From kbarr216 at yahoo.com Mon Jul 12 17:56:36 2004 From: kbarr216 at yahoo.com (Ken Barr) Date: Wed Mar 22 16:04:11 2006 Subject: [Hotspot] hotspot/wattch integration Message-ID: <20040712215636.4956.qmail@web51706.mail.yahoo.com> Just checking: out-of-the box Wattch does not appear to be set up to produce a trace of the form required by Hotspot, right? I need to take care of the running average and periodic output myself? (if I've overlooked some obvious dump_stats(int period) function, let me know!) I'm not sure if you use wattch, but here's a procedural question: do you tend to call Hotspot on the fly or use it (as in sim-template) to post-process a trace? Ken __________________________________ Do you Yahoo!? New and Improved Yahoo! Mail - 100MB free storage! http://promotions.yahoo.com/new_mail From skadron at cs.virginia.edu Wed Jul 14 17:01:50 2004 From: skadron at cs.virginia.edu (Kevin Skadron) Date: Wed Mar 22 16:04:11 2006 Subject: [Hotspot] Re: HotSpot post from kbarr216@yahoo.com requires approval In-Reply-To: References: Message-ID: <40F59F3E.1000805@cs.virginia.edu> > Subject: hotspot/wattch integration > From: Ken Barr > Date: Mon, 12 Jul 2004 14:56:36 -0700 (PDT) > Just checking: out-of-the box Wattch does not appear > to be set up to produce a trace of the form required > by Hotspot, right? I need to take care of the running > average and periodic output myself? As far as I know, yes. > I'm not sure if you use wattch, but here's a > procedural question: do you tend to call Hotspot on > the fly or use it (as in sim-template) to post-process > a trace? We actually do it both ways, depending on the expt. When you do something that changes the instruction flow, you need to either generate a new trace or run HotSpot on the fly. But for many experiments, post-processing a trace is enough. Please let us know if you have further questions! /K From skadron at cs.virginia.edu Wed Jul 14 17:06:47 2004 From: skadron at cs.virginia.edu (Kevin Skadron) Date: Wed Mar 22 16:04:11 2006 Subject: [Hotspot] We need info from HotSpot users Message-ID: <40F5A067.5080607@cs.virginia.edu> Now that HotSpot has been available for over a year, we'd like to collect info about how people are using it. We'd be interested to hear: - What kind of topics you are studying - What kind of simulator infrastructure you are using it with (SimpleScalar, Wattch, Liberty, Rsim, Turandot, etc.) - Any publications you have using HotSpot -- not just refereed pubs, but also TRs, class reports, etc. - Any other info you think might be relevant Also, we are working on v3.0, and would like to hear about any problems you have encountered or other suggestions you have for improving HotSpot. And if you have made major modifications that you think might be useful to the HotSpot user community, we'd be interested to hear about those too. Thanks, --Kevin Skadron, Mircea Stan, Wei Huang, Karthik Sankaranarayanan, Shougata Ghosh, Yingmin Li, and Siva Velusamy -- skadron@cs.virginia.edu | Assistant Professor http://www.cs.virginia.edu/~skadron | Dept. of Computer Science voice: (434) 982-2042 | University of Virginia fax: (434) 982-2214 | Charlottesville, VA 22904 From yongkui at acad.umass.edu Wed Jul 14 21:07:29 2004 From: yongkui at acad.umass.edu (yongkui@acad.umass.edu) Date: Wed Mar 22 16:04:11 2006 Subject: [Hotspot] cannot reach the steady state temperature? In-Reply-To: <40F59F3E.1000805@cs.virginia.edu> References: <40F59F3E.1000805@cs.virginia.edu> Message-ID: <1089853649.40f5d8d1956cd@mail-www4.oit.umass.edu> hi, I integrated hotspot into wattch, and use it to simultate the cache temperature. but I found from the simulation results that the steady state temperature of units (which is calculated with the function steady_state_temp().) is always 10 degrees higher than the actual temperature that unit will reach finally (calling compute_temp() periodically). I think that theoretically the final temperature should reach this steady state temperature finally, but it seems not. Did you find this problem before? Can you explain this for me? Unit Steady Transient Icache 85.08 75.74 Dcache 84.49 75.15 I checked my calling way of hotspot interface, I think I am using the recommended way of calling hotspot. yongkui From yl3e at cs.virginia.edu Mon Jul 26 15:39:00 2004 From: yl3e at cs.virginia.edu (Yingmin Li) Date: Wed Mar 22 16:04:11 2006 Subject: [Hotspot] cannot reach the steady state temperature? In-Reply-To: <1089853649.40f5d8d1956cd@mail-www4.oit.umass.edu> Message-ID: Hi, What's the initial temperature of your transient simulation? and how long did you simulate? I think you should set the steady state temperature as the initial temperature for transient simulation , otherwise, in some cases, it takes quite long simulation to reach steady state temperature, for example, several billions instructions.... /Yingmin On Wed, 14 Jul 2004 yongkui@acad.umass.edu wrote: > hi, > > I integrated hotspot into wattch, and use it to simultate the cache temperature. > > but I found from the simulation results that the steady state temperature of > units (which is calculated with the function steady_state_temp().) is always 10 > degrees higher than the actual temperature that unit will reach finally > (calling compute_temp() periodically). > I think that theoretically the final temperature should reach this steady state > temperature finally, but it seems not. > > Did you find this problem before? Can you explain this for me? > > Unit Steady Transient > Icache 85.08 75.74 > Dcache 84.49 75.15 > > I checked my calling way of hotspot interface, I think I am using the > recommended way of calling hotspot. > > yongkui > > > _______________________________________________ > HotSpot mailing list > HotSpot@cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > From skadron at cs.virginia.edu Mon Jul 26 15:39:31 2004 From: skadron at cs.virginia.edu (Kevin Skadron) Date: Wed Mar 22 16:04:11 2006 Subject: [Hotspot] Re: cannot reach the steady state temperature? Message-ID: <41055DF3.6090105@cs.virginia.edu> Just wanted to update everyone on the answer to Yongkui's question. --Kevin ---------------------- Hello Yongkui, The simulation time is the issue here. Longer simulation time (for like 100 seconds) should result in matching transient and steady state numbers. The time constant for the heat sink is very large and hence is this slow behaviour. As a convincing exercise, I tried the following: 1. Changed the power file to one with much fewer number of samples. To obtain this, I just averaged the power numbers you had sent. 2. Increased the sampling interval to a large amount so that the total simulation time was close to like 100 seconds. The steady state temperature then matched the transient. Hope this answers your question. Thanks -karthik > -------- Original Message -------- > Subject: Re: [Fwd: HotSpot post from yongkui@acad.umass.edu requires > approval] > Date: Thu, 22 Jul 2004 05:09:26 -0400 > From: yongkui@acad.umass.edu > To: skadron@cs.virginia.edu > References: <40F6B37A.1050307@cs.virginia.edu> > <1090393248.40fe14a037960@mail-www4.oit.umass.edu> > <40FE77C0.2090704@cs.virginia.edu> > > hi, > > I am very sorry, I gave you a wrong gcc.steady file last time. > > I include the new correct file in the atatchment gcc2.tar.gz file, and hope > this can make the problem clear. > > I set the initial temperature to 80C for all blocks. and the sampling > interval > is 3.33 microseconds, the simulation time is 6465*3.33=21528us=21ms, > maybe it > is because the simulation time is too short? > > but I think the transient temperature should be decreasing from 80C to > 70C, but > actually the temperature of all blocks are increasing from 80C. So i > think it > is strange. > > thank you very much. > > yongkui > > yongkui@acad.umass.edu wrote: > > > > > hi, > > > > > > I found a problem with HotSpot 2.0: > > > I found from the simulation results that the steady state temperature of > > > units (which is calculated with the function steady_state_temp().) is not > > same > > > as the actual temperature that unit will reach finally > > > (calling compute_temp() periodically). > > > I think that theoretically the final temperature should reach this steady > > state > > > temperature finally, but it seems not. > > > > > > I included the wattch_base.flp, gcc.p, gcc.t, sim-template.out file in the > > > > > attachments, I use the same floorplan as you gave to me. > > > > > > the steady temperature is in sixties degree C, and the initial temperature > > is > > > 80 C, but in gcc.t output file, the temperature is keep increasing from 80 > > C. > > > > > > What is wrong with my data and simulation? > > > > > > thanks. > > > > > > yongkui From kalpana at email.arizona.edu Fri Jul 30 03:25:47 2004 From: kalpana at email.arizona.edu (kalpana@email.arizona.edu) Date: Wed Mar 22 16:04:11 2006 Subject: [Hotspot] question reg. version 2.0 Message-ID: <1091172347.7a97b8db35df1@www.email.arizona.edu> Hi, I have a question reg the DAC paper "Compact thermal modeling for temperature-aware design". The paper says - The thermal model for the primary heat flow path in [the previous work "temperature aware microarchitecture"] is extended by making the model grid like. the 3-block die is divided into 3 by 3 grid cells. what I dont understand is: so each grid is of uniform size. As I understand the hotspot model on the website, a fully specified floorplan of the chip is needed as input. Obviously each module in the floorplan has different dimensions. I dont understand the concept of dividing the chip into unfiorm size grid cells. and moreover on the website information about the changes added to version 2.0 and in the documentation there is no this grid thing, so I could not correlate. Can you please explain me what this grid-like model means. I very much appreciate your help. Thanks kalpana vakati From skadron at cs.virginia.edu Fri Jul 30 09:17:52 2004 From: skadron at cs.virginia.edu (Kevin Skadron) Date: Wed Mar 22 16:04:11 2006 Subject: [Hotspot] question reg. version 2.0 In-Reply-To: <1091172347.7a97b8db35df1@www.email.arizona.edu> References: <1091172347.7a97b8db35df1@www.email.arizona.edu> Message-ID: <410A4A80.1070609@cs.virginia.edu> Short answer: the grid model isn't in the public-release version of HotSpot yet! Long answer: The grid model is not yet fully incorporated into HotSpot. It was developed in a code branch that we're still integrating. That will be v3, coming out this fall. We are also looking into a hybrid grid-lumpy model, so part of the floorplan can be modeled the "old" way (by arch. blocks) and other parts can be modeled with a grid (e.g. a cache). We're still working on the math for this hybrid and can't promise that for the fall though. Basically, the grid model is intended for fine-grained modeling, and really grew out of our desire to use HotSpot in a CAD setting. But it's easy to map a lumpy floorplan onto a grid. When setting up a simulation, each cell simply looks up which architectural blocks cover that cell. In most cases there's only one block, but when there are multiple blocks, an area-weighted average is taken. This will be part of v3. Hope this answers your questions. Please holler if you have more questions; we're happy to help. --Kevin kalpana@email.arizona.edu wrote: > Hi, > > I have a question reg the DAC paper "Compact thermal modeling for > temperature-aware design". > The paper says - The thermal model for the primary heat flow path in [the > previous work "temperature aware microarchitecture"] is extended by making the > model grid like. the 3-block die is divided into 3 by 3 grid cells. > > what I dont understand is: > so each grid is of uniform size. As I understand the hotspot model on the > website, a fully specified floorplan of the chip is needed as input. Obviously > each module in the floorplan has different dimensions. I dont understand the > concept of dividing the chip into unfiorm size grid cells. and moreover on the > website information about the changes added to version 2.0 and in the > documentation there is no this grid thing, so I could not correlate. > > Can you please explain me what this grid-like model means. > I very much appreciate your help. > > Thanks > kalpana vakati > > > _______________________________________________ > HotSpot mailing list > HotSpot@cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot -- skadron@cs.virginia.edu | Assistant Professor http://www.cs.virginia.edu/~skadron | Dept. of Computer Science voice: (434) 982-2042 | University of Virginia fax: (434) 982-2214 | Charlottesville, VA 22904