From chzhang at cs.ucr.edu Thu Aug 7 23:43:13 2003 From: chzhang at cs.ucr.edu (Chuanjun Zhang) Date: Wed Mar 22 16:04:08 2006 Subject: [Hotspot] model for a function unit Message-ID: Hi, It seems that the HotSpot only models the function unit level temperature. Does it model the temperature in a particular function unit ? Such as if the ALU is modified, can I evaluate the temperature changes of the new ALU by using the HotSpot model ? Thanks. -- Chuanjun Zhang 909-787-2373(o) From ks4kk at cs.virginia.edu Fri Aug 8 02:51:39 2003 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Wed Mar 22 16:04:08 2006 Subject: [Hotspot] model for a function unit In-Reply-To: References: Message-ID: Hi Chuanjun Zhang, HotSpot currently models temperature at the functional block granularity. The way to work around this would be to split the functional block of interest (ALU in this case) into multiple sub-blocks and specify each as a separate functional block in the floorplan file. Note that the power simulator should then provide power numbers for these individual sub-blocks. Thanks, -karthik On Thu, 7 Aug 2003, Chuanjun Zhang wrote: > Hi, > It seems that the HotSpot only models the function unit level temperature. > Does it model the temperature in a particular function unit ? > Such as if the ALU is modified, can I evaluate the temperature changes of > the new ALU by using the HotSpot model ? > Thanks. > -- > Chuanjun Zhang > 909-787-2373(o) > > > > _______________________________________________ > HotSpot mailing list > HotSpot@cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > From gomezjo at imec.be Wed Aug 13 13:11:13 2003 From: gomezjo at imec.be (Jose Ignacio Gomez) Date: Wed Mar 22 16:04:08 2006 Subject: [Hotspot] HotSpot usage Message-ID: <3F3A7131.10601@imec.be> Hi, I've recently installed HotSpot and I'm now trying to make it run, but I have some problems. If I understood the steps well, it's first required to run the application just to compute the steady temperatures. Then, I set the initial temperatures according to the previous results and do a new simulation, this time calling regularly to compute_temp (I just call once to steady_state at the end). That way, I've tried to get similar results to those you present in the ISCA paper (fig 3b), but I never get a range of 40 degrees. I'm trying it with the gcc power values provided in sim-template, iterating over compute_temp, but the temperature range I get is much smaller (around 10 degrees) Could you please inidicate me what am I doing wrong? (and, if it's possible, send me a complete example of the right usage?) Thank you very much, Nacho From ks4kk at cs.virginia.edu Wed Aug 13 17:15:58 2003 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Wed Mar 22 16:04:08 2006 Subject: [Hotspot] HotSpot usage In-Reply-To: <3F3A7131.10601@imec.be> References: <3F3A7131.10601@imec.be> Message-ID: Hello Nacho, The figure you are referring to (figure 3b) starts from a cool chip (300K) while the HotSpot code starts from the steady state. That is one reason for the 10 degree range as opposed to 40 degrees in the figure. So, in HotSpot code, if you start from initial temperatures equal to the ambient, you should observe a much larger range. Secondly, the power numbers used in figure 3b are not from Wattch - they are just trial power numbers we used. Since the experiment is a validation effort, it just replicates the floorplan and the thermal R and C values. The power density is just a trial value. Hence, the steady state value of the temperature, which is determined by the power density, can differ for other power numbers. If this explanation doesn't answer your question fully, e-mail me and we can talk more. Thanks -karthik On Wed, 13 Aug 2003, Jose Ignacio Gomez wrote: > > Hi, > I've recently installed HotSpot and I'm now trying to make > it run, but I have some problems. > If I understood the steps well, it's first required to run > the application just to compute the steady temperatures. Then, I set the > initial temperatures according to the previous results and do a new > simulation, this time calling regularly to compute_temp (I just call > once to steady_state at the end). > > That way, I've tried to get similar results to those you present > in the ISCA paper (fig 3b), but I never get a range of 40 degrees. I'm > trying it with the gcc power values provided in sim-template, iterating > over compute_temp, but the temperature range I get is much smaller > (around 10 degrees) > > Could you please inidicate me what am I doing wrong? (and, if > it's possible, send me a complete example of the right usage?) > > Thank you very much, > Nacho > > > > > > > _______________________________________________ > HotSpot mailing list > HotSpot@cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot >