From ks4kk at cs.virginia.edu Fri Jun 20 13:18:20 2003 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Wed Mar 22 16:04:07 2006 Subject: [Hotspot] Calling compute temp In-Reply-To: <001f01c3373d$1c063f10$2714cb82@mdl.cse.psu.edu> References: <200305251509.LAA10189@webmail9.cac.psu.edu><001301c32470$57d24140$2714cb82@mdl.cse.psu.edu> <001f01c3373d$1c063f10$2714cb82@mdl.cse.psu.edu> Message-ID: Hi Ananth, Yes the release version is available on the web. Please visit http://lava.cs.virginia.edu/HotSpot I feel the problem you are facing might be due to wrong initial temperatures. If you start from a value vastly different from the steady state temperature, then it might take very long to heat up. I would urge you to replace your current version (that was our internal version) with the release version. It has many aesthetic changes and a few important bug fixes. Some of those aesthetic changes were motivated by the questions you and others asked in the HotSpot mailing list - thank you. Please read the documentation with the release version (README file). It provides a detailed, step-by-step 'things-to-do' list for using HotSpot. I hope it is useful. Thanks -karthik On Fri, 20 Jun 2003, Ananth Hegde wrote: > Karthik, > I was wondering if the new release of HotSpot was available yet. I am still > encountering problems or may be its not a problem at all. > I took a processor configuration at 180nm ran 132.ijpeg for 961085029 cycles > calling compute_temp every 10000 cycles.While the temperatures were in the > range of 299-302 starting at initial temp of 298, the corresponding steady > state temperatures were as high as 355-380. I tried reducing the interval to > 1000 cycles. It didnt help either. In your paper you say that hotspot has > been validated only for steady state temp. Is it because of that? Or am i > running it for too few cycles for the chip to heat up? > > Thanks > Ananth > > > ----- Original Message ----- > From: "Karthik Sankaranarayanan" > To: "Ananth Hegde" > Sent: Tuesday, May 27, 2003 1:12 PM > Subject: Re: [Hotspot] Calling compute temp > > > > Ananth, > > > > > This is what I think. Correct me if I am wrong. For finding the averages > > > "elapsed_time" means the no. of cycles (not in seconds) and for > > > "compute_temp","elapsed_time" means no. of cycles*PERIOD.(it is in > seconds) > > > > Yes, exactly - you are correct. > > > > -karthik > > > > > >