From TAYW0034 at e.ntu.edu.sg Fri Jun 1 01:35:56 2012 From: TAYW0034 at e.ntu.edu.sg (#TAY WEI CHOON#) Date: Fri, 1 Jun 2012 08:35:56 +0000 Subject: [Hotspot] Hotspot mechanism Message-ID: <9FC1D3EE5F47314887481926A2836C91057EECF2@HKNPRD0111MB385.apcprd01.prod.exchangelabs.com> Dear Sir, I would like to find out the mechanism of the Hotspot. For multi-layer structure, does it simulates the overall 3D structure? Or it simulates every 2D layer? if its through a 2D layer by layer, how is the heat transfer from one layer to another? Regards, Shawn -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120601/d4b1755c/attachment.html From skadron at cs.virginia.edu Fri Jun 1 05:48:04 2012 From: skadron at cs.virginia.edu (Kevin Skadron) Date: Fri, 01 Jun 2012 08:48:04 -0400 Subject: [Hotspot] Hotspot mechanism In-Reply-To: <9FC1D3EE5F47314887481926A2836C91057EECF2@HKNPRD0111MB385.apcprd01.prod.exchangelabs.com> References: <9FC1D3EE5F47314887481926A2836C91057EECF2@HKNPRD0111MB385.apcprd01.prod.exchangelabs.com> Message-ID: <4FC8BA04.90509@cs.virginia.edu> Yes, there are indeed vertical resistors between the layers. So based on your floorplan and layer information, when you initialize HotSpot, it constructs the equivalent RC circuit, with horizontal resistors between the blocks, vertical resistors between the layers, and capacitors at every node. Then it solves the circuit at every time step in your power trace (or just once, of course, if you are only doing steady-state). For more information, it is probably best if you read our papers. I would start with the ACM TACO'04 paper, then if you want to read further, the IEEE Trans. Computers '08. /K On 6/1/2012 4:35 AM, #TAY WEI CHOON# wrote: > Dear Sir, > > I would like to find out the mechanism of the Hotspot. For multi-layer > structure, does it simulates the overall 3D structure? Or it simulates > every 2D layer? if its through a 2D layer by layer, how is the heat > transfer from one layer to another? > > Regards, > > Shawn > > > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot From waqasch86 at gmail.com Wed Jun 6 04:09:11 2012 From: waqasch86 at gmail.com (waqas ch) Date: Wed, 6 Jun 2012 13:09:11 +0200 Subject: [Hotspot] Constriction/Spreading resistance Message-ID: Hi I am trying to figure out how the spreading/constriction thermal lateral resistances are calculated by hotspot, it seems that you are not applying the full formula from the ASME paper on constriction resistance calculation (published 1995), but just multiplying the material resistance per unit length with the shared length with the adjacent blocks. It would be great if you can explain a bit on this and refer to me the code file which does the full implementation for spreading resistance calculation. Regards Waqas -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120606/bf7030df/attachment.html From wh6p at virginia.edu Wed Jun 6 13:57:27 2012 From: wh6p at virginia.edu (Wei Huang) Date: Wed, 6 Jun 2012 15:57:27 -0500 Subject: [Hotspot] Constriction/Spreading resistance In-Reply-To: References: Message-ID: I took a look at the code since it was a long time ago when we last touched this part of the code. I think we did update the calculation with a simpler approach, because we found some discrepancies among different ways of computing constriction/spreading resistance. So, we divided the thermal resistance between two blocks that have different sizes into two parts, each part corresponds to the area that is shared by one of the blocks. Our validation results show this approach results in good accuracy. It also simplifies things significantly and makes the model more intuitive. If you are aware of an authoritative source of constricted/spreading resistance calculation, please do let us know. Thanks, -Wei On Wed, Jun 6, 2012 at 6:09 AM, waqas ch wrote: > Hi > > I am trying to figure out how the spreading/constriction thermal lateral > resistances are calculated by hotspot, it seems that you are not applying > the full formula from the ASME paper on constriction resistance calculation > (published 1995), but just multiplying the material resistance per unit > length with the shared length with the adjacent blocks. > > It would be great if you can explain a bit on this and refer to me the > code file which does the full implementation for spreading resistance > calculation. > > Regards > > Waqas > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > > -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120606/cfbe1836/attachment-0001.html From no-reply at dropboxmail.com Wed Jun 13 22:41:03 2012 From: no-reply at dropboxmail.com (Dropbox) Date: Thu, 14 Jun 2012 05:41:03 +0000 Subject: [Hotspot] yan jiaqi invited you to Dropbox Message-ID: <20120614054103.39A40B40535@sjc-batch3.sjc.dropbox.com> yan jiaqi wants you to try Dropbox! Dropbox lets you bring all your photos, docs and videos with you anywhere and share them easily. Get started here: http://www.dropbox.com/el/?r=/link/20.L2EVF0MNEy/NjIyMDQ5MjI4ODc%3Fsrc%3Dreferrals_bulk9%26eh%3D3c0b37b&b=clk:None:11444843878871228338:438:331&z=304d3a05f9 - The Dropbox Team ____________________________________________________ To stop receiving invites from Dropbox, please go to http://www.dropbox.com/el/?r=/bl/23dc285fbab5/hotspot%2540cs.virginia.edu&b=clk:None:11444843878871228338:438:331&z=5a7575c092 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120614/953cd4ef/attachment.html From bartosz.wojciechowski at pwr.wroc.pl Fri Jun 15 05:34:58 2012 From: bartosz.wojciechowski at pwr.wroc.pl (Bartosz Wojciechowski) Date: Fri, 15 Jun 2012 14:34:58 +0200 Subject: [Hotspot] problem with modelling cpu temperature - digital sensor error Message-ID: Dear All, I'm currently trying to model the thermal behavior of my i7 processor with HotSpot. However I have some troubles with thermal sensor readings. During idle periods there are differences of more than 10K between cores. During high load, with constant and equal load on all cores, the differences are smaller but still above 5K. I know at least one of the sensors is wrong, because the idle temps shown are smaller than ambient temperature (25 deg C vs 29 in computer case). But I'm not convinced that the sole reason of differences between reported core temperatures are the sensors. I know Digital Thermal Sensors have limited accuracy (intel whitepaper on DTS/PECI), but over 10 Kelvins seems a little too much of an error. The problem is that there are 3 possible sources of differences in reported temperature that I can think off: sensor error, differences in power consumption between cores and position of cores relative to other heat sources (floorplan). Does anyone know of any other possible reasons, and a way to mittigate this problem? If you could point me to paper on the subject or have any knowledge about possible sources of such temperature readings, please do not hesitate to write me :-) I've been trying to devise a sure way to validate the relative results from all the cores by using DVFS (and the fact that heat flows from hotter to cooler core, regardless of what the sensor shows), but my results are still inconclusive. This is not strictly HotSpot-related, but I hope it will be interesting to all those in temperature modelling. Regards, Bartosz -- grad student Wroclaw, Poland http://zak.iiar.pwr.wroc.pl From wh6p at virginia.edu Fri Jun 15 08:21:00 2012 From: wh6p at virginia.edu (Wei Huang) Date: Fri, 15 Jun 2012 10:21:00 -0500 Subject: [Hotspot] problem with modelling cpu temperature - digital sensor error In-Reply-To: References: Message-ID: Hi, It seems you'll have to do some calibrations of the DTS's by yourself. What you described by using DVFS and/or power-saving vs. active modes for validation/calibration makes sense to me -- at different utilization levels of the cores, the relative DTS errors might not be the same. I would assume the actual temperature difference among cores would be small (<5K) when all cores running same workload, since it is mainly due to core-to-core variations. After the calibration, you'll have a plot, x being the measured temperature, y being the error for that temperature at each DTS. Then you can do corrections based on the errors for each sensor. This is what I can think of right now, there should be more complicated ways for better accuracy.... Hope this is helpful. -Wei On Fri, Jun 15, 2012 at 7:34 AM, Bartosz Wojciechowski < bartosz.wojciechowski at pwr.wroc.pl> wrote: > Dear All, > > I'm currently trying to model the thermal behavior of my i7 processor with > HotSpot. However I have some troubles with thermal sensor readings. During > idle > periods there are differences of more than 10K between cores. During high > load, > with constant and equal load on all cores, the differences are smaller but > still above 5K. I know at least one of the sensors is wrong, because the > idle > temps shown are smaller than ambient temperature (25 deg C vs 29 in > computer > case). But I'm not convinced that the sole reason of differences between > reported core temperatures are the sensors. > > I know Digital Thermal Sensors have limited accuracy (intel whitepaper on > DTS/PECI), but over 10 Kelvins seems a little too much of an error. The > problem > is that there are 3 possible sources of differences in reported temperature > that I can think off: sensor error, differences in power consumption > between > cores and position of cores relative to other heat sources (floorplan). > > Does anyone know of any other possible reasons, and a way to mittigate this > problem? If you could point me to paper on the subject or have any > knowledge about > possible sources of such temperature readings, please do not hesitate to > write > me :-) > > I've been trying to devise a sure way to > validate the relative results from all the cores by using DVFS (and the > fact > that heat flows from hotter to cooler core, regardless of what the sensor > shows), but my results are still inconclusive. > > This is not strictly HotSpot-related, but I hope it will be interesting to > all > those in temperature modelling. > > Regards, > Bartosz > > -- > grad student > Wroclaw, Poland > http://zak.iiar.pwr.wroc.pl > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20120615/d6d9d6bc/attachment.html