From michaelkadin at gmail.com Tue Oct 16 14:57:55 2007 From: michaelkadin at gmail.com (Mike Kadin) Date: Tue, 16 Oct 2007 17:57:55 -0400 Subject: [Hotspot] Question About HotSpot's Included Floorplan and Power Traces In-Reply-To: <34954c9d0710161422rc40cd74w4da136feaa4b3700@mail.gmail.com> References: <34954c9d0710161422rc40cd74w4da136feaa4b3700@mail.gmail.com> Message-ID: <34954c9d0710161457y1fee75b5nf1d3340bedb580fd@mail.gmail.com> I just had a quick and easy question. In HotSpot 3.0 there are some given files that I have been using, and my research directs me towards knowing where they came from. I understand the floorplan is modeled after the EV7, is this the 130nm version? And also, I assume you got the power numbers in the gcc.ptrace from running the SPEC benchmark gcc through Wattch. What was the frequency that you ran the processor at to give these numbers? Thanks, Mike Kadin Undergraduate Researcher Brown University - Division of Engineering -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20071016/c80dbc56/attachment.html From ks4kk at cs.virginia.edu Tue Oct 16 16:47:13 2007 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Tue, 16 Oct 2007 19:47:13 -0400 (EDT) Subject: [Hotspot] Question About HotSpot's Included Floorplan and Power Traces In-Reply-To: <34954c9d0710161457y1fee75b5nf1d3340bedb580fd@mail.gmail.com> References: <34954c9d0710161422rc40cd74w4da136feaa4b3700@mail.gmail.com> <34954c9d0710161457y1fee75b5nf1d3340bedb580fd@mail.gmail.com> Message-ID: Hello Mike, Please note that the sample power numbers that come with HotSpot release are for illustrative purposes only. In fact gcc.ptrace records the execution of only a couple of million instructions starting at a random simulation point. Those numbers were obtained while modeling an ev6-like microarchitecture running at 3 GHz at 130 nm. Thanks, -karthik On Tue, 16 Oct 2007, Mike Kadin wrote: > I just had a quick and easy question. In HotSpot 3.0 there are some given > files that I have been using, and my research directs me towards knowing > where they came from. I understand the floorplan is modeled after the EV7, > is this the 130nm version? And also, I assume you got the power numbers in > the gcc.ptrace from running the SPEC benchmark gcc through Wattch. What was > the frequency that you ran the processor at to give these numbers? > > Thanks, > > Mike Kadin > Undergraduate Researcher > Brown University - Division of Engineering > From shangupt at umich.edu Thu Oct 18 08:30:11 2007 From: shangupt at umich.edu (Shantanu Gupta) Date: Thu, 18 Oct 2007 11:30:11 -0400 Subject: [Hotspot] Fixing a constant temperature for an architectural block Message-ID: <1192721411.25071.141.camel@aslan.eecs.umich.edu> Hi, I have a question regarding the working of the HotSpot solver. I am using the latest 4.0 version. Can I somehow fix the temperature of an architectural block/module (say an ALU) for the entire duration of HotSpot simulation ? In such a setup, the power trace for the ALU will be ignored by the HotSpot solver. And this is while the other blocks/modules in the floorplan update their temperatures using the power trace. I would think that this can be done in the same way the ambient temperature is kept fixed within the solver, but do not have any idea how that is done either.. Thanks, Shantanu From ks4kk at cs.virginia.edu Thu Oct 18 12:13:01 2007 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Thu, 18 Oct 2007 15:13:01 -0400 (EDT) Subject: [Hotspot] Fixing a constant temperature for an architectural block In-Reply-To: <1192721411.25071.141.camel@aslan.eecs.umich.edu> References: <1192721411.25071.141.camel@aslan.eecs.umich.edu> Message-ID: Hi Shantanu, This can very well be done by explicitly setting the corresponding temperatures to the desired values in the input temperature vector before every call to compute_temp inside hotspot.c. The particular block's temperature in the silicon layer and the layers vertically beneath it have to be set. You will have to set the values appropriately accounting for the corresponding decrease in temperature layer-after-layer (due to the vertical thermal resistance) from silicon to ambient. Hope this helps. Thanks -karthik On Thu, 18 Oct 2007, Shantanu Gupta wrote: > Hi, > > I have a question regarding the working of the HotSpot solver. I am > using the latest 4.0 version. > > Can I somehow fix the temperature of an architectural block/module (say > an ALU) for the entire duration of HotSpot simulation ? In such a setup, > the power trace for the ALU will be ignored by the HotSpot solver. And > this is while the other blocks/modules in the floorplan update their > temperatures using the power trace. > > I would think that this can be done in the same way the ambient > temperature is kept fixed within the solver, but do not have any idea > how that is done either.. > > Thanks, > > Shantanu > > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > From skadron at cs.virginia.edu Thu Oct 18 13:31:48 2007 From: skadron at cs.virginia.edu (Kevin Skadron) Date: Thu, 18 Oct 2007 16:31:48 -0400 Subject: [Hotspot] Fixing a constant temperature for an architectural block In-Reply-To: References: <1192721411.25071.141.camel@aslan.eecs.umich.edu> Message-ID: <4717C2B4.30201@mail.cs.virginia.edu> Karthik is right, but a caveat: Depending on what effect you are trying to model, I'm not sure you would want to fix the temperature in the lower layers. That will disrupt the lateral heat transfer in the packaging. /K Karthik Sankaranarayanan wrote: > Hi Shantanu, > > This can very well be done by explicitly setting the corresponding > temperatures to the desired values in the input temperature vector before > every call to compute_temp inside hotspot.c. > > The particular block's temperature in the silicon layer and the layers > vertically beneath it have to be set. You will have to set the values > appropriately accounting for the corresponding decrease in temperature > layer-after-layer (due to the vertical thermal resistance) from silicon to > ambient. > > Hope this helps. > Thanks > -karthik > > > On Thu, 18 Oct 2007, Shantanu Gupta wrote: > >> Hi, >> >> I have a question regarding the working of the HotSpot solver. I am >> using the latest 4.0 version. >> >> Can I somehow fix the temperature of an architectural block/module (say >> an ALU) for the entire duration of HotSpot simulation ? In such a setup, >> the power trace for the ALU will be ignored by the HotSpot solver. And >> this is while the other blocks/modules in the floorplan update their >> temperatures using the power trace. >> >> I would think that this can be done in the same way the ambient >> temperature is kept fixed within the solver, but do not have any idea >> how that is done either.. >> >> Thanks, >> >> Shantanu >> >> >> _______________________________________________ >> HotSpot mailing list >> HotSpot at mail.cs.virginia.edu >> http://www.cs.virginia.edu/mailman/listinfo/hotspot >> > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot From shangupt at umich.edu Thu Oct 18 14:11:10 2007 From: shangupt at umich.edu (Shantanu Gupta) Date: Thu, 18 Oct 2007 17:11:10 -0400 Subject: [Hotspot] Fixing a constant temperature for an architectural block In-Reply-To: <4717C2B4.30201@mail.cs.virginia.edu> References: <1192721411.25071.141.camel@aslan.eecs.umich.edu> <4717C2B4.30201@mail.cs.virginia.edu> Message-ID: <1192741870.25071.177.camel@aslan.eecs.umich.edu> Thanks for the clarification. I agree with this problem of disrupting lateral heat transfer in the packaging. My original intent behind doing all this was to approximate the neighbor effect in a multicore system. Given the power trace for a core in the center of CMP, I wanted to fix the temperature of the surrounding (neighbor) cores. And then see the impact of varying this fixed temperature of the neighboring cores on the central core under observation. I understand that this is a gross approximation, but I was just curious to see if their is some observable lateral heat transfer in a CMP. Because some of the previous works like: * A Temperature-Aware Design Issues for SMT and CMP Architectures (WCED 04) * Heat-and-Run: Leveraging SMT and CMP to Manage Power Density Through the Operating System (ASPLOS 04) ..tend to ignore this effect. Regards, Shantanu On Thu, 2007-10-18 at 16:31 -0400, Kevin Skadron wrote: > Karthik is right, but a caveat: Depending on what effect you are trying > to model, I'm not sure you would want to fix the temperature in the > lower layers. That will disrupt the lateral heat transfer in the > packaging. > /K > > Karthik Sankaranarayanan wrote: > > Hi Shantanu, > > > > This can very well be done by explicitly setting the corresponding > > temperatures to the desired values in the input temperature vector before > > every call to compute_temp inside hotspot.c. > > > > The particular block's temperature in the silicon layer and the layers > > vertically beneath it have to be set. You will have to set the values > > appropriately accounting for the corresponding decrease in temperature > > layer-after-layer (due to the vertical thermal resistance) from silicon to > > ambient. > > > > Hope this helps. > > Thanks > > -karthik > > > > > > On Thu, 18 Oct 2007, Shantanu Gupta wrote: > > > >> Hi, > >> > >> I have a question regarding the working of the HotSpot solver. I am > >> using the latest 4.0 version. > >> > >> Can I somehow fix the temperature of an architectural block/module (say > >> an ALU) for the entire duration of HotSpot simulation ? In such a setup, > >> the power trace for the ALU will be ignored by the HotSpot solver. And > >> this is while the other blocks/modules in the floorplan update their > >> temperatures using the power trace. > >> > >> I would think that this can be done in the same way the ambient > >> temperature is kept fixed within the solver, but do not have any idea > >> how that is done either.. > >> > >> Thanks, > >> > >> Shantanu > >> > >> > >> _______________________________________________ > >> HotSpot mailing list > >> HotSpot at mail.cs.virginia.edu > >> http://www.cs.virginia.edu/mailman/listinfo/hotspot > >> > > _______________________________________________ > > HotSpot mailing list > > HotSpot at mail.cs.virginia.edu > > http://www.cs.virginia.edu/mailman/listinfo/hotspot > > > From michaelkadin at gmail.com Tue Oct 23 07:36:17 2007 From: michaelkadin at gmail.com (Mike Kadin) Date: Tue, 23 Oct 2007 10:36:17 -0400 Subject: [Hotspot] Voltage for Included Power Trace Message-ID: Hey, I had asked previously about what frequency and technology size was used for the included gcc power trace that comes with hotspot. Do you know what voltage this simulated processor was running at? Thanks, Mike Kadin From ks4kk at cs.virginia.edu Tue Oct 23 10:36:13 2007 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Tue, 23 Oct 2007 13:36:13 -0400 (EDT) Subject: [Hotspot] Voltage for Included Power Trace In-Reply-To: References: Message-ID: I believe it was at 1.3V however, please keep in mind the caveat that the power numbers are for illustrative purposes only. -karthik On Tue, 23 Oct 2007, Mike Kadin wrote: > Hey, > > I had asked previously about what frequency and technology size was used for > the included gcc power trace that comes with hotspot. Do you know what > voltage this simulated processor was running at? > > > Thanks, > Mike Kadin > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot >