don't know what I'd do without PL making it easy to alter the environment.All this talk of cake & pizza, whats next? Beer?
QuoteResearchers from Intel Labs Barcelona have developed â€œAnaphaseâ€ technology, which is a novel hardware/software hybrid approach to leverage multiple cores in order to improve single-thread performance on multi-core processors. This research focuses on different speculative techniques to automatically partition single thread applications to be processed on multiple cores.
Quote from: bertie97 on August 15, 2012, 01:24:46 PMBenYeeHua - the scalibq is a useful blog article for clarification of CPU interaction on threading.
QuoteMy cores are dynamically clocked by the AOD software - hence the temp. change range I guess. My impression is the PC is smoother & behaving more dynamically, & I am inclined to thank PL for that. That I suspect has a knock-on effect to the heat generated - useless cycles & thrashing being avoided where possible.
Quote from: Victek on August 14, 2012, 07:15:55 PMI thought the drink of preference was Mountain Dew? Anyway, I expect Jeremy will jump in soon with some new information.
QuoteI have noticed since disabling CP in probalance that the PC is faster (illusion? Maybe PL#6 is just doing a better job), it certainly seems much quicker in ToW2 than it was. Even when I am in a <200 units battle it's smooth. (The game sims every bullet/shell fired, trajectory & collision/richochet event - so the CPU takes a serious thrashing.)
Quote from: bertie97As I am running AMD I have real physical cores - no HT.
Quote from: BenYeeHua on August 18, 2012, 03:24:29 PM1.Using newest VS2012 is getting performance, but losing compatibility as user need to install the update?
Quote2.Did Intel compiler still not supporting the new instructions(AVX, for example) for non-intel CPU?
Quote3.Will ParkControl become more functional?
Quote4.The disabling of core-parking in ProBalance working >90% of computer(that support core-parking)?
Quote5.Using HPET can getting more performance for gaming and other software?(I only know that changing this for lose-control for speed of software/game when overclocking FSB)
Quote5.What do you think about Piledriver before it came out?
Quote from: bitsum.support on August 18, 2012, 04:46:19 PMMicrosoft decided to restore support for XP / 2003 in the end, BUT it won't come until some yet-to-be-announced update 'later this Fall'.
Quote(Maybe the 4 thread swapped to 2/4 cores most of the time as it is just unparking some fake-core, making the core usage not balanced).
Quote..... RESPONSE #2 is PENDING.
Quote from: bitsum.support on August 19, 2012, 01:37:39 PMSorry for the wait ;o. I will catch up and finish as soon as I issue this next minor update. It is very important, a lot of people have been waiting on it. Heck, the beta didn't even work in XP for the last 3 days, and didn't work in XP x64 up until now. I've got to test everything pretty rapidly and get it out, as some are seeing crashes in the current final (with minidumps). Of course, I can't release it until I think it is ready. Since VS2012 went final, that is no longer a blocker, but I still must do some more testing, and there are some other known issues I want to address.
Quote from: bitsum.support on August 22, 2012, 10:11:03 AM... and then my anniversary came, so had to tend to the wife . I will finish this discussion today. I can finally get back into my routine here, whew.
QuoteThe maximum percentage of logical processors (in terms of all logical processors that are enabled on the system) that can be in the unparked state at any given time. For example, on a system with 16 logical processors, configuring the value of this setting to 50% ensures that no more than 8 logical processors are ever in the unparked state at the same time. The Core Parking algorithm is disabled if the value of this setting is not greater than the value of the Processor Performance Core Parking Minimum Cores setting
QuoteExamples are device driver I/O, the kernel itself, and other internal components.
Quote from: BenYeeHua on August 22, 2012, 03:31:12 PMAnd hope you having a some nice days on trip
Quote7.Can I making core-parking more smartly(and disabling fake-core, EIST etc)? (Because it is "some" important, so I repeat asking 2 times)
Quote8.When I disable core-parking, it still going C-state right?
Quote9.What benefit can I get from memory trim?Faster GC?and it can be using by other process?(Just pushing into Standby list, not pushing it into page-file)
Quote10.Changing AdditionalCriticalWorkerThreads and AdditionalDelayedWorkerThreads to 0x10 (16), what can I get?Less DPC latency?
Quote11.Can I using this(Interrupt-Affinity Policy (IntPolicy) tool) to reduce DPC latency?(by reducing swapping when disable core-parking)
Quote12.And why the responsiveness of PL and CPU eater(when running it) will decreasing by time, but it still responsive at it should(I found it become smooth, it was lag before, I don't know this is because it is 1ms timer resolution, or after changing into HPET)And your forum is missing the function to upload attachments, you better find a time to fix it
Quote7.As this picture, I found that core-parking having so many function, is that safe/can disabling using of fake-core by changing core-parking max core?And if I can, making core-parking more smartly?
QuoteSo it can be parking forever for fake-core.Right?
Quote8.Core-parking just making it sleep faster(going deeper C-state faster), but when I disable core-parking, it still will going C-state right?
Quote9.What benefit I can get from memory trim?
QuoteAs I know, I can using trim to reduce the memory used by process(not push it into page-file, is push it into Standby list)And I tried trim Standby list(by using RAMMap), and it is pushing into page-file.So that system is pushing the useless data(Standby list)first, then data using by process, correct?
QuoteAnd did doing this (Just push it into Standby list) making software GC(garbage collector) a "bit" faster(even it having memory lacks)?(And the best way is restart program or waiting developers reducing it.)
QuoteTCP/IP one is the most in my memory, as SpeedGuide have collected data on it website
Quote1.The core-parking(ProBalance) still no working here, and it is making core-parking disable after reboot/changing PowerPlan. Only ParkControl apply real-time.
QuoteMore later, I promise . Your questions are in-depth and really take some time to answer correctly.
QuoteI will investigate this in today's work, I was not aware this to be the case! EDIT: I found the cause, and will be fixing it in the next minor version. EDIT2: FIXED. Note that it will actually work in some cases, but not immediately as it should in all cases. I am confident in the fix though, it was an oversight. After review and few other odds and ends I will publish the update.
QuoteEDIT: Do note that the governor may need be to running escalated in most NT6+ environments.
Quote from: bitsum.support on August 11, 2012, 06:13:14 PMFirst, I do see the issue with the dialogs. I am looking into that for the next update. They may not correctly reflect the Enabled/Disabled state. 100 means disabled and 50 means enabled.
Quote& to continue the red herrings - I can only hope that BenYeeHua was given some Mountain Dew for ä¸ƒå¤•èŠ,
QuoteAlso need to test some video encoding I think.
QuoteWas wondering in ref to the above what does 0 represent as a value in the dialog? Neither 1 or the other?
QuoteJust updated to beta x.69 & am going to try some gaming with CP set to disabled & see if anything feels different.
QuoteIn its' current guise PL is helping me maximize the performance I'm seeing so I'm happy. The core activity graph is now showing much more balanced access/load across the group.I am now able to add items to profiles & assign CP state to those profiles so it's looking good.
QuoteAs I am running an ancient 1090 I don't actually have much knowledge of Bulldozer. I looked at it briefly & thought it was less of an advance than I would have liked. This may be unfair on my part but ... I really ought to take another look as the info above has piqued my curiosity.
Quoteif you see what I mean.
QuoteTime vs performance per CPU?
QuoteOne reason for this is that the intended proper behavior of a Windows application is not what is seen in the real world. They are meant to do a brief burst of CPU use, then enter wait state. Instead, they often are poorly designed and end up in live-locked and other 100% CPU use situations. It is impossible for Microsoft to know for sure what application really needs that CPU and which doesn't. With a little 'market intelligence', a little logic, and some help from the user Process Lasso's ProBalance can do better -- saving your PC from a total stall or simply increasing responsiveness.
QuoteMission statement vs 'reality'
Quote from: bitsum.support on August 26, 2012, 01:59:58 PM@BenYeeHua: There may have been a rush to final and this was missed. I'll fix it right up ASAP. The code wasn't missed, so there must be a malfunction of some sort.
Quoteis there any possibility that your previous setting had cpu parking disabled?
QuoteFirst, sorry for any lost posts. I had some server issues here. Web server administrator is one of my several full time jobs ;p.