From wh6p at virginia.edu  Fri Mar  1 08:33:38 2013
From: wh6p at virginia.edu (Wei Huang)
Date: Fri, 1 Mar 2013 10:33:38 -0600
Subject: [Hotspot] Hotspot Color Map
In-Reply-To: <CANq=v5eDs9rktxemKBumHYEzV7hSE=51FxADNqiYij9ZAwEMbA@mail.gmail.com>
References: <CANq=v5eDs9rktxemKBumHYEzV7hSE=51FxADNqiYij9ZAwEMbA@mail.gmail.com>
Message-ID: <CANYeXBo7qT7VAD0gxArDJSDA07Q3sSByTtPWjd_oqcxpQ9=AYA@mail.gmail.com>

Hi,

Can you specify which thermal map are you looking at and the command line
you used to generate it? Is it the ExcelUI thermal map or is it generated
by yourself?

Thanks,

-Wei

On Thu, Feb 28, 2013 at 7:45 PM, Santanu Sarma <smilingsantanu at gmail.com>wrote:

> Hi All,
>   I have few queries for the thermal color map generated by the block
> based model. I assume that the steady state temperature for 30 blocks (in
> EV6 floor plan, 5.02 version), should be mapped to 30 color points.
> However, there are color variation even with in the blocks. How are the
> steady state temperatures mapped to the blocks in the thermal color map?
> Are there any additional temperature points computed when generating the
> thermal color map?
>
>   Also, can anyone share or point me to the a power trace file other than
> the default gcc benchmark for EV6 floor plan as included in the 5.02
> versions.
>
> Thanks
> Santanu
>
>
> On Thu, Feb 28, 2013 at 11:01 AM, Wei Huang <wh6p at virginia.edu> wrote:
>
>> Hi Johann,
>>
>> Please see my comments...
>> 1) your approach seems correct. As long as all layers have the same area,
>> adding dummy, zero-power blocks is the correct way to do this.
>> 2) I am not sure about the BU package. For HotSpot itself, you may want
>> to check whether the leakage model is enabled in the config file you are
>> using.
>> 3) HotSpot can simulate "point" heat source only at the granularity of an
>> individual grid cell. Although I doubt any tool can simulate true point
>> source, you can always use smaller grid size, at the expense of longer
>> simulation time. Also, since HotSpot grid model is more accurate when
>> simulating cell sizes greater than the die thickness, for tiny heat
>> sources, you may consider slicing the die in z direction into multiple
>> thinner layers to achieve better accuracy. Of course, this would also add
>> computation complexity. You can also try other FEM thermal simulators such
>> as ANSYS or FloTherm.
>>
>> Hope this helps.
>> -Wei
>>
>>
>> On Wed, Feb 27, 2013 at 3:59 AM, Johann Knechtel <johann.knechtel at ifte.de
>> > wrote:
>>
>>> Hi all,
>>>
>>> I am using the HotSpot extension from BU for modeling 3D ICs and got
>>> some questions.
>>>
>>> 1) I do 3D floorplanning with non-packed, fixed-outline configuration.
>>> Thus, I experience layouts where not every die's outline is defined by
>>> the block arrangement, and I get an error from HotSpot indicating the
>>> layer mismatch. In order to resolve this, I add a dummy block in each
>>> die covering the desired outline w/ zero value for specific heat and
>>> thermal resistivity. This should be reasonable and not affecting the
>>> simulation, right?
>>>
>>> 2) Is the thermal-leakage loop also working in the 3D BU extension? For
>>> some initial experiments I couldn't see any difference in the results,
>>> but I am not sure whether this is an issue with my settings or an
>>> general issue.
>>>
>>> 3) During floorplanning-optimization loops, I want to use power blurring
>>> for fast thermal estimation and then subsequently evaluate the results
>>> using HotSpot thermal simulation. For power blurring, I need a thermal
>>> mask (impulse response of layers obtained by applying point heat
>>> sources), which I also generated using HotSpot until now.
>>> However, I am not sure whether the accuracy of the grid-based simulation
>>> is sufficient; for a point source w/ large power values, the thermal
>>> distribution seems very limited to some grid bins, even in cases where
>>> the power source is in an adjacent layer and thus thermal spreading
>>> should occur.
>>> Can somebody comment on that? Is HotSpot in general suitable for
>>> modelling such point sources?
>>>
>>> Thanks,
>>> Johann
>>> --
>>> http://www.ifte.de/english/staff/knechtel.html
>>> TU Dresden, IFTE, BAR II/47, Phone: +49 351 463 39612
>>> _______________________________________________
>>> 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
>>
>>
>
> _______________________________________________
> HotSpot mailing list
> HotSpot at mail.cs.virginia.edu
> http://www.cs.virginia.edu/mailman/listinfo/hotspot
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20130301/1bb7929f/attachment.html 

From soumyaj at ece.iitkgp.ernet.in  Tue Mar  5 02:52:55 2013
From: soumyaj at ece.iitkgp.ernet.in (Soumya J)
Date: Tue, 5 Mar 2013 16:22:55 +0530 (IST)
Subject: [Hotspot] Power values at different time instances
In-Reply-To: <536327496.129580.1362480522269.JavaMail.root@mailstore1.iitkgp.ac.in>
Message-ID: <2061148925.129604.1362480775258.JavaMail.root@mailstore1.iitkgp.ac.in>

Hi,

What is the time interval between the power values given in .ptrace file.
Is it same as the sampling interval? If so, if i have different power values for the blocks at different time instances, how to specify those values in ptrace file? The time difference between two different power values is not same.
-- 
SOUMYA.J
Research Scholar
E&ECE Department
IIT Kharagpur

From ks4kk at virginia.edu  Tue Mar  5 08:16:26 2013
From: ks4kk at virginia.edu (Karthik Sankaranarayanan)
Date: Tue, 5 Mar 2013 11:16:26 -0500
Subject: [Hotspot] Power values at different time instances
In-Reply-To: <2061148925.129604.1362480775258.JavaMail.root@mailstore1.iitkgp.ac.in>
References: <536327496.129580.1362480522269.JavaMail.root@mailstore1.iitkgp.ac.in>
	<2061148925.129604.1362480775258.JavaMail.root@mailstore1.iitkgp.ac.in>
Message-ID: <CAPE+kwBcNW7VYcq2JN23WuwihfRu7=fRfZwv3N_KuWz3rq4HQg@mail.gmail.com>

Hi Soumya,

What is the time interval between the power values given in .ptrace file.
> Is it same as the sampling interval?


Yes. That is correct.


> If so, if i have different power values for the blocks at different time
> instances, how to specify those values in ptrace file? The time difference
> between two different power values is not same.
>

You have two options:

1) Sample your non-uniform data to be uniform

2) Modify hotspot.c to read the time interval from the trace file as well

Hope this helps.
-karthik
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20130305/2abbc10e/attachment.html 

From mircea at virginia.edu  Tue Mar  5 10:01:38 2013
From: mircea at virginia.edu (Mircea R. Stan)
Date: Tue, 05 Mar 2013 13:01:38 -0500
Subject: [Hotspot] Power values at different time instances
In-Reply-To: <CAPE+kwBcNW7VYcq2JN23WuwihfRu7=fRfZwv3N_KuWz3rq4HQg@mail.gmail.com>
References: <536327496.129580.1362480522269.JavaMail.root@mailstore1.iitkgp.ac.in>
	<2061148925.129604.1362480775258.JavaMail.root@mailstore1.iitkgp.ac.in>
	<CAPE+kwBcNW7VYcq2JN23WuwihfRu7=fRfZwv3N_KuWz3rq4HQg@mail.gmail.com>
Message-ID: <web-4517994@neon.mail.virginia.edu>

> You have two options:
> 
> 1) Sample your non-uniform data to be uniform
> 
> 2) Modify hotspot.c to read the time interval from the trace file as well

I think option 1) is more straightforward as it can be done outside the 
Hotspot
simulator loop. What you really want to do is to *resample* your power data.
Resampling is a relatively well understood issue in communication theory and
signal processing with solid theory and algorithms available. Hope this helps!

M

> 
> Hope this helps.
> -karthik

Mircea Stan

The views expressed in this email represent the personal
views of the sender and do not represent the official
position of the University of Virginia

From soumyaj at ece.iitkgp.ernet.in  Thu Mar  7 05:21:26 2013
From: soumyaj at ece.iitkgp.ernet.in (Soumya J)
Date: Thu, 7 Mar 2013 18:51:26 +0530 (IST)
Subject: [Hotspot] Floorplan with empty spaces
In-Reply-To: <2042174168.146179.1362662284131.JavaMail.root@mailstore1.iitkgp.ac.in>
Message-ID: <264778321.146185.1362662486090.JavaMail.root@mailstore1.iitkgp.ac.in>

Hi,

I have a floorplan with empty spaces in between.
1.I gave floorplan with empty spaces as input to Hotspot without giving any information about empty spaces.
2.I gave floorplan as input by giving coordinates of empty spaces and gave power value as zero to those blocks.

I observed both the cases result in the same temperature profile. Is it ok if i ignore the empty spaces?

-- 
SOUMYA.J
Research Scholar
E&ECE Department
IIT Kharagpur

From smillican at wisc.edu  Thu Mar  7 08:44:51 2013
From: smillican at wisc.edu (Spencer Millican)
Date: Thu, 7 Mar 2013 10:44:51 -0600
Subject: [Hotspot] Floorplan with empty spaces
In-Reply-To: <264778321.146185.1362662486090.JavaMail.root@mailstore1.iitkgp.ac.in>
References: <2042174168.146179.1362662284131.JavaMail.root@mailstore1.iitkgp.ac.in>
	<264778321.146185.1362662486090.JavaMail.root@mailstore1.iitkgp.ac.in>
Message-ID: <CAFT6x5eZ637H2zvk-fnALy3n-wjoGj4eDWXxZCerrP-7_XSmWg@mail.gmail.com>

Souma,

If you give empty spaces in your floor plan  it will reek chaos in your
temperature simulations. Fill in your empty spaces with "dummy modules"
with no power dissipation.

-Spencer, UW Madison

On Thu, Mar 7, 2013 at 7:21 AM, Soumya J <soumyaj at ece.iitkgp.ernet.in>wrote:

> Hi,
>
> I have a floorplan with empty spaces in between.
> 1.I gave floorplan with empty spaces as input to Hotspot without giving
> any information about empty spaces.
> 2.I gave floorplan as input by giving coordinates of empty spaces and gave
> power value as zero to those blocks.
>
> I observed both the cases result in the same temperature profile. Is it ok
> if i ignore the empty spaces?
>
> --
> SOUMYA.J
> Research Scholar
> E&ECE Department
> IIT Kharagpur
> _______________________________________________
> HotSpot mailing list
> HotSpot at mail.cs.virginia.edu
> http://www.cs.virginia.edu/mailman/listinfo/hotspot
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20130307/d5dc275d/attachment.html 

From wh6p at virginia.edu  Tue Mar 12 07:52:20 2013
From: wh6p at virginia.edu (Wei Huang)
Date: Tue, 12 Mar 2013 10:52:20 -0500
Subject: [Hotspot] Queries_on_Hotspot
In-Reply-To: <CALC-P5eQBEfdzx4u2Fii7pnGMMJ4CA5jKyUwuSOEhj+8+14v-Q@mail.gmail.com>
References: <CALC-P5eQBEfdzx4u2Fii7pnGMMJ4CA5jKyUwuSOEhj+8+14v-Q@mail.gmail.com>
Message-ID: <CANYeXBoY7JAA4cZywY1wQyvDF3iAoX=XW+1ZyWbuBrowOzJ-fg@mail.gmail.com>

Hi Payel,

You would need a floorplan file for the SOC you are modeling. Please make
sure there is no blank spaces or overlapping areas between blocks. For
actual blank spaces in silicon, you can model them as blocks that dissipate
zero power in HotSpot. Sample floorplan files can be found in the release
as *.flp files. All dimensions are in meters.

You would also need a power trace file, which has columns corresponding to
the power of the blocks defined in your floorplan file. It seems you want
to know steady state temperatures of each block -- so you only need to
provide one line in the power trace file. Power unit is Watt.

If you know some details about the thermal package for this SOC, you would
also need to go through hotspot.config and make adjustments there.

Command-line details to run HotSpot can be found at
http://lava.cs.virginia.edu/HotSpot/HotSpot-HOWTO.htm

Hope this is helpful.

-Wei


On Tue, Mar 12, 2013 at 7:27 AM, Payel Ghosh <ghoshpayel89 at gmail.com> wrote:

> Hotspot Team Members,
>
> I am using Hotspot ( Version 5.02 ) for the fulfillment of my project
> work.
> I am sending this mail to you because I am facing some problems
> regarding it's uses.
>
> 1) Could you please tell me how can I get temperature for a specific SOC?
> I have the information about SOC like the
> power values of its different cores and area related information( which
> are calculated with the help of ITC'02 benchmark information ).
> I am sending you the information of SOC d695 for example.
>
> ------------------------------
> --------------------------------------------------------------------------------------------------------------------------------------
>
> Core_id        Width(m)    Height(m)    Left-x        Bottom-y    Power
> values
>
> 1        0.016000    0.038000    0.000000    0.000000    660
> 2        0.016000    0.813000    0.001600    0.000000    602
> 3        0.016000    2.507000    0.032000    0.000000    823
> 4        0.010667    3.000000    0.000000    0.040000    275
> 5        0.018000    3.000067    0.013000    1.001000    690
> 6        0.446000    3.000000    0.030000    2.509800    354
> 7        0.212667    3.000000    0.133900    0.000000    530
> 8        0.178000    3.000000    0.129000    3.200000    753
> 9        0.059667    3.000700    0.074000    0.000000    641
> 10      0.576000    3.000700    0.074000    3.200000    1144
>
> -----------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
>
> Also tell me what are the different input files I should modify
> accordingly and how to get all those input values.
>
> 2) Secondly, How can I  use block model?
>
> I will be very much obliged if you send necessary answers regarding above
> queries.
>
>
> Thanks and Regards,
> Payel Ghosh
> M.Tech (Information Technology)
> Bengal Engineering and Science University,Shibpur.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20130312/93760dd8/attachment.html 

From littlepretty at yahoo.cn  Thu Mar 14 14:06:07 2013
From: littlepretty at yahoo.cn (Jiaqi Yan)
Date: Thu, 14 Mar 2013 22:06:07 +0000 (UTC)
Subject: [Hotspot] Invitation to connect on LinkedIn
Message-ID: <1155463281.6633735.1363298767281.JavaMail.app@ela4-app2320.prod>

LinkedIn
------------



I'd like to add you to my professional network on LinkedIn.

- Jiaqi

Jiaqi Yan
Student at Beijing Normal University
China

Confirm that you know Jiaqi Yan:
https://www.linkedin.com/e/-uqlkxy-heagvx5y-z/isd/11683556108/YjyGzWFT/?hs=false&tok=0Nhy0rfZ5KJ5E1

--
You are receiving Invitation to Connect emails. Click to unsubscribe:
http://www.linkedin.com/e/-uqlkxy-heagvx5y-z/Ksd9H1Y52gQPGBRkWqZ-7L3EUrLJp-Bzz0Qy/goo/hotspot%40cs%2Evirginia%2Eedu/20061/I3867372292_1/?hs=false&tok=1Mkan-zm9KJ5E1

(c) 2012 LinkedIn Corporation. 2029 Stierlin Ct, Mountain View, CA 94043, USA.


  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20130314/332a5ed5/attachment.html 

From khurshid at wisc.edu  Sun Mar 24 11:34:19 2013
From: khurshid at wisc.edu (Mushfique Khurshid)
Date: Sun, 24 Mar 2013 11:34:19 -0500
Subject: [Hotspot] Difference between transient and steady state temperatures
Message-ID: <CAKLY5E=+Js3-BLNN=HLhSuJpRZPYRALZhwawAdYnXwJnrcRz9A@mail.gmail.com>

Hi,

I am simulating a multilayer 3d stacked chip. I removed heat sink and heat
spreader by making their thicknesses 10 microns as suggested in a previous
digest. So now I simulate the hotspot with a power trace having constant
power for a long time. The temperature trace file gives us temperature
trace (transient) while in the end I print out the STEADY STATE temperature
of all the layers. The transient temperatures are significantly lower than
the steady state temperature. The transient temperatures are in the order
of 347K while steady state is 385K. In the transient trace, I never see the
temperature eventually rise to 385K in the ttrace file. At the end of the
trace it is 347K. Why is this difference?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20130324/ba1b08aa/attachment.html 

From mircea at virginia.edu  Mon Mar 25 07:23:45 2013
From: mircea at virginia.edu (Mircea Stan)
Date: Mon, 25 Mar 2013 08:23:45 -0400
Subject: [Hotspot] Difference between transient and steady state
	temperatures
In-Reply-To: <CAKLY5E=+Js3-BLNN=HLhSuJpRZPYRALZhwawAdYnXwJnrcRz9A@mail.gmail.com>
References: <CAKLY5E=+Js3-BLNN=HLhSuJpRZPYRALZhwawAdYnXwJnrcRz9A@mail.gmail.com>
Message-ID: <9C6262DD-84D5-441A-9EE7-FF0E4AB26816@virginia.edu>

I think the problem is that you made the heat sink too thin. Hotspot is an approximate 2D solution 
to a 3D problem, with trade offs between accuracy and computation speed; thus there is
always the issue of what is an acceptable range for various parameters. In general, if you
try unrealistic parameters you should not expect reasonable results - 10 micron thick heatsink
is not realistic. As to why the difference between steady state and transient happens to be
large I think it has to do with the difference in solvers, but again, the real problem is that the
assumptions are unrealistic so probably both sets of results are not to be trusted. Hope this
helps!

Mircea Stan

On Mar 24, 2013, at 12:34 PM, Mushfique Khurshid <khurshid at wisc.edu> wrote:

> Hi,
> 
> I am simulating a multilayer 3d stacked chip. I removed heat sink and heat spreader by making their thicknesses 10 microns as suggested in a previous digest. So now I simulate the hotspot with a power trace having constant power for a long time. The temperature trace file gives us temperature trace (transient) while in the end I print out the STEADY STATE temperature of all the layers. The transient temperatures are significantly lower than the steady state temperature. The transient temperatures are in the order of 347K while steady state is 385K. In the transient trace, I never see the temperature eventually rise to 385K in the ttrace file. At the end of the trace it is 347K. Why is this difference?
> _______________________________________________
> HotSpot mailing list
> HotSpot at mail.cs.virginia.edu
> http://www.cs.virginia.edu/mailman/listinfo/hotspot


From skadron at cs.virginia.edu  Mon Mar 25 09:45:45 2013
From: skadron at cs.virginia.edu (Kevin Skadron)
Date: Mon, 25 Mar 2013 10:45:45 -0400
Subject: [Hotspot] Difference between transient and steady state
	temperatures
In-Reply-To: <9C6262DD-84D5-441A-9EE7-FF0E4AB26816@virginia.edu>
References: <CAKLY5E=+Js3-BLNN=HLhSuJpRZPYRALZhwawAdYnXwJnrcRz9A@mail.gmail.com>
	<9C6262DD-84D5-441A-9EE7-FF0E4AB26816@virginia.edu>
Message-ID: <51506319.5020201@cs.virginia.edu>

It would help us if we can see your floorplan, power trace file, and 
configuration file. Please send directly to me and CC Mircea and Wei.
Thanks,
/K

On 3/25/2013 8:23 AM, Mircea Stan wrote:
> I think the problem is that you made the heat sink too thin. Hotspot is an approximate 2D solution
> to a 3D problem, with trade offs between accuracy and computation speed; thus there is
> always the issue of what is an acceptable range for various parameters. In general, if you
> try unrealistic parameters you should not expect reasonable results - 10 micron thick heatsink
> is not realistic. As to why the difference between steady state and transient happens to be
> large I think it has to do with the difference in solvers, but again, the real problem is that the
> assumptions are unrealistic so probably both sets of results are not to be trusted. Hope this
> helps!
>
> Mircea Stan
>
> On Mar 24, 2013, at 12:34 PM, Mushfique Khurshid <khurshid at wisc.edu> wrote:
>
>> Hi,
>>
>> I am simulating a multilayer 3d stacked chip. I removed heat sink and heat spreader by making their thicknesses 10 microns as suggested in a previous digest. So now I simulate the hotspot with a power trace having constant power for a long time. The temperature trace file gives us temperature trace (transient) while in the end I print out the STEADY STATE temperature of all the layers. The transient temperatures are significantly lower than the steady state temperature. The transient temperatures are in the order of 347K while steady state is 385K. In the transient trace, I never see the temperature eventually rise to 385K in the ttrace file. At the end of the trace it is 347K. Why is this difference?
>> _______________________________________________
>> 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 wh6p at virginia.edu  Mon Mar 25 11:09:06 2013
From: wh6p at virginia.edu (Wei Huang)
Date: Mon, 25 Mar 2013 11:09:06 -0500
Subject: [Hotspot] Difference between transient and steady state
	temperatures
In-Reply-To: <51506319.5020201@cs.virginia.edu>
References: <CAKLY5E=+Js3-BLNN=HLhSuJpRZPYRALZhwawAdYnXwJnrcRz9A@mail.gmail.com>
	<9C6262DD-84D5-441A-9EE7-FF0E4AB26816@virginia.edu>
	<51506319.5020201@cs.virginia.edu>
Message-ID: <CANYeXBqOq8ZFbz8m9K=P90NvX9Qs0mDpF2rpKZhXxD5pgvDjNA@mail.gmail.com>

Hi,

you mentioned you simulated for a long time. Could you elaborate how long
it is? this can be calculated by number of entries in your power trace file
and the sampling rate settings in you config file. Usually, if one wants to
simulate the thermal warm-up phase, which usually takes minutes in an
actual package, and probably days in software simulations in tools such as
HotSpot. We normally wouldn't recommend to simulate the thermal warm-up
phase. But if that is really what you want, you may have to run MUCH longer
than what you did.

Another setting that might worth checking is to make sure your initial
temperature is the same as ambient temperature.

-Wei


On Mon, Mar 25, 2013 at 9:45 AM, Kevin Skadron <skadron at cs.virginia.edu>wrote:

> It would help us if we can see your floorplan, power trace file, and
> configuration file. Please send directly to me and CC Mircea and Wei.
> Thanks,
> /K
>
>
> On 3/25/2013 8:23 AM, Mircea Stan wrote:
>
>> I think the problem is that you made the heat sink too thin. Hotspot is
>> an approximate 2D solution
>> to a 3D problem, with trade offs between accuracy and computation speed;
>> thus there is
>> always the issue of what is an acceptable range for various parameters.
>> In general, if you
>> try unrealistic parameters you should not expect reasonable results - 10
>> micron thick heatsink
>> is not realistic. As to why the difference between steady state and
>> transient happens to be
>> large I think it has to do with the difference in solvers, but again, the
>> real problem is that the
>> assumptions are unrealistic so probably both sets of results are not to
>> be trusted. Hope this
>> helps!
>>
>> Mircea Stan
>>
>> On Mar 24, 2013, at 12:34 PM, Mushfique Khurshid <khurshid at wisc.edu>
>> wrote:
>>
>>  Hi,
>>>
>>> I am simulating a multilayer 3d stacked chip. I removed heat sink and
>>> heat spreader by making their thicknesses 10 microns as suggested in a
>>> previous digest. So now I simulate the hotspot with a power trace having
>>> constant power for a long time. The temperature trace file gives us
>>> temperature trace (transient) while in the end I print out the STEADY STATE
>>> temperature of all the layers. The transient temperatures are significantly
>>> lower than the steady state temperature. The transient temperatures are in
>>> the order of 347K while steady state is 385K. In the transient trace, I
>>> never see the temperature eventually rise to 385K in the ttrace file. At
>>> the end of the trace it is 347K. Why is this difference?
>>> ______________________________**_________________
>>> HotSpot mailing list
>>> HotSpot at mail.cs.virginia.edu
>>> http://www.cs.virginia.edu/**mailman/listinfo/hotspot<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<http://www.cs.virginia.edu/mailman/listinfo/hotspot>
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20130325/ed482531/attachment.html 

From wh6p at virginia.edu  Mon Mar 25 16:48:05 2013
From: wh6p at virginia.edu (Wei Huang)
Date: Mon, 25 Mar 2013 16:48:05 -0500
Subject: [Hotspot] Difference between transient and steady state
	temperatures
In-Reply-To: <CAKLY5E=XDFGTsdOKXvUUG6BE=p-Q+wNsY5tFYZcbg9bo1dcJTg@mail.gmail.com>
References: <CAKLY5E=+Js3-BLNN=HLhSuJpRZPYRALZhwawAdYnXwJnrcRz9A@mail.gmail.com>
	<9C6262DD-84D5-441A-9EE7-FF0E4AB26816@virginia.edu>
	<51506319.5020201@cs.virginia.edu>
	<CAKLY5E=XDFGTsdOKXvUUG6BE=p-Q+wNsY5tFYZcbg9bo1dcJTg@mail.gmail.com>
Message-ID: <CANYeXBoaDyiuy_Lk9SqEMJBcgEpQsnWTgfokYyVwhqHn0rnjpA@mail.gmail.com>

Hi,

Your power trace only has 526 lines (and by the way, it is not constant
power), together with a setting of sampling interval of every 3.33us (in
.config file), you are only able to simulate about 1.5ms worth of time with
this setup. That is FAR from getting temperature to steady state.

Hope this helps.

-Wei


On Mon, Mar 25, 2013 at 12:34 PM, Mushfique Khurshid <khurshid at wisc.edu>wrote:

> Attached is my lcf, flp, ptrace and config files as requested. I get same
> difference between transient and steady state even after i changed the
> thicknesses of sink/spreader back to normal
>
>
>
> On Mon, Mar 25, 2013 at 9:45 AM, Kevin Skadron <skadron at cs.virginia.edu>wrote:
>
>> It would help us if we can see your floorplan, power trace file, and
>> configuration file. Please send directly to me and CC Mircea and Wei.
>> Thanks,
>> /K
>>
>>
>> On 3/25/2013 8:23 AM, Mircea Stan wrote:
>>
>>> I think the problem is that you made the heat sink too thin. Hotspot is
>>> an approximate 2D solution
>>> to a 3D problem, with trade offs between accuracy and computation speed;
>>> thus there is
>>> always the issue of what is an acceptable range for various parameters.
>>> In general, if you
>>> try unrealistic parameters you should not expect reasonable results - 10
>>> micron thick heatsink
>>> is not realistic. As to why the difference between steady state and
>>> transient happens to be
>>> large I think it has to do with the difference in solvers, but again,
>>> the real problem is that the
>>> assumptions are unrealistic so probably both sets of results are not to
>>> be trusted. Hope this
>>> helps!
>>>
>>> Mircea Stan
>>>
>>> On Mar 24, 2013, at 12:34 PM, Mushfique Khurshid <khurshid at wisc.edu>
>>> wrote:
>>>
>>>  Hi,
>>>>
>>>> I am simulating a multilayer 3d stacked chip. I removed heat sink and
>>>> heat spreader by making their thicknesses 10 microns as suggested in a
>>>> previous digest. So now I simulate the hotspot with a power trace having
>>>> constant power for a long time. The temperature trace file gives us
>>>> temperature trace (transient) while in the end I print out the STEADY STATE
>>>> temperature of all the layers. The transient temperatures are significantly
>>>> lower than the steady state temperature. The transient temperatures are in
>>>> the order of 347K while steady state is 385K. In the transient trace, I
>>>> never see the temperature eventually rise to 385K in the ttrace file. At
>>>> the end of the trace it is 347K. Why is this difference?
>>>> ______________________________**_________________
>>>> HotSpot mailing list
>>>> HotSpot at mail.cs.virginia.edu
>>>> http://www.cs.virginia.edu/**mailman/listinfo/hotspot<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<http://www.cs.virginia.edu/mailman/listinfo/hotspot>
>>>
>>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20130325/8e733609/attachment.html