From ljin at cs.ucr.edu Thu Oct 23 20:12:46 2003 From: ljin at cs.ucr.edu (Ling Ling Jin) Date: Wed Mar 22 16:04:08 2006 Subject: [Hotspot] just want temperature for a single block Message-ID: Hi: I just downloaded Hotspot and read through the paper. To my understanding, hotspot estimates the temperature for a block based on (1) the power consumed on the block (2) infection from other block's temperature. So, if I just want to know the temperature for like cache(and I have power estimation for cache block), but I don't know other blocks' power consumption, I won't be able to use hotspot and get the temperature estimation. Am I right? Thanks a lot! Lingling From skadron at cs.virginia.edu Fri Oct 24 10:14:25 2003 From: skadron at cs.virginia.edu (Kevin Skadron) Date: Wed Mar 22 16:04:08 2006 Subject: [Hotspot] just want temperature for a single block References: Message-ID: <3F9933C1.99C4FC81@cs.virginia.edu> If you only know power density for the cache, you certainly won't be able to account for lateral thermal diffusion (in either direction). Our experiments suggest that this is an important effect generally speaking, but depending on your specific experiments may or may not be essential. For the kind of experiment you're describing, you could still use HotSpot and simply use a simple floorplan consisting just of the cache and some surrounding silicon that has some average power density. That won't properly account for thermal diffusion but is better than the simplistic model we described in our first work back in HPCA 2002. Hope this helps! /K -- 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 Ling Ling Jin wrote: > > Hi: > > I just downloaded Hotspot and read through the paper. To my understanding, > hotspot estimates the temperature for a block based on (1) the power > consumed on the block (2) infection from other block's temperature. > > So, if I just want to know the temperature for like cache(and I have > power estimation for cache block), but I don't know other blocks' > power consumption, I won't be able to use hotspot and get the > temperature estimation. Am I right? > > Thanks a lot! > Lingling > > _______________________________________________ > HotSpot mailing list > HotSpot@cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot From wh6p at cms.mail.virginia.edu Fri Oct 24 10:16:33 2003 From: wh6p at cms.mail.virginia.edu (Wei Huang) Date: Wed Mar 22 16:04:08 2006 Subject: [Hotspot] just want temperature for a single block In-Reply-To: References: Message-ID: <236880468.1066990593@Lonely_Wolves> Hi Lingling, In your case, I think HotSpot can still give you reasonable cache temperature estimations. I would suggest you try the following: -- if you know or can somehow estimate the processor's total power, you may put powers other than the cache's uniformly across the die. Please note that the absolute temperature calculated from HotSpot can be different for different packages. The package in HotSpot now is a typical one that we consider to be representative. You may try different package, in that case you need to change the thermal resistance and capacitances for the package. If you have further questions, please let us know. Thanks. Wei --On 2003??10??23?? 17:12 -0700 Ling Ling Jin wrote: > Hi: > > I just downloaded Hotspot and read through the paper. To > my understanding, hotspot estimates the temperature for a > block based on (1) the power consumed on the block (2) > infection from other block's temperature. > > So, if I just want to know the temperature for like > cache(and I have power estimation for cache block), but I > don't know other blocks' power consumption, I won't be > able to use hotspot and get the temperature estimation. > Am I right? > > Thanks a lot! > Lingling > > > _______________________________________________ > HotSpot mailing list > HotSpot@cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot From wh6p at cms.mail.virginia.edu Wed Oct 29 10:49:49 2003 From: wh6p at cms.mail.virginia.edu (Wei Huang) Date: Wed Mar 22 16:04:08 2006 Subject: [Hotspot] just want temperature for a single block In-Reply-To: References: Message-ID: <83088843.1067424589@hilopc3.ee.Virginia.EDU> Hi Lingling, Maybe you can try to find some power numbers for existing ALPHA processors on the web, just a suggestion... I don't know much about SimpleScalar and Wattch, maybe other guys in the mailing list can offer some more help... Wei --On Tuesday, October 28, 2003 10:22 PM -0800 Ling Ling Jin wrote: > Hi, Wei: > > Thanks a lot for your suggestions! However, I'm having > problem with the processor power estimation part. I'm > using sim-alpha to run the simulation, and now, I'm using > sim-wattch just to estimate the cache power. But > modifying sim-wattch to estimate the whole alpha > processor's power would be too much work... Is there any > easier way to roughly get the processor's total power? > > Thanks a lot! > Lingling > From ljin at cs.ucr.edu Wed Oct 29 01:22:29 2003 From: ljin at cs.ucr.edu (Ling Ling Jin) Date: Wed Mar 22 16:04:08 2006 Subject: [Hotspot] just want temperature for a single block In-Reply-To: <236880468.1066990593@Lonely_Wolves> Message-ID: Hi, Wei: Thanks a lot for your suggestions! However, I'm having problem with the processor power estimation part. I'm using sim-alpha to run the simulation, and now, I'm using sim-wattch just to estimate the cache power. But modifying sim-wattch to estimate the whole alpha processor's power would be too much work... Is there any easier way to roughly get the processor's total power? Thanks a lot! Lingling > Hi Lingling, > > In your case, I think HotSpot can still give you reasonable > cache temperature estimations. I would suggest you try the > following: > -- if you know or can somehow estimate the processor's > total power, you may put powers other than the cache's > uniformly across the die. > Please note that the absolute temperature calculated from > HotSpot can be different for different packages. The > package in HotSpot now is a typical one that we consider to > be representative. You may try different package, in that > case you need to change the thermal resistance and > capacitances for the package. > > If you have further questions, please let us know. Thanks. > > Wei From skadron at cs.virginia.edu Thu Oct 30 16:53:55 2003 From: skadron at cs.virginia.edu (Kevin Skadron) Date: Wed Mar 22 16:04:09 2006 Subject: [Hotspot] just want temperature for a single block References: Message-ID: <3FA18873.C77C648B@cs.virginia.edu> Since all you want is processor-wide average power, can't you just use published numbers? (eg, from Microprocessor Report, Alpha documentation, etc.) /K Ling Ling Jin wrote: > > Hi, Wei: > > Thanks a lot for your suggestions! However, I'm having problem with the > processor power estimation part. I'm using sim-alpha to run the > simulation, and now, I'm using sim-wattch just to estimate the cache > power. But modifying sim-wattch to estimate the whole alpha processor's > power would be too much work... Is there any easier way to roughly get the > processor's total power? > > Thanks a lot! > Lingling > > > Hi Lingling, > > > > In your case, I think HotSpot can still give you reasonable > > cache temperature estimations. I would suggest you try the > > following: > > -- if you know or can somehow estimate the processor's > > total power, you may put powers other than the cache's > > uniformly across the die. > > > Please note that the absolute temperature calculated from > > HotSpot can be different for different packages. The > > package in HotSpot now is a typical one that we consider to > > be representative. You may try different package, in that > > case you need to change the thermal resistance and > > capacitances for the package. > > > > If you have further questions, please let us know. Thanks. > > > > Wei > > _______________________________________________ > 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