From ks4kk at cs.virginia.edu Mon Jun 2 09:01:06 2008 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Mon, 2 Jun 2008 12:01:06 -0400 (EDT) Subject: [Hotspot] The power trace input file for HotSpot In-Reply-To: <2b8513b80805291042h57e125b1i9be60e7acb69d043@mail.gmail.com> References: <2b8513b80805291042h57e125b1i9be60e7acb69d043@mail.gmail.com> Message-ID: Hello Yu-Hsun Lin, The -ptrace option in Simplescalar/Wattch is for pipeline tracing and not power tracing. In order to generate a power trace file for HotSpot from Wattch, you would have to add printfs in Wattch source code at the appropriate places. Please see the sim-template.c and hotspot.c (main function) files that come with the HotSpot distribution for illustrations. HotSpot's power trace file format is a straightforward one - the first line gives the names of the functional blocks (tab-separated) and subsequent lines give the corresponding power values (also tab-separated) per calling interval. See the example.ptrace and gcc.ptrace files for samples. Hope this helps, -karthik On Fri, 30 May 2008, ?L???V wrote: > Hi, > > I am new to use HotSpot, and I have a basic question. > I try to generate the power trace input file for HotSpot. > But the -ptrace option in Wattch doesn't generate the power trace for HotSpot. > > Could someone give me a full example to generate the power trace file > for HotSpot? > Thank you very much. > > best regards. > > -- > Yu-Hsun Lin ?L???V > System Software Lab, CS, NTHU, Taiwan > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > > From augusto.vega at bsc.es Mon Jun 16 02:37:06 2008 From: augusto.vega at bsc.es (Augusto Vega) Date: Mon, 16 Jun 2008 11:37:06 +0200 Subject: [Hotspot] 3D-stacking in HotSpot Message-ID: <48563442.1020604@bsc.es> Hi, I'm trying to do thermal modeling of 3D-stacked architectures. As I can see in the documentation, this feature is supported by HotSpot, and it's pretty clear how this works. However, I've some doubts: 1) I've written my own layer configuration file with DIFFERENT floorplans in each layer, and I run HotSpot with the flag: "-f ev6.flp". What does this flag mean? If I have different floorplans in my 3D-stacked chip, why should I specify only one? 2) I also use the "-grid_steady_file" flag in order to obtain a ".t" trace that I will later transform in a SVG image. If I have several layers, which does the ".t" file correspond to? Any help will be appreciated. -- Augusto Vega BSC-CNS Barcelona Supercomputing Center Centro Nacional de Supercomputaci?n Telf.: +34 93 413 79 46 e-mail: augusto.vega at bsc.es From ks4kk at cs.virginia.edu Mon Jun 16 06:21:10 2008 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Mon, 16 Jun 2008 09:21:10 -0400 (EDT) Subject: [Hotspot] 3D-stacking in HotSpot In-Reply-To: <48563442.1020604@bsc.es> References: <48563442.1020604@bsc.es> Message-ID: > 1) I've written my own layer configuration file with DIFFERENT > floorplans in each layer, and I run HotSpot with the flag: "-f ev6.flp". > What does this flag mean? If I have different floorplans in my > 3D-stacked chip, why should I specify only one? The -f flag for 3D chips is just a redundant "backward compatibility" option. It is overridden by the options in the layer configuration file - as indicated by the warning message. > 2) I also use the "-grid_steady_file" flag in order to obtain a ".t" > trace that I will later transform in a SVG image. If I have several > layers, which does the ".t" file correspond to? It corresponds to just the top layer. If you need the entire grid of temperatures, it is available in model->grid->last_steady->cuboid[][][] and model->grid->last_steady->extra. Also please take a look at the dump_steady_temp_grid function in temperature_grid.c Hope this helps. -karthik From augusto.vega at bsc.es Wed Jun 18 10:55:03 2008 From: augusto.vega at bsc.es (Augusto Vega) Date: Wed, 18 Jun 2008 19:55:03 +0200 Subject: [Hotspot] 3D-stacking in HotSpot Message-ID: <48594BF7.5050906@bsc.es> Hi, With the help of Karthik I could obtain the temperature profiles for different layers in my 3D-stacked chip. However, I have found something that seems strange. Let's take a look to the following code (in the "big while" in the function main() of hotspot.c): /* for computing average */ for(i=0; i < n; i++) overall_power[i] += power[i]; lines++; Iteration by iteration, instantaneous power values are accumulated in 'overall_power'. The value 'n' is the total number of blocks that dissipate power. Now, I'm modeling a chip with 4 layers: Layer 0: silicon (dissipates power) Layer 1: thermal interface material (no dissipation) Layer 2: silicon (dissipates power) Layer 3: thermal interface material (no dissipation) Layer 0 has 30 blocks and layer 2 has also 30 blocks (so, n is 60). In the 'for' loop I wrote at the beginning we are only taking into account the first 60 elements of 'power' that in my case means I'm only considering power values for layer 0 and missing power values for layer 2 (that reside in the second half of the 'power' array). The results I've obtained were as having only one layer (layer 0) dissipating power. I've modified the code in order to consider the whole 'power' array when computing the overall power values and now it seems to work fine. But I wonder if this is really a problem or I'm doing something wrong. So, I'll be very grateful if somebody can clarify me this point. Thank you in advance. -- Augusto Vega BSC-CNS Barcelona Supercomputing Center Centro Nacional de Supercomputaci?n Telf.: +34 93 413 79 46 e-mail: augusto.vega at bsc.es From ks4kk at cs.virginia.edu Wed Jun 18 11:37:35 2008 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Wed, 18 Jun 2008 14:37:35 -0400 (EDT) Subject: [Hotspot] 3D-stacking in HotSpot In-Reply-To: <48594BF7.5050906@bsc.es> References: <48594BF7.5050906@bsc.es> Message-ID: Hello Augusto, This was a problem in previous versions of HotSpot and was fixed in the most recent release (4.1). It appears that you are using an older version of HotSpot. We encourage you to use Ver 4.1 - apart from bugfixes, it also has a much faster steady state solver for the grid model. Hope this helps. -karthik On Wed, 18 Jun 2008, Augusto Vega wrote: > Hi, > > With the help of Karthik I could obtain the temperature profiles for > different layers in my 3D-stacked chip. However, I have found something > that seems strange. > > Let's take a look to the following code (in the "big while" in the > function main() of hotspot.c): > > /* for computing average */ > for(i=0; i < n; i++) > overall_power[i] += power[i]; > lines++; > > Iteration by iteration, instantaneous power values are accumulated in > 'overall_power'. The value 'n' is the total number of blocks that > dissipate power. > > Now, I'm modeling a chip with 4 layers: > > Layer 0: silicon (dissipates power) > Layer 1: thermal interface material (no dissipation) > Layer 2: silicon (dissipates power) > Layer 3: thermal interface material (no dissipation) > > Layer 0 has 30 blocks and layer 2 has also 30 blocks (so, n is 60). > > In the 'for' loop I wrote at the beginning we are only taking into > account the first 60 elements of 'power' that in my case means I'm only > considering power values for layer 0 and missing power values for layer > 2 (that reside in the second half of the 'power' array). The results > I've obtained were as having only one layer (layer 0) dissipating power. > > I've modified the code in order to consider the whole 'power' array when > computing the overall power values and now it seems to work fine. But I > wonder if this is really a problem or I'm doing something wrong. > > So, I'll be very grateful if somebody can clarify me this point. > > Thank you in advance. > > -- > Augusto Vega > > BSC-CNS > Barcelona Supercomputing Center > Centro Nacional de Supercomputaci?n > Telf.: +34 93 413 79 46 > e-mail: augusto.vega at bsc.es > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > > From ks4kk at cs.virginia.edu Thu Jun 26 19:44:49 2008 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Thu, 26 Jun 2008 22:44:49 -0400 (EDT) Subject: [Hotspot] Question about hotspot In-Reply-To: <1F03DBDB34089C6CE2D6FDEB@mstu2> References: <002701c8d7a0$317e2930$91103d80@Mk> <4863C5A4.6090107@cs.virginia.edu> <1F03DBDB34089C6CE2D6FDEB@mstu2> Message-ID: Hello Minki, The following command sequence illustrates what you would like to accomplish: hotspot -c hotspot.config -f ev6.flp -p gcc.ptrace \ -steady_file gcc.steady -model_type grid \ -grid_steady_file gcc.grid.steady grid_thermal_map.pl ev6.flp gcc.grid.steady > gcc.svg convert -font Helvetica svg:gcc.svg gcc.pdf Please see the HotSpot HOWTO (http://lava.cs.virginia.edu/HotSpot/HotSpot-HOWTO.htm) for more detailed step-by-step guidelines. Hope this helps, -karthik >>> --On Thursday, June 26, 2008 11:20 AM -0400 Minki Cho >>> wrote: >>> >>>> >>>> >>>> >>>> Hi! >>>> >>>> I am Minki Cho and studying at GeorgiaTech now. >>>> >>>> I am trying to figure out how to use hotspot. >>>> Unfortunately, I could not understand totally how to >>>> deal with. >>>> >>>> I want to test my arbitrary floorplan with hotspot >>>> simulator. How do I make that picture such as >>>> example.pdf or example.svg? >>>> >>>> When I run a hotspot executable file, that gave me a set >>>> of numbers. It looks like average temperature. >>>> Additionally, I have created floorplan and random power >>>> number for hotspot simulation. What else do I need to >>>> run simulator properly to make the grid picture of >>>> temperature. Thank you. >>>> >>>> >>>> >>>> Your sincerely, >>>> >>>> Minki Cho From ks4kk at cs.virginia.edu Thu Jun 26 20:10:51 2008 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Thu, 26 Jun 2008 23:10:51 -0400 (EDT) Subject: [Hotspot] HotSpot HOWTO and Version 4.1 Message-ID: Hello Everyone, As the news of the release of HotSpot version 4.1 has not been shared in this forum before, we would like to make that announcement now. Version 4.1 incorporates a significantly faster steady-state solver for the grid model. The previous Gauss-Seidel solver has been upgraded to a multigrid solver. For typical cases, resulting speedup per steady-state call is greater than 20x! Apart from regular enhancements like minor bugfixes, this version also includes better support for integration with existing power simulators. You can download version 4.1 here: http://lava.cs.virginia.edu/HotSpot/download_form2.html Furthermore, recent questions posted to this list indicate the lack of a coherent HOWTO document for HotSpot. Hence, we would also like to announce the upgraded HOWTO with step-by-step guidelines on the usage of HotSpot. The HOWTO can be found here: http://lava.cs.virginia.edu/HotSpot/HotSpot-HOWTO.htm Please feel free let us know your thoughts and any questions you might have. Thank you. -karthik ---------------------------------- Karthik Sankaranarayanan, PhD Student, Department of Computer Science, University of Virginia, 151 Engineer's Way, PO Box 400740 Charlottesville, VA 22904-4740 Office: Olsson 227-12 Email: karthick at cs.virginia.edu ---------------------------------- From mohamed_elsawaf at yahoo.com Fri Jun 27 04:15:19 2008 From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf) Date: Fri, 27 Jun 2008 04:15:19 -0700 (PDT) Subject: [Hotspot] unexpected blocks temperatures In-Reply-To: Message-ID: <499863.65616.qm@web53706.mail.re2.yahoo.com> Hi can you please help me? how can i check that the center of my floor-plan is on the same access as heat-sink center? I?am getting unexpected blocks temperatures.Iam suspecting that chip floor-plan is shifted to the left of from heat-sink center. Thank you Mohamed Elsawaf -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20080627/86573810/attachment.html From wh6p at cms.mail.virginia.edu Fri Jun 27 07:32:22 2008 From: wh6p at cms.mail.virginia.edu (Wei Huang) Date: Fri, 27 Jun 2008 10:32:22 -0400 Subject: [Hotspot] unexpected blocks temperatures In-Reply-To: <499863.65616.qm@web53706.mail.re2.yahoo.com> References: <499863.65616.qm@web53706.mail.re2.yahoo.com> Message-ID: Hi Mohamed, HotSpot automatically aligns chip center with heat sink/spreader centers. What is the size of your chip? You may want to check if it is larger than the default sizes of heat spreader and heat sink in HotSpot. If that is the case, increasing the default values to proper ones should solve your problem. -Wei --On Friday, June 27, 2008 4:15 AM -0700 Mohamed Elsawaf wrote: > Hi > > can you please help me? > how can i check that the center of my floor-plan is on > the same access as heat-sink center? > I?am getting unexpected blocks temperatures.Iam > suspecting that chip floor-plan is shifted to the left of > from heat-sink center. > > Thank you > Mohamed Elsawaf > > > From mohamed_elsawaf at yahoo.com Fri Jun 27 11:07:17 2008 From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf) Date: Fri, 27 Jun 2008 11:07:17 -0700 (PDT) Subject: [Hotspot] unexpected blocks temperatures In-Reply-To: Message-ID: <643909.10254.qm@web53704.mail.re2.yahoo.com> Hi Wei Thank you for your fast response My chip area = 0.000514234 m^2 # chip specs # chip thickness in meters -t_chip 0.0005 # heat sink specs # convection capacitance in J/K -c_convec 140.4 # convection resistance in K/W -r_convec 0.1 # heatsink side in meters -s_sink 0.12 # heatsink thickness in meters -t_sink 0.0069 # heat spreader specs # spreader side in meters -s_spreader 0.03 # spreader thickness in meters -t_spreader 0.001 # interface material specs # interface material thickness in meters -t_interface 0.000075 waiting for your feedback Thank you Mohamed Elsawaf --- On Fri, 6/27/08, Wei Huang wrote: From: Wei Huang Subject: Re: [Hotspot] unexpected blocks temperatures To: mohamed_elsawaf at yahoo.com Cc: "Karthik Sankaranarayanan" , hotspot at mail.cs.virginia.edu Date: Friday, June 27, 2008, 8:32 AM Hi Mohamed, HotSpot automatically aligns chip center with heat sink/spreader centers. What is the size of your chip? You may want to check if it is larger than the default sizes of heat spreader and heat sink in HotSpot. If that is the case, increasing the default values to proper ones should solve your problem. -Wei --On Friday, June 27, 2008 4:15 AM -0700 Mohamed Elsawaf wrote: > Hi > > can you please help me? > how can i check that the center of my floor-plan is on > the same access as heat-sink center? > I?am getting unexpected blocks temperatures.Iam > suspecting that chip floor-plan is shifted to the left of > from heat-sink center. > > Thank you > Mohamed Elsawaf > > > -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20080627/01d320cd/attachment-0001.html From sudarshanml at gmail.com Fri Jun 27 11:38:18 2008 From: sudarshanml at gmail.com (Sudarshan Reddy M) Date: Sat, 28 Jun 2008 00:08:18 +0530 Subject: [Hotspot] Conjugate Gradient Method Message-ID: Hi, Why is Conjugate gradient method based solvers are not used in steady state analysis. It really speed up the performance compared to the existing multigrid solvers. If required I can help you out in this regard. Let me know if there is any specific reason for not having CG. It is really good and easily implementable algorithm for Ax=b when A is symmetric and positive definite(which is the case in steady state analysis). Regards -- Sudarshan Reddy M M. Tech Control and Computing IIT Bombay From mohamed_elsawaf at yahoo.com Mon Jun 30 23:31:57 2008 From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf) Date: Mon, 30 Jun 2008 23:31:57 -0700 (PDT) Subject: [Hotspot] unexpected blocks temperatures In-Reply-To: <35BBFB82E5C70D38A63CCA01@mstu2> Message-ID: <554934.53575.qm@web53705.mail.re2.yahoo.com> Hi Wei Thank you for your help, i will try Hostspot 4.1 meanwhile please find the simulation temperatures distribution output [per block per core] and my floorplan reassembling to IBM power4 MCM [ 8 cores] i scale it to 45 nm technology i can see that cores C0 and C6 having higher temperatures at steady state than the other cores while all cores having the same power consumed in gcc.ptrace file also i don't have any explanation regarding this unexpected higher temperature thank you Mohamed Elsawaf --- On Fri, 6/27/08, Wei Huang wrote: From: Wei Huang Subject: Re: [Hotspot] unexpected blocks temperatures To: mohamed_elsawaf at yahoo.com Cc: "Karthik Sankaranarayanan" Date: Friday, June 27, 2008, 12:49 PM Hi Mohamed, As you can see, your chip size is 2.26cm on each side, which is smaller than the sizes of spreader (3cm) and heat sink (12cm). So I don't think your problem is caused by the size. How much higher the temperatures are than what you expected? From your configuration file, it seems either you have changed it (which is OK) or you are using an old version of HotSpot. The newest version is 4.1, which fixes some bugs in previous versions and includes a faster grid model solver. So I would recommend you to download v4.1 and try it again. -Wei --On Friday, June 27, 2008 11:07 AM -0700 Mohamed Elsawaf wrote: > Hi Wei > > Thank you for your fast response > > My chip area = 0.000514234 m^2 > ># chip specs ># chip thickness in meters > -t_chip 0.0005 > ># heat sink specs ># convection capacitance in J/K > -c_convec 140.4 ># convection resistance in K/W > -r_convec 0.1 ># heatsink side in meters > -s_sink 0.12 ># heatsink thickness in meters > -t_sink 0.0069 > ># heat spreader specs ># spreader side in meters > -s_spreader 0.03 ># spreader thickness in meters > -t_spreader 0.001 > ># interface material specs ># interface material thickness in meters > -t_interface 0.000075 > > waiting for your feedback > > Thank you > > Mohamed Elsawaf > > > --- On Fri, 6/27/08, Wei Huang > wrote: > > From: Wei Huang > Subject: Re: [Hotspot] unexpected blocks temperatures > To: mohamed_elsawaf at yahoo.com > Cc: "Karthik Sankaranarayanan" , > hotspot at mail.cs.virginia.edu > Date: Friday, June 27, 2008, 8:32 AM > > > Hi Mohamed, HotSpot automatically aligns chip center > with heat sink/spreader centers. What is the size of > your chip? You may want to check if it is larger than > the default sizes of heat spreader and heat sink in > HotSpot. If that is the case, increasing the default > values to proper ones should solve your problem. -Wei > --On Friday, June 27, 2008 4:15 AM -0700 Mohamed Elsawaf > wrote: > Hi > > can you > please help me? > how can i check that the center of my > floor-plan is on > the same access as > heat-sink center? > I?am getting unexpected blocks > temperatures.Iam > suspecting that chip floor-plan is > shifted to the left of > from heat-sink center. > > Thank > you > Mohamed Elsawaf > > > > > > -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20080630/8f802e81/attachment-0001.html -------------- next part -------------- A non-text attachment was scrubbed... Name: temp.jpg Type: image/jpeg Size: 36391 bytes Desc: not available Url : http://www.cs.virginia.edu/pipermail/hotspot/attachments/20080630/8f802e81/attachment-0002.jpg -------------- next part -------------- A non-text attachment was scrubbed... Name: flp.jpg Type: image/jpeg Size: 46771 bytes Desc: not available Url : http://www.cs.virginia.edu/pipermail/hotspot/attachments/20080630/8f802e81/attachment-0003.jpg