From wh6p at cms.mail.virginia.edu  Tue Apr  1 03:56:50 2008
From: wh6p at cms.mail.virginia.edu (Wei Huang)
Date: Tue, 01 Apr 2008 07:56:50 -0400
Subject: [Hotspot] [Fwd: HotSpot and heat sinks]
In-Reply-To: <47F19595.6070003@cs.virginia.edu>
References: <47F19595.6070003@cs.virginia.edu>
Message-ID: <13C726C4BE77FF1E4038B648@[192.168.1.102]>

Hi Brett,

HotSpot 4.0 is able to model chips without heatsink. What 
you can do is to change the material properties of 
sink/spreader/TIM to model for example the 
ceramic/plastic/metal lid that connects the chip with 
ambient air. You may also want to set the area of 
"sink/spreader" to the desired convection area (but their 
areas have to satisfy A_sink>A_spreader>A_TIM=A_chip). The 
next thing to do is to apply the right heat transfer 
coefficient h in temperature.h, h=1/(Rth*Area), where Rth 
is the convective thermal resistance that is usually 
specified in the package datasheet. This should do most of 
what you want.

One possible caveat is that without a heatsink, the heat 
transferred through the secondary heat transfer path (C4 
pads, solder balls and PCB) might not be negligible 
depending on your choice of package/materials. HotSpot 
cannot be explicitly configured to model this path yet. Let 
us know if it turns out that you need to do this, and we 
may figure something out.

Hope this helps,

-Wei

--On Monday, March 31, 2008 9:53 PM -0400 Kevin Skadron 
<skadron at cs.virginia.edu> wrote:

> Wei, can you respond?
> /K
>
> -------- Original Message --------
> Subject: HotSpot and heat sinks
> Date: Mon, 31 Mar 2008 14:57:28 -0400
> From: Brett Meyer <brettmeyer at cmu.edu>
> To: Kevin Skadron <skadron at cs.virginia.edu>
>
> Professor Skadron,
>
> I'm a graduate student at Carnegie Mellon University
> investigating
> temperature-dependent reliability in embedded systems,
> and we are
> considering HotSpot for our temperature modeling needs.
> Many such
> systems have minimal cooling budgets, and perhaps lack a
> heat sink
> entirely.  Has HotSpot been constructed in such a way
> that the
> assumption of a heat sink could be adjusted for our
> domain, and if so,
> have you or anyone else pursued this matter?  In
> particular, I'm
> looking for guidance on how command-line parameters to
> HotSpot
> (thermal conductivity, thermal resistance, etc.) might be
> changed to
> emulate a system with no heat sink.
>
> Thank you for your time,
>
> Brett Meyer
> Department of Electrical and Computer Engineering
> Carnegie Mellon University






From eby_buendia at hotmail.com  Sun Apr 20 18:29:16 2008
From: eby_buendia at hotmail.com (Encarna Buendia)
Date: Mon, 21 Apr 2008 01:29:16 +0000
Subject: [Hotspot] on the power-temperature relationship
Message-ID: <BLU124-W49EC1F33218F3EC64025BD9EE10@phx.gbl>


I guess I have two basic questions that go beyond the Hotspot tool itself and fall more into how power and temperature relate to each other:
 
1/ static power depends exponentially on the temperature. So, if we have two blocks A and B side by side, the temperature of block A will affect the power consumption of block B. Higher power consumption in block B will raise its temperature, affecting the power of block A, and so on and forth. Then, what prevents the temperature in both blocks from skyrocketing? Is it that at certain power level, the increase in temperature is negligible so the system stabilizes?
 
2/ the Hostpot tool requires a power trace to obtain the temperature map. But again, how accurate that methodology is? If the power of a block gets affected by the temperature of surrounding blocks, the power W of block B at time T offered in the trace may actually be X% optimistic if that block is in the middle of a hotspot at time T, or Y% pessimistic if it is in a cool part of the die. So, is the methodology of feeding a power trace correct, or should the temperature info somehow fed back to the power trace?
 
Thanks,
Eby

_________________________________________________________________
Get in touch in an instant. Get Windows Live Messenger now.
http://www.windowslive.com/messenger/overview.html?ocid=TXT_TAGLM_WL_Refresh_getintouch_042008
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20080421/f9c2aff0/attachment.html 

From bernauer at informatik.uni-tuebingen.de  Mon Apr 21 00:45:15 2008
From: bernauer at informatik.uni-tuebingen.de (Andreas Bernauer)
Date: Mon, 21 Apr 2008 09:45:15 +0200
Subject: [Hotspot] on the power-temperature relationship
In-Reply-To: <BLU124-W49EC1F33218F3EC64025BD9EE10@phx.gbl>
References: <BLU124-W49EC1F33218F3EC64025BD9EE10@phx.gbl>
Message-ID: <1208763915.6996.12.camel@lava>


On Mon, 2008-04-21 at 01:29 +0000, Encarna Buendia wrote:
> 1/ static power depends exponentially on the temperature. So, if we
> have two blocks A and B side by side, the temperature of block A will
> affect the power consumption of block B. Higher power consumption in
> block B will raise its temperature, affecting the power of block A,
> and so on and forth. Then, what prevents the temperature in both
> blocks from skyrocketing? 

Usually, the chip is not in isolation, but surrounded by other matter
and air.  Thus, heat dissipates into the surrounding, either by heat
conduction or convection.  That's why on modern desktop systems you have
a fan on top of the CPU: you increase (force) heat convection.  If the
heat generated by A and B is less then or equal to the heat dissipation,
the system's temperature cannot skyrocket.

>  
> 2/ the Hostpot tool requires a power trace to obtain the temperature
> map. But again, how accurate that methodology is? If the power of a
> block gets affected by the temperature of surrounding blocks, the
> power W of block B at time T offered in the trace may actually be X%
> optimistic if that block is in the middle of a hotspot at time T, or Y
> % pessimistic if it is in a cool part of the die. So, is the
> methodology of feeding a power trace correct, or should the
> temperature info somehow fed back to the power trace?
>  

Within a certain range you can neglect the effect of the increased
temperature on the power consumption.  However, if you want to be
accurate, you have need to have Hotspot 'in the loop', that is you must
estimate power and temperature value simultaneously (as we do in our
simulations).

Regards,

Andreas.
________________________________________________________________________
-- 
WSI/TI, Sand 13, 72076 T?bingen, +49 70 71 29 75 940
http://www.ti.uni-tuebingen.de/Andreas_Bernauer.227.0.html
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://www.cs.virginia.edu/pipermail/hotspot/attachments/20080421/8445ec6a/attachment.bin 

From mrs8n at cms.mail.virginia.edu  Mon Apr 21 07:27:35 2008
From: mrs8n at cms.mail.virginia.edu (Mircea R. Stan)
Date: Mon, 21 Apr 2008 10:27:35 -0400
Subject: [Hotspot] on the power-temperature relationship
In-Reply-To: <BLU124-W49EC1F33218F3EC64025BD9EE10@phx.gbl>
References: <BLU124-W49EC1F33218F3EC64025BD9EE10@phx.gbl>
Message-ID: <web-4332264@koi.mail.virginia.edu>

On Mon, 21 Apr 2008 01:29:16 +0000
  Encarna Buendia <eby_buendia at hotmail.com> wrote:
> 
> I guess I have two basic questions that go beyond the Hotspot tool itself 
>and fall more into how power and temperature relate to each other:
> 
> 1/ static power depends exponentially on the temperature. So, if we have 
>two blocks A and B side by side, the temperature of block A will affect the 
>power consumption of block B. Higher power consumption in block B will 
>raise its temperature, affecting the power of block A, and so on and forth. 
>Then, what prevents the temperature in both blocks from skyrocketing? Is it 
>that at certain power level, the increase in temperature is negligible so 
>the system stabilizes?

One way to think about this situation is like a system with feedback,
you have block A which is the input (the power being dissipated and
the corresponding raise in temperature), and block B which provides the
feedback (through it's increased leakage due to the raise in temperature,
which affects block A). If the "gain" of this feedback loop is larger
than 1, then you have an unstable system and thermal runaway,
if the gain is less than 1 then the temperature stabilizes at some value.
The two mechanisms to affect this gain are: package, basically a
smaller thermal resistance of the package decreases the "gain"
of this feedback loop, or floorplanning, by placing the blocks farther
apart again you decrease the "gain". Hope this helps!

> 
> 2/ the Hostpot tool requires a power trace to obtain the temperature map. 
>But again, how accurate that methodology is? If the power of a block gets 
>affected by the temperature of surrounding blocks, the power W of block B 
>at time T offered in the trace may actually be X% optimistic if that block 
>is in the middle of a hotspot at time T, or Y% pessimistic if it is in a 
>cool part of the die. So, is the methodology of feeding a power trace 
>correct, or should the temperature info somehow fed back to the power 
>trace?

This is called "self-consistent" leakage calculation, and it involves using
Hotspot in the simulation loop, making sure you iterate over the leakage
and temperature until the two converge (assuming you don't get into
thermal runaway when the temperature would keep increasing). Again
I hope this helps!

> 
> Thanks,
> Eby
> 
> _________________________________________________________________
> Get in touch in an instant. Get Windows Live Messenger now.
> http://www.windowslive.com/messenger/overview.html?ocid=TXT_TAGLM_WL_Refresh_getintouch_042008

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 wh6p at cms.mail.virginia.edu  Mon Apr 21 07:36:15 2008
From: wh6p at cms.mail.virginia.edu (Wei Huang)
Date: Mon, 21 Apr 2008 10:36:15 -0400
Subject: [Hotspot] on the power-temperature relationship
In-Reply-To: <1208763915.6996.12.camel@lava>
References: <BLU124-W49EC1F33218F3EC64025BD9EE10@phx.gbl>
	<1208763915.6996.12.camel@lava>
Message-ID: <1B20ED01CE0549CE274B13E1@mstu2>

--On Monday, April 21, 2008 9:45 AM +0200 Andreas Bernauer 
<bernauer at informatik.uni-tuebingen.de> wrote:
> On Mon, 2008-04-21 at 01:29 +0000, Encarna Buendia wrote:
>> 1/ static power depends exponentially on the
>> temperature. So, if we have two blocks A and B side by
>> side, the temperature of block A will affect the power
>> consumption of block B. Higher power consumption in
>> block B will raise its temperature, affecting the power
>> of block A, and so on and forth. Then, what prevents the
>> temperature in both blocks from skyrocketing?
>
> Usually, the chip is not in isolation, but surrounded by
> other matter and air.  Thus, heat dissipates into the
> surrounding, either by heat conduction or convection.
> That's why on modern desktop systems you have a fan on
> top of the CPU: you increase (force) heat convection.  If
> the heat generated by A and B is less then or equal to
> the heat dissipation, the system's temperature cannot
> skyrocket.

The situation Eby described is called "thermal runaway". As 
Andreas had answered, as long as the thermal package is 
"good" enough, the leakage-temperature loop will finally 
converge. If we plot Power on x-axis and Temperature on 
y-axis, the thermal package can be approximately  modeled 
as a straight line through the origin, and the slope of 
that line the heat dissipation capability of the package. 
The leakage-temperature relationship can be modeled as an 
exponential curve. As long as the leakage-temperature curve 
is below the thermal package line, there won't be thermal 
runaway.

>
>>
>> 2/ the Hostpot tool requires a power trace to obtain the
>> temperature map. But again, how accurate that
>> methodology is? If the power of a block gets affected by
>> the temperature of surrounding blocks, the power W of
>> block B at time T offered in the trace may actually be X%
>> optimistic if that block is in the middle of a hotspot
>> at time T, or Y % pessimistic if it is in a cool part of
>> the die. So, is the methodology of feeding a power trace
>> correct, or should the temperature info somehow fed back
>> to the power trace?
>>
>
> Within a certain range you can neglect the effect of the
> increased temperature on the power consumption.  However,
> if you want to be accurate, you have need to have Hotspot
> 'in the loop', that is you must estimate power and
> temperature value simultaneously (as we do in our
> simulations).

In an upcoming release, HotSpot will integrate the 
leakage-temperature loop, allowing users to choose between 
their favorite leakage models or an embedded leakage model 
in HotSpot. Before that happens, what Andreas said is the 
best solution if you want leakage to be included.

Thanks,
-Wei





From mohamed_elsawaf at yahoo.com  Wed Apr 23 23:50:33 2008
From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf)
Date: Wed, 23 Apr 2008 23:50:33 -0700 (PDT)
Subject: [Hotspot] inordinate floorplan size
In-Reply-To: <Pine.SOC.4.64.0802041400400.6854@cobra.cs.Virginia.EDU>
Message-ID: <342239.1757.qm@web53704.mail.re2.yahoo.com>

Hi

i am trying to load Floorplan close to IBM Power PC
MCM chip into hotspot 4
i got the following error: inordinate floorplan size!
what does it mean this error message "inordinate
floorplan size!"?
do we have limitation on floor plan size?

Note: 

I made reverse engineering of IBM Power PC MCM chip
floor plan and replot it using Matlab
mcm4_R2.flp is the original floor plan containing 254
blocks
mcm4_R3.flp is the converted floor plan containing 750
blocks

please advise

Thank you
Mohamed Elsawaf
-------------------------------------------------------

[root at hs HotSpot-4.0]# ./hotspot -c
/GSDC/hotspot.config -f /GSDC/mcm4_R2.flp -p
/GSDC/gcc.ptrace -o /GSDC/gcc.ttrace_out -steady_file
/GSDC/gcc.steady > /GSDC/temperature
error: inordinate floorplan size!
[root at hs HotSpot-4.0]# echo $?
1
[root at hs HotSpot-4.0]# ./hsconvert-flp.pl
/GSDC/mcm4_R2.flp > /GSDC/mcm4_R3.flp
[root at hs HotSpot-4.0]# echo $?
0
[root at hs HotSpot-4.0]# ./hotspot -c
/GSDC/hotspot.config -f /GSDC/mcm4_R3.flp -p
/GSDC/gcc.ptrace -o /GSDC/gcc.ttrace_out -steady_file
/GSDC/gcc.steady > /GSDC/temperature
error: inordinate floorplan size!
[root at hs HotSpot-4.0]# echo $?
1


      ____________________________________________________________________________________
Be a better friend, newshound, and 
know-it-all with Yahoo! Mobile.  Try it now.  http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ

From ks4kk at cs.virginia.edu  Thu Apr 24 00:27:49 2008
From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan)
Date: Thu, 24 Apr 2008 03:27:49 -0400 (EDT)
Subject: [Hotspot] inordinate floorplan size
In-Reply-To: <342239.1757.qm@web53704.mail.re2.yahoo.com>
References: <342239.1757.qm@web53704.mail.re2.yahoo.com>
Message-ID: <Pine.SOC.4.64.0804240323250.17215@cobra.cs.Virginia.EDU>

Hi Mohamed,

The floorplan dimensions might be greater than that of the spreader and/or 
sink. Please check to see if the width and height of the chip are each 
greater than those of the spreader and sink.

Hope this helps.
-karthik

On Wed, 23 Apr 2008, Mohamed Elsawaf wrote:

> Hi
>
> i am trying to load Floorplan close to IBM Power PC
> MCM chip into hotspot 4
> i got the following error: inordinate floorplan size!
> what does it mean this error message "inordinate
> floorplan size!"?
> do we have limitation on floor plan size?
>
> Note:
>
> I made reverse engineering of IBM Power PC MCM chip
> floor plan and replot it using Matlab
> mcm4_R2.flp is the original floor plan containing 254
> blocks
> mcm4_R3.flp is the converted floor plan containing 750
> blocks
>
> please advise
>
> Thank you
> Mohamed Elsawaf
> -------------------------------------------------------
>
> [root at hs HotSpot-4.0]# ./hotspot -c
> /GSDC/hotspot.config -f /GSDC/mcm4_R2.flp -p
> /GSDC/gcc.ptrace -o /GSDC/gcc.ttrace_out -steady_file
> /GSDC/gcc.steady > /GSDC/temperature
> error: inordinate floorplan size!
> [root at hs HotSpot-4.0]# echo $?
> 1
> [root at hs HotSpot-4.0]# ./hsconvert-flp.pl
> /GSDC/mcm4_R2.flp > /GSDC/mcm4_R3.flp
> [root at hs HotSpot-4.0]# echo $?
> 0
> [root at hs HotSpot-4.0]# ./hotspot -c
> /GSDC/hotspot.config -f /GSDC/mcm4_R3.flp -p
> /GSDC/gcc.ptrace -o /GSDC/gcc.ttrace_out -steady_file
> /GSDC/gcc.steady > /GSDC/temperature
> error: inordinate floorplan size!
> [root at hs HotSpot-4.0]# echo $?
> 1
>
>
>      ____________________________________________________________________________________
> Be a better friend, newshound, and
> know-it-all with Yahoo! Mobile.  Try it now.  http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
> _______________________________________________
> HotSpot mailing list
> HotSpot at mail.cs.virginia.edu
> http://www.cs.virginia.edu/mailman/listinfo/hotspot
>

From ks4kk at cs.virginia.edu  Thu Apr 24 00:31:04 2008
From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan)
Date: Thu, 24 Apr 2008 03:31:04 -0400 (EDT)
Subject: [Hotspot] inordinate floorplan size
In-Reply-To: <Pine.SOC.4.64.0804240323250.17215@cobra.cs.Virginia.EDU>
References: <342239.1757.qm@web53704.mail.re2.yahoo.com>
	<Pine.SOC.4.64.0804240323250.17215@cobra.cs.Virginia.EDU>
Message-ID: <Pine.SOC.4.64.0804240329530.17215@cobra.cs.Virginia.EDU>

Oops - sorry about the typo. Meant to say "Please check to see if the 
width and height of the chip are each *lesser* than those of the spreader 
and sink"

Thanks
-karthik

On Thu, 24 Apr 2008, Karthik Sankaranarayanan wrote:

> Hi Mohamed,
>
> The floorplan dimensions might be greater than that of the spreader and/or
> sink. Please check to see if the width and height of the chip are each
> greater than those of the spreader and sink.
>
> Hope this helps.
> -karthik
>
> On Wed, 23 Apr 2008, Mohamed Elsawaf wrote:
>
>> Hi
>>
>> i am trying to load Floorplan close to IBM Power PC
>> MCM chip into hotspot 4
>> i got the following error: inordinate floorplan size!
>> what does it mean this error message "inordinate
>> floorplan size!"?
>> do we have limitation on floor plan size?
>>
>> Note:
>>
>> I made reverse engineering of IBM Power PC MCM chip
>> floor plan and replot it using Matlab
>> mcm4_R2.flp is the original floor plan containing 254
>> blocks
>> mcm4_R3.flp is the converted floor plan containing 750
>> blocks
>>
>> please advise
>>
>> Thank you
>> Mohamed Elsawaf
>> -------------------------------------------------------
>>
>> [root at hs HotSpot-4.0]# ./hotspot -c
>> /GSDC/hotspot.config -f /GSDC/mcm4_R2.flp -p
>> /GSDC/gcc.ptrace -o /GSDC/gcc.ttrace_out -steady_file
>> /GSDC/gcc.steady > /GSDC/temperature
>> error: inordinate floorplan size!
>> [root at hs HotSpot-4.0]# echo $?
>> 1
>> [root at hs HotSpot-4.0]# ./hsconvert-flp.pl
>> /GSDC/mcm4_R2.flp > /GSDC/mcm4_R3.flp
>> [root at hs HotSpot-4.0]# echo $?
>> 0
>> [root at hs HotSpot-4.0]# ./hotspot -c
>> /GSDC/hotspot.config -f /GSDC/mcm4_R3.flp -p
>> /GSDC/gcc.ptrace -o /GSDC/gcc.ttrace_out -steady_file
>> /GSDC/gcc.steady > /GSDC/temperature
>> error: inordinate floorplan size!
>> [root at hs HotSpot-4.0]# echo $?
>> 1
>>
>>
>>      ____________________________________________________________________________________
>> Be a better friend, newshound, and
>> know-it-all with Yahoo! Mobile.  Try it now.  http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
>> _______________________________________________
>> 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
>