Problems? See this thread at archive.org.
Is there any way to disassemble the bios file?
-
OK, thank you again. That clears it up.
Not as of yet, we need a new tool deal with the newer BIOS format, and none have leaked. I've searched for weeks and found nothing, only some russians claiming to have a newer version which I cannot find. Apparently it could be a hybrid BIOS/EFI format, and we need new versions of amibcp and/or mmtool to work with it.
Reference: http://forum.notebookreview.com/showpost.php?p=5436618&postcount=6 -
i've been fairly busy with school, but here's a new set of data for benching 196.34. verdict: 195.55 is still the king.
run1 [10473] very short stuttering in SM2 test 2. full brightness, iphone charging, full kbd, razer mouse, stock asus clocks:
![[IMG]](images/storyImages/19634-stock15.png)
run2 [10556] no stuttering. 1 brightness:
![[IMG]](images/storyImages/19634-stock1.png)
run3 [11200] visible slowdowns in SM2 test 2. 1 brightness, 550/1375/950 clocks:
![[IMG]](images/storyImages/19634-regoc1.png)
run4 [9263] no visible slowdowns. not exactly sure what was happening in this one, everything was just slower by a lot. full brightness, 550/1375/950 clocks. throttlestop settings: clock modulation @ 100, chipset modulation @ 100, set multi @ 13.
![[IMG]](images/storyImages/19634-regoc15-tscm.png)
run5 [10345] visible slowdowns in the usual parts. slowdowns correspond with a drop in multiplier (as reported by throttlestop) and GPU usage cut in half. full brightness, 550/1375/950 clocks. throttlestop settings: clock modulation @ 100, set multi @ 13.
![[IMG]](images/storyImages/19634-regoc15-ts.png)
turning on 'set multiplier' in throttlestop makes the reported multiplier go berserk, eg. no synchronization on any core and all jumping between 8 and 12.
gl and hth. excellent program btw.Attached Files:
-
-
-
yea, the vx doesn't have a 45W CPU
for anyone who follows this thread, i may have found a real-world application where this throttling affects performance and a solid reason for asus to issue a response that won't include "this only happens on synthetics, not in everyday use". Dolphin. i'll be making a summary of my testing shortly. -
tmonitor might not be using the proper method, but in the end i use performance and temperatures as the final judgement. the logs are only there as formalities.
below is the log of what happens + screenshot. in general the multiplier is all over, load is <10% (whereas task manager says <2% at any given time). i am aware that it switches many hundreds of times a second, so it may not be an issue at all (seems to be at 7x a lot more on XP), but with the clocks jumping around like that it takes a bit more to point out throttling.
the only reason i said it was is because i'm not sure if 2 logical cores mapped to the same physical core can have different multipliers.
edit: added a short log from xp.
edit2: prime95 gets all cores up to 12.75 on win7. it could be something with my install, as i know it's not exactly "stable". i went through and disabled all services i thought unneeded (homegroup, wmp sharing, search, dns cache to name a few) and probably some experimental bluetooth driver stuff that (shouldn't) be there.
edit3: realtemp shows 933, 1000, and 1066 all apx. 1/3rd of the time (win7).Attached Files:
-
-
i assumed that if there are two threads running independently on two logical cores mapped to the same physical core, then one active would imply the other active as well. mistake on my part (wouldn't make sense).
i'll rerun the tests with throttlestop, hopefully to some visible effect. for a suggestion, having a multiplier column for each logical core in the log would be nice for times like these perhaps to the right of DTS. the green = off and red = on could also get confusing for people who rely on the tray.
+1 for the explanation (must spread some around first, but i won't forget). too much time on XP made me used to seeing it at min for most of the time
edit: and yea, setting it to Always On produces the same effect as on win7. -
I faced the double negatives since day 1 on this project. Does throttle stop on mean you are telling your computer to throttle or telling it to stop throttling? What about throttle stop? Does stop mean you are stopping the throttling or stopping the program.
It's a big confusing mess. Same with the colors. Red is the color of a stop sign so it originally made sense that the system tray would have the same color. When I added a second color to the system tray I didn't want to do a color swap or previous users would be confused. I should have put more thought into this originally but too late now.
The multiplier in the log file is the multiplier on the hardest working core. This gives a good indication of how hard your computer is working so if only a single thread is working hard then it will report the multiplier for that core or thread. You don't need to see all the threads in the log. For most users it would just become too much information. If you see a 7.0 in the log file then that means that no other core is higher than that so throttling is obvious.
I might add an option to report all of the multis in the log but it would be a user selectable option in the INI file or something like that so no one gets over run with numbers jumping all over the place in the log. -
as long as "throttlestop is enabled" -> stop throttling, i have no objections
from the pattern i saw during dolphin, it doesn't set the CPU all the way down but instead limits it slightly to 10-12x on two threads. sometimes it changes with one at 7 and one at 15, but reporting 15 makes it seem as though there's nothing unusual going on even though performance takes an even bigger hit.
if it takes more than an hour to implement don't worry too much, it's a unique case that would most likely ever be used by me. it took asus 3 months to fix the BSOD problem, and the new BIOS still isn't on the official site (hoping they change the limit or remove the throttle altogether...). -
It's OK if the multipliers are dropping down on cores or threads that have a low C0% since those threads are spending most of their time asleep. The average multiplier for them will be lower.
The multiplier reported in the log file is always the one from the hardest working core which better represents how hard the CPU is working. Post a screen shot or better yet a log file when you're doing the dolphin.
-
TMonitor uses the method outlined in the Turbo White Paper and is using the same timers that i7 Turbo uses so in theory the two programs should display the same results. When the CPU is lightly loaded, it is fudging the raw data and can report a multiplier much lower than what the data says the multiplier is at.
Here's an example on a Core i7:
![[IMG]](images/storyImages/tmonitorii.jpg)
RealTemp uses the same code as i7 Turbo and ThrottleStop. At idle, it reads these timers, sees that both timers for each thread are running at the same relative speed so this indicates the CPU is using its default multiplier which is 23.0 for an i7-950. TMonitor sees that the CPU is idle so it assumes that SpeedStep is enabled and the multiplier is at the minimum 12.0. That's not what the internal timers are saying according to the method that Intel recommends.
Here's an example with a Core 2 Duo Desktop CPU.
![[IMG]](images/storyImages/tmonitor1.png)
Based on the internal timers, ThrottleStop reports that the CPU is using its default maximum multiplier of 10.5. TMonitor shows that it is using the minimum 6.0 multiplier even though the Minimum processor state is set to 100%.
![[IMG]](images/storyImages/tmonitor2.png)
In the second picture, the Minimum processor state is adjusted to a very low number which lets the multiplier drop down at idle. ThrottleStop, using the internal timers, reports the 6.0 multiplier. TMonitor also reports 6.0 but it reports that regardless of the Minimum processor state setting.
![[IMG]](images/storyImages/tmonitor3.png)
In the last example I'm using my load tester to fully load the first core. The internal timers (ThrottleStop) are saying that both cores are at the maximum multiplier of 10.5. TMonitor shows one core using the maximum multiplier while the other core is using the minimum multiplier of 6.0. That's not what the timers are saying and to the best of my knowledge, that's a physical impossibility for a Core 2 Desktop CPU. The regular dropouts in the Core 1 TMonitor graph are also not accurate given that the first core has a constant 100% load on it.
I apologize for getting off topic but too many people are putting too much faith in TMonitor. It may be 100% correct but if it is then Intel's documentation must be 100% wrong. I'll believe Intel before I'll believe TMonitor. -
Thanks.
I've seen the momentary one-or-two-bar-wide "speed" dips and rises in Tmonitor, I guessed it was background processes waking up the other cores and lowering turboboost as a result, or just momentary brainfarts, so I guess I never assumed it to be 100% correct anyway (not when it records frequency as >5GHz fr example).
I understand you don't know what's going on inside Tmonitor, but from what you say it seems like it's not misleading me (and perhaps others) too much in what we think is going on due to the throttling stuff.
---
Interestingly, I noticed something regarding turboboost and the power4gear overclock profiles:
Doing a 1-thread Prime95 (locked to core 0), the average multipler was:
No turbo: ~19.1
Turbo: ~19.0
Extreme Turbo: ~19.3
...now I would have thought it would be *lower* if anything. I would quantify but I've just received a 2 week legal deadline to change our website so I may be a tad busy
-
When running a benchmark like Prime95 Small FFTs, there shouldn't be any "2 bar dips" that TMonitor constantly shows. That benchmark puts an incredibly consistent amount of load on a CPU core. There are no significant dips like TMonitor constantly shows. I was getting those dips at a constant frequency of about once every 5 seconds. That tells me something is not right. (It was exactly 4.6 seconds when tested with a stop watch.)
When I find a monitoring tool that has too many brain farts or shows me data that is not accurate then I can't depend on anything it is telling me. A lot of the time at full load it is probably very accurate but there are other situations when it is not. The data coming from ThrottleStop is very consistent and I know that it closely follows the method that Intel recommends.
The amount of turbo boost you get during any benchmark depends on how many active cores there are during the test and it is impossible to control that because there is no way to control all background processes. There is always going to be some random variation in the average multiplier during a single threaded benchmark like Super PI. I'm not sure how much variation should be considered normal. That would vary from one computer to the next depending on what background tasks they have running including what antivirus software they are using.
Statistically speaking, your results lead me to believe that all 3 power profiles are probably exactly the same and the difference is just normal variation due to background activity which causes turbo boost to constantly engage and disengage depending on the number of active cores.
No more testing. It sounds like it is time for you to use your computer and to get to work.
-
interesting, i haven't noticed the drops. not during prime95 at least.
temperatures and unexpected drop in performance are still the best way to check for throttling. both programs report 7x occurrences correctly, but hopefully the CPU will cooperate enough that i can use throttlestop for the partial-throttling in the emulators.
one more major assignment due at 10pm tomorrow and then it's game time (no reading in my "reading week")
i modified the throttle and BSOD wiki pages a fair bit, in case anyone follows them. what we need are ideas+screenshots on other programs/tests that would have this in normal usage. -
Oh don't worry, in the last month on top of all the normal stuff I've played through Borderlands twice, Clear sky (laziest lamest sequel ever...), gone about 1/4 through Far Cry 2, spent quite a few hours trying to port all the scripts/mods/packs across from the install of X3 Terran Conflict from the old lappy, watched all of Yes Minister & Yes Prime Minister, a bit of Scrubs, a lot of The Shield (fknel that was a good show) and the pretty lame 12th and 13th seasons of Southpark and thats just on this laptop
Sleep, what's that?
-
-
I'm pretty sure I experienced throttling when playing Borderlands while zipping GBs of stuff with 7-Zip (normal compression) last night.
I didn't see anything abnormal in terms of fps dips (like most fps' there are stutters with huge explosions going off right next to you with heaps of smoke etc etc) and Rivatuner graph showed pretty consistent fps
It was only when I quit, I saw the Rivatuner log of "CPU Frequency" showing heaps of dips to 933. Now it may just be misreporting idle/low util% speeds so I'd very much like to doublecheck it later, but all other times playing games I've not seen speed dips like that in the logs, but then all other times I've not been doing 7zip in the background.
?
-
[Fixed/Workaround] Asus G51J(x) CPU throttling investigation
Discussion in 'ASUS Gaming Notebook Forum' started by thalanix, Jan 20, 2010.
