From mohamed_elsawaf at yahoo.com  Tue Jan 13 23:41:26 2009
From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf)
Date: Tue, 13 Jan 2009 23:41:26 -0800 (PST)
Subject: [Hotspot] initial temperature file during Transient analysis
Message-ID: <961195.83947.qm@web53705.mail.re2.yahoo.com>

Hi

I am calling Hotspot 4.1 simulator as sub-rooting within PI control loop for DVFS.?
i need to know how can i specify initial temperature file when calling Hotspot 4.1 for Transient analysis in both mode Block and Grid.
What will be the format of this initial temperature file??
What changes are required in hotspot.config?

as i am ruining Hotspot 4.1 with gcc.ptrace files have 200,000 lines which is taking long time to get gccttrace file

/HS/HotSpot-4.1/hotspot -c ./HS_configuration/hotspot.config -f ./HS_configuration/MSC45nm_R10.flp -p ./Simulation_Results_Transient/gccptrace.2 -o ./Simulation_Results_Transient/gcc.ttrace.2 -steady_file ./Simulation_Results_Transient/gcc.steady.2 > ./Simulation_Results_Transient/Estimated_SS_Temperature.2

Thank youM Elsawaf




      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20090113/04503c98/attachment.html 

From wh6p at cms.mail.virginia.edu  Wed Jan 14 06:31:29 2009
From: wh6p at cms.mail.virginia.edu (Wei Huang)
Date: Wed, 14 Jan 2009 09:31:29 -0500
Subject: [Hotspot] initial temperature file during Transient analysis
In-Reply-To: <961195.83947.qm@web53705.mail.re2.yahoo.com>
References: <961195.83947.qm@web53705.mail.re2.yahoo.com>
Message-ID: <C4630380A38587153CCBB5E2@mstu1>

Hi, you can have initial temperatures in a file with the 
same format as the sample .steady file, and then use the 
-init_file flag. This is already mentioned in detail in the 
README files.

-Wei

--On Tuesday, January 13, 2009 11:41 PM -0800 Mohamed 
Elsawaf <mohamed_elsawaf at yahoo.com> wrote:

> Hi
>
> I am calling Hotspot 4.1 simulator as sub-rooting within
> PI control loop for DVFS.?
> i need to know how can i specify initial temperature file
> when calling Hotspot 4.1 for Transient analysis in both
> mode Block and Grid.
> What will be the format of this initial temperature file??
> What changes are required in hotspot.config?
>
> as i am ruining Hotspot 4.1 with gcc.ptrace files have
> 200,000 lines which is taking long time to get gccttrace
> file
>
> /HS/HotSpot-4.1/hotspot -c
> ./HS_configuration/hotspot.config -f
> ./HS_configuration/MSC45nm_R10.flp -p
> ./Simulation_Results_Transient/gccptrace.2 -o
> ./Simulation_Results_Transient/gcc.ttrace.2 -steady_file
> ./Simulation_Results_Transient/gcc.steady.2 >
> ./Simulation_Results_Transient/Estimated_SS_Temperature.2
>
>
> Thank you
>
> M Elsawaf
>
>
>






From mohamed_elsawaf at yahoo.com  Wed Jan 14 09:46:47 2009
From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf)
Date: Wed, 14 Jan 2009 09:46:47 -0800 (PST)
Subject: [Hotspot] initial temperature file during Transient analysis
In-Reply-To: <C4630380A38587153CCBB5E2@mstu1>
Message-ID: <515686.32670.qm@web53702.mail.re2.yahoo.com>

Hi Wei

Thank you for your fast response,  steady file at cycles n+1 requires the temperatures of blocks, iface, hsp, hsink and inode;  i can extract blocks temperatures only from gcc.ttrace at cycles n

so how can i these remaining values at calling interval n?

Yours, M Elsawaf


--- On Wed, 1/14/09, Wei Huang <wh6p at cms.mail.virginia.edu> wrote:
From: Wei Huang <wh6p at cms.mail.virginia.edu>
Subject: Re: [Hotspot] initial temperature file during Transient analysis
To: mohamed_elsawaf at yahoo.com, hotspot at mail.cs.virginia.edu
Date: Wednesday, January 14, 2009, 7:31 AM

Hi, you can have initial temperatures in a file with the 
same format as the sample .steady file, and then use the 
-init_file flag. This is already mentioned in detail in the 
README files.

-Wei

--On Tuesday, January 13, 2009 11:41 PM -0800 Mohamed 
Elsawaf <mohamed_elsawaf at yahoo.com> wrote:

> Hi
>
> I am calling Hotspot 4.1 simulator as sub-rooting within
> PI control loop for DVFS.?
> i need to know how can i specify initial temperature file
> when calling Hotspot 4.1 for Transient analysis in both
> mode Block and Grid.
> What will be the format of this initial temperature file??
> What changes are required in hotspot.config?
>
> as i am ruining Hotspot 4.1 with gcc.ptrace files have
> 200,000 lines which is taking long time to get gccttrace
> file
>
> /HS/HotSpot-4.1/hotspot -c
> ./HS_configuration/hotspot.config -f
> ./HS_configuration/MSC45nm_R10.flp -p
> ./Simulation_Results_Transient/gccptrace.2 -o
> ./Simulation_Results_Transient/gcc.ttrace.2 -steady_file
> ./Simulation_Results_Transient/gcc.steady.2 >
> ./Simulation_Results_Transient/Estimated_SS_Temperature.2
>
>
> Thank you
>
> M Elsawaf
>
>
>







      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20090114/eb702485/attachment.html 

From wh6p at cms.mail.virginia.edu  Wed Jan 14 11:09:02 2009
From: wh6p at cms.mail.virginia.edu (Wei Huang)
Date: Wed, 14 Jan 2009 14:09:02 -0500
Subject: [Hotspot] initial temperature file during Transient analysis
In-Reply-To: <515686.32670.qm@web53702.mail.re2.yahoo.com>
References: <515686.32670.qm@web53702.mail.re2.yahoo.com>
Message-ID: <8C8A778BB69C9FCAEFC624F7@mstu1>

Hi Mohamed,

You just need to specify the initial temperatures once in 
your transient simulations. One way to obtain the initial 
temperatures without warming up everything from the ambient 
temperature is to run a quick steady-state simulation with 
average powers for each floorplan unit, and use the 
resulting steady file as initial temperatures. Then you can 
run some more instructions to "warm up" the architectural 
states (see our ISCA'03 paper for more details). I believe 
the *.steady file already has temperatures for all the 
nodes.

Hope this helps.
-Wei

--On Wednesday, January 14, 2009 9:46 AM -0800 Mohamed 
Elsawaf <mohamed_elsawaf at yahoo.com> wrote:

>
> Hi Wei
>
> Thank you for your fast response, steady file at cycles
> n+1 requires the temperatures of blocks, iface, hsp,
> hsink and inode; i can extract blocks temperatures only
> from gcc.ttrace at cycles n
>
> so how can i these remaining values at calling interval n?
>
> Yours, M Elsawaf
>
>
> --- On Wed, 1/14/09, Wei Huang
> <wh6p at cms.mail.virginia.edu> wrote:
>
> From: Wei Huang <wh6p at cms.mail.virginia.edu>
> Subject: Re: [Hotspot] initial temperature file during
> Transient analysis
> To: mohamed_elsawaf at yahoo.com,
> hotspot at mail.cs.virginia.edu
> Date: Wednesday, January 14, 2009, 7:31 AM
>
>
> Hi, you can have initial temperatures in a file with the
> same format as the sample .steady file, and then use the
> -init_file flag. This is already mentioned in detail in
> the
> README files.
>
> -Wei
>
> --On Tuesday, January 13, 2009 11:41 PM -0800 Mohamed
> Elsawaf <mohamed_elsawaf at yahoo.com> wrote:
>
>> Hi
>>
>> I am calling Hotspot 4.1 simulator as sub-rooting within
>> PI control loop for DVFS.?
>> i need to know how can i specify initial temperature file
>> when calling Hotspot 4.1 for Transient analysis in both
>> mode Block and Grid.
>> What will be the format of this initial temperature
>> file?? What changes are required in hotspot.config?
>>
>> as i am ruining Hotspot 4.1 with gcc.ptrace files have
>> 200,000 lines which is taking long time to get gccttrace
>> file
>>
>> /HS/HotSpot-4.1/hotspot -c
>> ./HS_configuration/hotspot.config -f
>> ./HS_configuration/MSC45nm_R10.flp -p
>> ./Simulation_Results_Transient/gccptrace.2 -o
>> ./Simulation_Results_Transient/gcc.ttrace.2 -steady_file
>> ./Simulation_Results_Transient/gcc.steady.2 >
>> ./Simulation_Results_Transient/Estimated_SS_Temperature.2
>>
>>
>> Thank you
>>
>> M Elsawaf
>>
>>
>>
>
>
>
>
>
>
>
>






From skadron at cs.virginia.edu  Wed Jan 14 12:26:24 2009
From: skadron at cs.virginia.edu (Kevin Skadron)
Date: Wed, 14 Jan 2009 15:26:24 -0500
Subject: [Hotspot] initial temperature file during Transient analysis
In-Reply-To: <8C8A778BB69C9FCAEFC624F7@mstu1>
References: <515686.32670.qm@web53702.mail.re2.yahoo.com>
	<8C8A778BB69C9FCAEFC624F7@mstu1>
Message-ID: <496E4A70.1000603@cs.virginia.edu>

PS If you're interested in this topic of "warming up" the simulation 
state in the presence of dynamic thermal management, there was also an 
extensive treatment by Adve's group at Illinois.
/K

Wei Huang wrote:
> Hi Mohamed,
> 
> You just need to specify the initial temperatures once in 
> your transient simulations. One way to obtain the initial 
> temperatures without warming up everything from the ambient 
> temperature is to run a quick steady-state simulation with 
> average powers for each floorplan unit, and use the 
> resulting steady file as initial temperatures. Then you can 
> run some more instructions to "warm up" the architectural 
> states (see our ISCA'03 paper for more details). I believe 
> the *.steady file already has temperatures for all the 
> nodes.
> 
> Hope this helps.
> -Wei
> 
> --On Wednesday, January 14, 2009 9:46 AM -0800 Mohamed 
> Elsawaf <mohamed_elsawaf at yahoo.com> wrote:
> 
>> Hi Wei
>>
>> Thank you for your fast response, steady file at cycles
>> n+1 requires the temperatures of blocks, iface, hsp,
>> hsink and inode; i can extract blocks temperatures only
>> from gcc.ttrace at cycles n
>>
>> so how can i these remaining values at calling interval n?
>>
>> Yours, M Elsawaf
>>
>>
>> --- On Wed, 1/14/09, Wei Huang
>> <wh6p at cms.mail.virginia.edu> wrote:
>>
>> From: Wei Huang <wh6p at cms.mail.virginia.edu>
>> Subject: Re: [Hotspot] initial temperature file during
>> Transient analysis
>> To: mohamed_elsawaf at yahoo.com,
>> hotspot at mail.cs.virginia.edu
>> Date: Wednesday, January 14, 2009, 7:31 AM
>>
>>
>> Hi, you can have initial temperatures in a file with the
>> same format as the sample .steady file, and then use the
>> -init_file flag. This is already mentioned in detail in
>> the
>> README files.
>>
>> -Wei
>>
>> --On Tuesday, January 13, 2009 11:41 PM -0800 Mohamed
>> Elsawaf <mohamed_elsawaf at yahoo.com> wrote:
>>
>>> Hi
>>>
>>> I am calling Hotspot 4.1 simulator as sub-rooting within
>>> PI control loop for DVFS. 
>>> i need to know how can i specify initial temperature file
>>> when calling Hotspot 4.1 for Transient analysis in both
>>> mode Block and Grid.
>>> What will be the format of this initial temperature
>>> file?  What changes are required in hotspot.config?
>>>
>>> as i am ruining Hotspot 4.1 with gcc.ptrace files have
>>> 200,000 lines which is taking long time to get gccttrace
>>> file
>>>
>>> /HS/HotSpot-4.1/hotspot -c
>>> ./HS_configuration/hotspot.config -f
>>> ./HS_configuration/MSC45nm_R10.flp -p
>>> ./Simulation_Results_Transient/gccptrace.2 -o
>>> ./Simulation_Results_Transient/gcc.ttrace.2 -steady_file
>>> ./Simulation_Results_Transient/gcc.steady.2 >
>>> ./Simulation_Results_Transient/Estimated_SS_Temperature.2
>>>
>>>
>>> Thank you
>>>
>>> M Elsawaf
>>>
>>>
>>>
>>
>>
>>
>>
>>
>>
>>
> 
> 
> 
> 
> 
> _______________________________________________
> HotSpot mailing list
> HotSpot at mail.cs.virginia.edu
> http://www.cs.virginia.edu/mailman/listinfo/hotspot

From ks4kk at cs.virginia.edu  Wed Jan 14 14:17:40 2009
From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan)
Date: Wed, 14 Jan 2009 17:17:40 -0500 (EST)
Subject: [Hotspot] initial temperature file during Transient analysis
In-Reply-To: <515686.32670.qm@web53702.mail.re2.yahoo.com>
References: <515686.32670.qm@web53702.mail.re2.yahoo.com>
Message-ID: <Pine.SOC.4.64.0901141711590.23914@cobra.cs.Virginia.EDU>


Hi Mohamed,

You would have to use the -steady_file command line option to generate the 
steady state temperature file (which is different from the temperature 
trace file). If this explanation is unclear, please take a look at the 
HotSpot HOWTO (section B-1 command lines a) and b)) at 
http://lava.cs.virginia.edu/HotSpot/HotSpot-HOWTO.htm - it details the 
command line to be used to generate a steady state temperature file and to 
re-use it as the initial temperature for the next run.

Hope this helps.

Thanks
-karthik


On Wed, 14 Jan 2009, Mohamed Elsawaf wrote:

> Hi Wei
>
> Thank you for your fast response, steady file at cycles n+1 requires the 
> temperatures of blocks, iface, hsp, hsink and inode;  i can extract 
> blocks temperatures only from gcc.ttrace at cycles n
>
> so how can i these remaining values at calling interval n?
>
> Yours, M Elsawaf
>
>
> --- On Wed, 1/14/09, Wei Huang <wh6p at cms.mail.virginia.edu> wrote:
> From: Wei Huang <wh6p at cms.mail.virginia.edu>
> Subject: Re: [Hotspot] initial temperature file during Transient analysis
> To: mohamed_elsawaf at yahoo.com, hotspot at mail.cs.virginia.edu
> Date: Wednesday, January 14, 2009, 7:31 AM
>
> Hi, you can have initial temperatures in a file with the
> same format as the sample .steady file, and then use the
> -init_file flag. This is already mentioned in detail in the
> README files.
>
> -Wei
>
> --On Tuesday, January 13, 2009 11:41 PM -0800 Mohamed
> Elsawaf <mohamed_elsawaf at yahoo.com> wrote:
>
>> Hi
>>
>> I am calling Hotspot 4.1 simulator as sub-rooting within
>> PI control loop for DVFS.?
>> i need to know how can i specify initial temperature file
>> when calling Hotspot 4.1 for Transient analysis in both
>> mode Block and Grid.
>> What will be the format of this initial temperature file??
>> What changes are required in hotspot.config?
>>
>> as i am ruining Hotspot 4.1 with gcc.ptrace files have
>> 200,000 lines which is taking long time to get gccttrace
>> file
>>
>> /HS/HotSpot-4.1/hotspot -c
>> ./HS_configuration/hotspot.config -f
>> ./HS_configuration/MSC45nm_R10.flp -p
>> ./Simulation_Results_Transient/gccptrace.2 -o
>> ./Simulation_Results_Transient/gcc.ttrace.2 -steady_file
>> ./Simulation_Results_Transient/gcc.steady.2 >
>> ./Simulation_Results_Transient/Estimated_SS_Temperature.2
>>
>>
>> Thank you
>>
>> M Elsawaf
>>
>>
>>
>
>
>
>
>

From mohamed_elsawaf at yahoo.com  Fri Jan 16 02:00:22 2009
From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf)
Date: Fri, 16 Jan 2009 02:00:22 -0800 (PST)
Subject: [Hotspot] Grid cell end index out of bounds
In-Reply-To: <Pine.SOC.4.64.0901141711590.23914@cobra.cs.Virginia.EDU>
Message-ID: <75166.68828.qm@web53708.mail.re2.yahoo.com>

HiI have problem running grid mode, the simulation is working OK with block mode.

I got below warning at grid mode:?

warning: overlap of functional blocks?
warning: overlap of functional blocks?
warning: overlap of functional blocks?
warning: overlap of functional blocks?
warning: overlap of functional blocks?
warning: overlap of functional blocks?
warning: overlap of functional blocks?

then below error


error: grid cell end index out of bounds!


I managed to avoid warning message by decreasing cols to 16 [-grid_cols 16] while keeping row 1024 [-grid_row 1024]
but still getting same error

error: grid cell end index out of bounds!
Note: number of warning increase by increasing number of cols
please advise

Thank you
M. Elsawaf






      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20090116/7c8c8bb5/attachment.html 

From ks4kk at cs.virginia.edu  Fri Jan 16 07:02:59 2009
From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan)
Date: Fri, 16 Jan 2009 10:02:59 -0500 (EST)
Subject: [Hotspot] Grid cell end index out of bounds
In-Reply-To: <75166.68828.qm@web53708.mail.re2.yahoo.com>
References: <75166.68828.qm@web53708.mail.re2.yahoo.com>
Message-ID: <Pine.SOC.4.64.0901161000070.28725@cobra.cs.Virginia.EDU>


Hi Mohamed,

It appears that your floorplan might have overlapping functional blocks. 
You can use tofig.pl tool to plot your floorplan and correct any errors 
that might be in the floorplan.

-karthik



On Fri, 16 Jan 2009, Mohamed Elsawaf wrote:

> HiI have problem running grid mode, the simulation is working OK with block mode.
>
> I got below warning at grid mode:?
>
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
>
> then below error
>
>
> error: grid cell end index out of bounds!
>
>
> I managed to avoid warning message by decreasing cols to 16 [-grid_cols 16] while keeping row 1024 [-grid_row 1024]
> but still getting same error
>
> error: grid cell end index out of bounds!
> Note: number of warning increase by increasing number of cols
> please advise
>
> Thank you
> M. Elsawaf
>
>
>
>

From mohamed_elsawaf at yahoo.com  Sat Jan 17 00:31:19 2009
From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf)
Date: Sat, 17 Jan 2009 00:31:19 -0800 (PST)
Subject: [Hotspot] Grid cell end index out of bounds
In-Reply-To: <Pine.SOC.4.64.0901161000070.28725@cobra.cs.Virginia.EDU>
Message-ID: <212409.82657.qm@web53709.mail.re2.yahoo.com>

Hi Karthik

it is really strange that grid mode see that i have overlapping functional blocks while block mode working ok!
How grid mode is expecting functional blocks shapes? As the current functional block are rectangular not squares or multiple of square
Do i need to plot floor-plan as grid cells?  ?

I used tofig.pl tool but i don't know how to use its output to plot my floor-plan, can you please send me more details?
waiting for your feedback.
Thank you
M. Elsawaf

--- On Fri, 1/16/09, Karthik Sankaranarayanan <ks4kk at cs.virginia.edu> wrote:
From: Karthik Sankaranarayanan <ks4kk at cs.virginia.edu>
Subject: Re: [Hotspot] Grid cell end index out of bounds
To: "Mohamed Elsawaf" <mohamed_elsawaf at yahoo.com>
Cc: hotspot at mail.cs.virginia.edu
Date: Friday, January 16, 2009, 8:02 AM

Hi Mohamed,

It appears that your floorplan might have overlapping functional blocks. 
You can use tofig.pl tool to plot your floorplan and correct any errors 
that might be in the floorplan.

-karthik



On Fri, 16 Jan 2009, Mohamed Elsawaf wrote:

> HiI have problem running grid mode, the simulation is working OK with
block mode.
>
> I got below warning at grid mode:?
>
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
> warning: overlap of functional blocks?
>
> then below error
>
>
> error: grid cell end index out of bounds!
>
>
> I managed to avoid warning message by decreasing cols to 16 [-grid_cols
16] while keeping row 1024 [-grid_row 1024]
> but still getting same error
>
> error: grid cell end index out of bounds!
> Note: number of warning increase by increasing number of cols
> please advise
>
> Thank you
> M. Elsawaf
>
>
>
>


      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20090117/baa3ff9e/attachment.html 

From ks4kk at cs.virginia.edu  Sat Jan 17 08:15:26 2009
From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan)
Date: Sat, 17 Jan 2009 11:15:26 -0500 (EST)
Subject: [Hotspot] Grid cell end index out of bounds
In-Reply-To: <212409.82657.qm@web53709.mail.re2.yahoo.com>
References: <212409.82657.qm@web53709.mail.re2.yahoo.com>
Message-ID: <Pine.SOC.4.64.0901171101130.5053@cobra.cs.Virginia.EDU>


Hi Mohamed,

No special floorplan is needed for the grid model - just the normal 
rectangular blocks would do. Just make sure there are no overlaps between 
the individual rectangles and that the envelop of all the rectangles is 
also a rectangle.

The block model assumes that the floorplan is OK (in the sense described 
above) and proceeds with its calculations while the grid model checks the 
floorplan before proceeding (when mapping it onto a grid). So, if the 
floorplan was erroneous, the block model would not have caught it 
(actually, a non-overlapping floorplan is a precondition for HotSpot).

Please read the HOWTO file 
(http://lava.cs.virginia.edu/HotSpot/HotSpot-HOWTO.htm) - it mentions the 
command line to explain the use of tofig.pl. The output of tofig.pl is in 
the .fig format which can be viewed using the XFig software. It can also 
be converted into other formats.

Hope this helps
-karthik

On Sat, 17 Jan 2009, Mohamed Elsawaf wrote:

> Hi Karthik
>
> it is really strange that grid mode see that i have overlapping 
> functional blocks while block mode working ok! How grid mode is 
> expecting functional blocks shapes? As the current functional block are 
> rectangular not squares or multiple of square Do i need to plot 
> floor-plan as grid cells?  ?
>
> I used tofig.pl tool but i don't know how to use its output to plot my 
> floor-plan, can you please send me more details? waiting for your 
> feedback. Thank you M. Elsawaf
>
> --- On Fri, 1/16/09, Karthik Sankaranarayanan <ks4kk at cs.virginia.edu> wrote:
> From: Karthik Sankaranarayanan <ks4kk at cs.virginia.edu>
> Subject: Re: [Hotspot] Grid cell end index out of bounds
> To: "Mohamed Elsawaf" <mohamed_elsawaf at yahoo.com>
> Cc: hotspot at mail.cs.virginia.edu
> Date: Friday, January 16, 2009, 8:02 AM
>
> Hi Mohamed,
>
> It appears that your floorplan might have overlapping functional blocks.
> You can use tofig.pl tool to plot your floorplan and correct any errors
> that might be in the floorplan.
>
> -karthik
>
>
>
> On Fri, 16 Jan 2009, Mohamed Elsawaf wrote:
>
>> HiI have problem running grid mode, the simulation is working OK with
> block mode.
>>
>> I got below warning at grid mode:?
>>
>> warning: overlap of functional blocks?
>> warning: overlap of functional blocks?
>> warning: overlap of functional blocks?
>> warning: overlap of functional blocks?
>> warning: overlap of functional blocks?
>> warning: overlap of functional blocks?
>> warning: overlap of functional blocks?
>>
>> then below error
>>
>>
>> error: grid cell end index out of bounds!
>>
>>
>> I managed to avoid warning message by decreasing cols to 16 [-grid_cols
> 16] while keeping row 1024 [-grid_row 1024]
>> but still getting same error
>>
>> error: grid cell end index out of bounds!
>> Note: number of warning increase by increasing number of cols
>> please advise
>>
>> Thank you
>> M. Elsawaf
>>
>>
>>
>>

From ks4kk at cs.virginia.edu  Sat Jan 17 08:23:20 2009
From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan)
Date: Sat, 17 Jan 2009 11:23:20 -0500 (EST)
Subject: [Hotspot] Grid cell end index out of bounds
In-Reply-To: <Pine.SOC.4.64.0901171101130.5053@cobra.cs.Virginia.EDU>
References: <212409.82657.qm@web53709.mail.re2.yahoo.com>
	<Pine.SOC.4.64.0901171101130.5053@cobra.cs.Virginia.EDU>
Message-ID: <Pine.SOC.4.64.0901171120450.5053@cobra.cs.Virginia.EDU>


In the description about a correct floorplan, I forgot to mention the 
condition that the area of the individual rectangles should add up to the 
area of the envelop rectangle (the die). This basically means that there 
should not be any gaps or overlaps in your floorplan.

Thanks,
-karthik


On Sat, 17 Jan 2009, Karthik Sankaranarayanan wrote:

>
> Hi Mohamed,
>
> No special floorplan is needed for the grid model - just the normal 
> rectangular blocks would do. Just make sure there are no overlaps between the 
> individual rectangles and that the envelop of all the rectangles is also a 
> rectangle.
>
> The block model assumes that the floorplan is OK (in the sense described 
> above) and proceeds with its calculations while the grid model checks the 
> floorplan before proceeding (when mapping it onto a grid). So, if the 
> floorplan was erroneous, the block model would not have caught it (actually, 
> a non-overlapping floorplan is a precondition for HotSpot).
>
> Please read the HOWTO file 
> (http://lava.cs.virginia.edu/HotSpot/HotSpot-HOWTO.htm) - it mentions the 
> command line to explain the use of tofig.pl. The output of tofig.pl is in the 
> .fig format which can be viewed using the XFig software. It can also be 
> converted into other formats.
>
> Hope this helps
> -karthik
>
> On Sat, 17 Jan 2009, Mohamed Elsawaf wrote:
>
>> Hi Karthik
>> 
>> it is really strange that grid mode see that i have overlapping functional 
>> blocks while block mode working ok! How grid mode is expecting functional 
>> blocks shapes? As the current functional block are rectangular not squares 
>> or multiple of square Do i need to plot floor-plan as grid cells?  ?
>> 
>> I used tofig.pl tool but i don't know how to use its output to plot my 
>> floor-plan, can you please send me more details? waiting for your feedback. 
>> Thank you M. Elsawaf
>> 
>> --- On Fri, 1/16/09, Karthik Sankaranarayanan <ks4kk at cs.virginia.edu> 
>> wrote:
>> From: Karthik Sankaranarayanan <ks4kk at cs.virginia.edu>
>> Subject: Re: [Hotspot] Grid cell end index out of bounds
>> To: "Mohamed Elsawaf" <mohamed_elsawaf at yahoo.com>
>> Cc: hotspot at mail.cs.virginia.edu
>> Date: Friday, January 16, 2009, 8:02 AM
>> 
>> Hi Mohamed,
>> 
>> It appears that your floorplan might have overlapping functional blocks.
>> You can use tofig.pl tool to plot your floorplan and correct any errors
>> that might be in the floorplan.
>> 
>> -karthik
>> 
>> 
>> 
>> On Fri, 16 Jan 2009, Mohamed Elsawaf wrote:
>> 
>>> HiI have problem running grid mode, the simulation is working OK with
>> block mode.
>>> 
>>> I got below warning at grid mode:?
>>> 
>>> warning: overlap of functional blocks?
>>> warning: overlap of functional blocks?
>>> warning: overlap of functional blocks?
>>> warning: overlap of functional blocks?
>>> warning: overlap of functional blocks?
>>> warning: overlap of functional blocks?
>>> warning: overlap of functional blocks?
>>> 
>>> then below error
>>> 
>>> 
>>> error: grid cell end index out of bounds!
>>> 
>>> 
>>> I managed to avoid warning message by decreasing cols to 16 [-grid_cols
>> 16] while keeping row 1024 [-grid_row 1024]
>>> but still getting same error
>>> 
>>> error: grid cell end index out of bounds!
>>> Note: number of warning increase by increasing number of cols
>>> please advise
>>> 
>>> Thank you
>>> M. Elsawaf
>>> 
>>> 
>>> 
>

From m.dehyadegari at ece.ut.ac.ir  Sun Jan 18 00:40:34 2009
From: m.dehyadegari at ece.ut.ac.ir (masoud Deh-Yadegari)
Date: Sun, 18 Jan 2009 12:10:34 +0330
Subject: [Hotspot] HotSpot 4.1 -- Power Trace
Message-ID: <21a6d640901180040n20ee7f92o49c44badbb8ce35f@mail.gmail.com>

Dear all,
I am using HotSpot4.1 but I have a problem about how to generate power trace
file for hotspot.
I am evaluating a synthesis algorithm for temperature meseaurment but I can
not how to generate power trace.
I am thankful if you give me information about it.
I should mention I can not use such tool like watch because it is for
processors.

Regards,
Masoud Deh-Yadehgari
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20090118/58fe79d7/attachment.html 

From skadron at cs.virginia.edu  Sun Jan 18 05:15:05 2009
From: skadron at cs.virginia.edu (Kevin Skadron)
Date: Sun, 18 Jan 2009 08:15:05 -0500
Subject: [Hotspot] HotSpot 4.1 -- Power Trace
In-Reply-To: <21a6d640901180040n20ee7f92o49c44badbb8ce35f@mail.gmail.com>
References: <21a6d640901180040n20ee7f92o49c44badbb8ce35f@mail.gmail.com>
Message-ID: <49732B59.6070308@cs.virginia.edu>

Are you looking to study transient temperature behavior, or only steady 
-state?
/K

masoud Deh-Yadegari wrote:
> Dear all,
> I am using HotSpot4.1 but I have a problem about how to generate power 
> trace file for hotspot.
> I am evaluating a synthesis algorithm for temperature meseaurment but I 
> can not how to generate power trace.
> I am thankful if you give me information about it.
> I should mention I can not use such tool like watch because it is for 
> processors.
>  
> Regards,
> Masoud Deh-Yadehgari
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> HotSpot mailing list
> HotSpot at mail.cs.virginia.edu
> http://www.cs.virginia.edu/mailman/listinfo/hotspot

From yyangpan at gmail.com  Mon Jan 19 12:53:35 2009
From: yyangpan at gmail.com (Pan, Yangyang)
Date: Mon, 19 Jan 2009 15:53:35 -0500
Subject: [Hotspot] A question about the BLAS from a new learner
Message-ID: <f524a5d40901191253m6aa4e133j55f97898028c6431@mail.gmail.com>

Hi everyone,

Good day. I am a new user of Hotspot. In the instruction 5  of Installation,
it tells the instruction about BLAS. But I do not know how to use the BLAS
in the Hotspot. Would you please provide some help about how to use the BLAS
and how to set the path about BLAS?  I really appreciate you help.

All the best,
Yangyang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20090119/4b28ece1/attachment.html 

From ks4kk at cs.virginia.edu  Mon Jan 19 17:51:08 2009
From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan)
Date: Mon, 19 Jan 2009 20:51:08 -0500 (EST)
Subject: [Hotspot] A question about the BLAS from a new learner
In-Reply-To: <f524a5d40901191253m6aa4e133j55f97898028c6431@mail.gmail.com>
References: <f524a5d40901191253m6aa4e133j55f97898028c6431@mail.gmail.com>
Message-ID: <Pine.SOC.4.64.0901192048030.26788@cobra.cs.Virginia.EDU>


Hi Yangyang,

Please take a look at the Makefile that comes with the HotSpot 
distribution. It has different sample paths for BLAS libraries of 
different targets. Uncomment the lines corresponding to your architecture 
and compile HotSpot using this modified Makefile.

As for the code that uses BLAS interfaces, take a look at RCutil.c. Most 
of the BLAS functions are called from there.

Thanks,
-karthik

On Mon, 19 Jan 2009, Pan, Yangyang wrote:

> Hi everyone,
>
> Good day. I am a new user of Hotspot. In the instruction 5  of Installation,
> it tells the instruction about BLAS. But I do not know how to use the BLAS
> in the Hotspot. Would you please provide some help about how to use the BLAS
> and how to set the path about BLAS?  I really appreciate you help.
>
> All the best,
> Yangyang
>

From mohamed_elsawaf at yahoo.com  Wed Jan 21 09:51:56 2009
From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf)
Date: Wed, 21 Jan 2009 09:51:56 -0800 (PST)
Subject: [Hotspot] blockTemperature is increase very fast
In-Reply-To: <Pine.SOC.4.64.0901171120450.5053@cobra.cs.Virginia.EDU>
Message-ID: <988735.41501.qm@web53707.mail.re2.yahoo.com>

HiI notice that a block Temperature is increase very fast as follow:

Block area: width 0.00067526  X height 0.00045017 =  3.03981E-07 m^2
Power consumed per calling interval: 1.2842 W
Power Density: 422.4607895 W/cm^2
technology: 45 nm
initial Temperature: 25 C
Temperature at 18265 calling interval: 70 C
hotspot calling interval - 10K cycles at 3GHz [default]

which mean that real time required to increase this block Temperature by 45 degree is as follow:?

10K cycles  X 18265 calling interval X 3GHz = 0.54795 second

i think 0.54795 second is too high for increase 45 C

am i using very high power density? or my real time calculation are incorrect? or something else?

Note:?
Same Block specs at 90 nm technology as follow:

Block area: width 0.00135051 X height 0.00090034 = 1.21592E-06 m^2
Power consumed per calling interval: 0.6421 W
Power Density: 52.80759869 W/cm^2

Please advise
thank you
M Elsawaf




      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20090121/8ddee556/attachment.html 

From wh6p at cms.mail.virginia.edu  Wed Jan 21 10:23:37 2009
From: wh6p at cms.mail.virginia.edu (Wei Huang)
Date: Wed, 21 Jan 2009 13:23:37 -0500
Subject: [Hotspot] blockTemperature is increase very fast
In-Reply-To: <988735.41501.qm@web53707.mail.re2.yahoo.com>
References: <988735.41501.qm@web53707.mail.re2.yahoo.com>
Message-ID: <3CF269C7581BC3641BF90384@mstu1>

The power density does seem a little bit too high 
(4.22W/mm^2). How did you get the power and area of that 
block for 45nm? What is the steady-state temperature of 
that block at 45nm? I would guess at 90nm, the temperature 
increasing rate is not that fast.

Localized heating within silicon can be very fast 
(~hundreds of milliseconds). After that, it takes minutes 
to heat up to the steady-state temperature due to the large 
thermal caps of heatsink etc.

Thanks,
-Wei

--On Wednesday, January 21, 2009 9:51 AM -0800 Mohamed 
Elsawaf <mohamed_elsawaf at yahoo.com> wrote:

>
> Hi
>
>
> I notice that a block Temperature is increase very fast
> as follow:
>
> Block area: width 0.00067526 X height 0.00045017 =
> 3.03981E-07 m^2
> Power consumed per calling interval: 1.2842 W
> Power Density: 422.4607895 W/cm^2
> technology: 45 nm
> initial Temperature: 25 C
> Temperature at 18265 calling interval: 70 C
> hotspot calling interval - 10K cycles at 3GHz [default]
>
> which mean that real time required to increase this block
> Temperature by 45 degree is as follow:?
>
> 10K cycles X 18265 calling interval X 3GHz = 0.54795
> second
>
> i think 0.54795 second is too high for increase 45 C
>
> am i using very high power density? or my real time
> calculation are incorrect? or something else?
>
> Note:?
> Same Block specs at 90 nm technology as follow:
>
> Block area: width 0.00135051 X height 0.00090034 =
> 1.21592E-06 m^2
> Power consumed per calling interval: 0.6421 W
> Power Density: 52.80759869 W/cm^2
>
> Please advise
> thank you
> M Elsawaf
>
>
>






From mohamed_elsawaf at yahoo.com  Thu Jan 22 00:01:17 2009
From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf)
Date: Thu, 22 Jan 2009 00:01:17 -0800 (PST)
Subject: [Hotspot] blockTemperature is increase very fast
In-Reply-To: <3CF269C7581BC3641BF90384@mstu1>
Message-ID: <214456.65292.qm@web53702.mail.re2.yahoo.com>

Hi Wei

Thank you for your help, are my calculation regarding real time correct?
10K cycles X 18265 calling interval X 3GHz = 0.54795 seconds

please find answer of your questions below

How did you get the power and area of that block for 45nm??
M Elsawaf: as the chip doesn't exists on 45 nm
?
?for power, i was try to scale power based on 90 nm information i extracted
?assuming power consumed at 45 nm will be only  2 X power consumed at 90 nm
?i think even in reality power consumed at 45 nm will be  4 X power consumed at 90 nm
?
?for area, i assumed that scaling floor-pan from 90 nm to 45 nm as optical zoom out
?Optical Shrink_scale = 45/90 , i know that scaling 90 nm into 45 nm technology is not just optical zoom out but from thermal point of view we can neglect the difference between technology scaling and optical zoom out.
?
?
What is the steady-state temperature of that block at 45nm?
M Elsawaf:
?steady-state temperature is 184.47 C
?
I would guess at 90nm, the temperature increasing rate is not that fast.
M Elsawaf: Yes



Thank youyours,?M Elsawaf

--- On Wed, 1/21/09, Wei Huang <wh6p at cms.mail.virginia.edu> wrote:
From: Wei Huang <wh6p at cms.mail.virginia.edu>
Subject: Re: [Hotspot] blockTemperature is increase very fast
To: mohamed_elsawaf at yahoo.com, hotspot at mail.cs.virginia.edu
Date: Wednesday, January 21, 2009, 11:23 AM

The power density does seem a little bit too high 
(4.22W/mm^2). How did you get the power and area of that 
block for 45nm? What is the steady-state temperature of 
that block at 45nm? I would guess at 90nm, the temperature 
increasing rate is not that fast.

Localized heating within silicon can be very fast 
(~hundreds of milliseconds). After that, it takes minutes 
to heat up to the steady-state temperature due to the large 
thermal caps of heatsink etc.

Thanks,
-Wei

--On Wednesday, January 21, 2009 9:51 AM -0800 Mohamed 
Elsawaf <mohamed_elsawaf at yahoo.com> wrote:

>
> Hi
>
>
> I notice that a block Temperature is increase very fast
> as follow:
>
> Block area: width 0.00067526 X height 0.00045017 =
> 3.03981E-07 m^2
> Power consumed per calling interval: 1.2842 W
> Power Density: 422.4607895 W/cm^2
> technology: 45 nm
> initial Temperature: 25 C
> Temperature at 18265 calling interval: 70 C
> hotspot calling interval - 10K cycles at 3GHz [default]
>
> which mean that real time required to increase this block
> Temperature by 45 degree is as follow:?
>
> 10K cycles X 18265 calling interval X 3GHz = 0.54795
> second
>
> i think 0.54795 second is too high for increase 45 C
>
> am i using very high power density? or my real time
> calculation are incorrect? or something else?
>
> Note:?
> Same Block specs at 90 nm technology as follow:
>
> Block area: width 0.00135051 X height 0.00090034 =
> 1.21592E-06 m^2
> Power consumed per calling interval: 0.6421 W
> Power Density: 52.80759869 W/cm^2
>
> Please advise
> thank you
> M Elsawaf
>
>
>







      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20090122/b10abf52/attachment.html 

From wh6p at cms.mail.virginia.edu  Thu Jan 22 08:39:46 2009
From: wh6p at cms.mail.virginia.edu (Wei Huang)
Date: Thu, 22 Jan 2009 11:39:46 -0500
Subject: [Hotspot] blockTemperature is increase very fast
In-Reply-To: <214456.65292.qm@web53702.mail.re2.yahoo.com>
References: <214456.65292.qm@web53702.mail.re2.yahoo.com>
Message-ID: <6918B134002440EEA3BB4281@mstu1>

Hi, please see my response below....

--On Thursday, January 22, 2009 12:01 AM -0800 Mohamed 
Elsawaf <mohamed_elsawaf at yahoo.com> wrote:

>
> Hi Wei
>
> Thank you for your help, are my calculation regarding
> real time correct?
> 10K cycles X 18265 calling interval X 3GHz = 0.54795
> seconds

10K cycles * 18265 samples * 0.33*10^(-9) seconds/cycle = 
0.06 seconds. Please doublecheck.

>
> please find answer of your questions below
>
> How did you get the power and area of that block for
> 45nm??
> M Elsawaf: as the chip doesn't exists on 45 nm
> ?
> ?for power, i was try to scale power based on 90 nm
> information i extracted
> ?assuming power consumed at 45 nm will be only 2 X power
> consumed at 90 nm
> ?i think even in reality power consumed at 45 nm will be
> 4 X power consumed at 90 nm

Power DENSITY scales about 1.66x/generation according to 
ITRS, so it is about 2.8x from 90 to 65nm. Your power 
density at 45nm is too high. It is about 8x higher than 
90nm.
> ?
> ?for area, i assumed that scaling floor-pan from 90 nm to
> 45 nm as optical zoom out
> ?Optical Shrink_scale = 45/90 , i know that scaling 90 nm
> into 45 nm technology is not just optical zoom out but
> from thermal point of view we can neglect the difference
> between technology scaling and optical zoom out.

If you scale area down, power should also be less, although 
not as fast as area scaling (that's why power density 
increases). Why did you assume power is 4 times higher? By 
the way, isn't the area scaled with (45/90)^2?

> ?
> ?
> What is the steady-state temperature of that block at
> 45nm?
> M Elsawaf:
> ?steady-state temperature is 184.47 C

For such a high steady-state temperature, you probably want 
to use lower package convection resistance to bring it 
down. This happens in real designs. Again, with less power 
density on your hot block, this temperature should also be 
much lower.

-Wei




From mohamed_elsawaf at yahoo.com  Sun Jan 25 00:02:38 2009
From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf)
Date: Sun, 25 Jan 2009 00:02:38 -0800 (PST)
Subject: [Hotspot] blockTemperature is increase very fast
In-Reply-To: <6918B134002440EEA3BB4281@mstu1>
Message-ID: <957866.29129.qm@web53708.mail.re2.yahoo.com>

Hi Mike and Wei?

Thank you for your help, i am reviewing the scaling keep you updated with the progress.
Thanks Mike for paper they are really useful
Wei, can you please tell me ITRS document name? having this scaling information: "Power DENSITY scales about 1.66x/generation according to ITRS"

so  power density scaling from 90 nm to 45 nm should be: 3 * 1.66x = 4.98x ?i am correct?

regarding the real time calculations a got confused my HotSpot configurations are as follow:

sampling_intvl		3.333e-06
base_proc_freq		3e+09

real time should be: 18265 samples * 3.33*10^(-6) seconds/cycle = 0.06082245 seconds

also i scaled Width and height by 45/90 so area scale is (45/90)^2
your,
M. Elsawaf




      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20090125/de4fb5f5/attachment.html 

From michaelkadin at gmail.com  Sun Jan 25 05:16:07 2009
From: michaelkadin at gmail.com (Mike Kadin)
Date: Sun, 25 Jan 2009 08:16:07 -0500
Subject: [Hotspot] blockTemperature is increase very fast
In-Reply-To: <957866.29129.qm@web53708.mail.re2.yahoo.com>
References: <6918B134002440EEA3BB4281@mstu1>
	<957866.29129.qm@web53708.mail.re2.yahoo.com>
Message-ID: <34954c9d0901250516g5a6dfd2dx25bbbc74ce80144f@mail.gmail.com>

Hi Mohamed,

Scaling from 90nm to 45nm is two technology generations:  90nm -> 65nm and
then 65nm->45nm.  So The power density should be multiplied by 1.66 TWICE
not THREE times.

Wei, I too would be interested in the source for the 1.66x number.  When I
was researching thermal management, I had a lot of trouble finding that
information.

Mohamed, your area and real-time calculations look legitimate.

Hope that was helpful,
Mike Kadin

Research Assistant
Brown University Division of Engineering
201-463-3029
MichaelKadin at gmail.com


On Sun, Jan 25, 2009 at 3:02 AM, Mohamed Elsawaf
<mohamed_elsawaf at yahoo.com>wrote:

> Hi Mike and Wei
>
> Thank you for your help, i am reviewing the scaling keep you updated with
> the progress.
>
> Thanks Mike for paper they are really useful
>
> Wei, can you please tell me ITRS document name? having this scaling
> information: "Power DENSITY scales about 1.66x/generation according to ITRS"
>
> so power density scaling from 90 nm to 45 nm should be: 3 * 1.66x = 4.98x
>
> i am correct?
>
> regarding the real time calculations a got confused my HotSpot
> configurations are as follow:
>
> sampling_intvl 3.333e-06
> base_proc_freq 3e+09
>
> real time should be: 18265 samples * 3.33*10^(-6) seconds/cycle =
> 0.06082245 seconds
>
> also i scaled Width and height by 45/90 so area scale is (45/90)^2
>
>
> your,
> M. Elsawaf
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20090125/e86173c8/attachment-0001.html 

From wh6p at cms.mail.virginia.edu  Mon Jan 26 05:56:00 2009
From: wh6p at cms.mail.virginia.edu (Wei Huang)
Date: Mon, 26 Jan 2009 08:56:00 -0500
Subject: [Hotspot] blockTemperature is increase very fast
In-Reply-To: <34954c9d0901250516g5a6dfd2dx25bbbc74ce80144f@mail.gmail.com>
References: <6918B134002440EEA3BB4281@mstu1>
	<957866.29129.qm@web53708.mail.re2.yahoo.com>
	<34954c9d0901250516g5a6dfd2dx25bbbc74ce80144f@mail.gmail.com>
Message-ID: <web-6885691@guppy.mail.virginia.edu>

Hi, Mohamed and Mike,

I think the scaling from 90 to 45 should k^2, where k is the scaling factor.
Regarding power density scaling, it is not explicitly mentioned in any ITRS 
document. But you can approximate by Dynamic Power/Area. Whether it is 1.66 
or 1.5 or 1.2 is not important. The trend of increasing PD is what we care 
about.

-Wei

On Sun, 25 Jan 2009 08:16:07 -0500
  Mike Kadin <michaelkadin at gmail.com> wrote:
> Hi Mohamed,
> 
> Scaling from 90nm to 45nm is two technology generations:  90nm -> 65nm and
> then 65nm->45nm.  So The power density should be multiplied by 1.66 TWICE
> not THREE times.
> 
> Wei, I too would be interested in the source for the 1.66x number.  When I
> was researching thermal management, I had a lot of trouble finding that
> information.
> 
> Mohamed, your area and real-time calculations look legitimate.
> 
> Hope that was helpful,
> Mike Kadin
> 
> Research Assistant
> Brown University Division of Engineering
> 201-463-3029
> MichaelKadin at gmail.com
> 
> 
> On Sun, Jan 25, 2009 at 3:02 AM, Mohamed Elsawaf
> <mohamed_elsawaf at yahoo.com>wrote:
> 
>> Hi Mike and Wei
>>
>> Thank you for your help, i am reviewing the scaling keep you updated with
>> the progress.
>>
>> Thanks Mike for paper they are really useful
>>
>> Wei, can you please tell me ITRS document name? having this scaling
>> information: "Power DENSITY scales about 1.66x/generation according to 
>>ITRS"
>>
>> so power density scaling from 90 nm to 45 nm should be: 3 * 1.66x = 4.98x
>>
>> i am correct?
>>
>> regarding the real time calculations a got confused my HotSpot
>> configurations are as follow:
>>
>> sampling_intvl 3.333e-06
>> base_proc_freq 3e+09
>>
>> real time should be: 18265 samples * 3.33*10^(-6) seconds/cycle =
>> 0.06082245 seconds
>>
>> also i scaled Width and height by 45/90 so area scale is (45/90)^2
>>
>>
>> your,
>> M. Elsawaf
>>
>>
>>

Wei Huang, PhD
Postdoctoral Researcher
Dept. of Computer Science
University of Virginia
Charlottesville, VA 22904

From mohamed_elsawaf at yahoo.com  Mon Jan 26 06:28:50 2009
From: mohamed_elsawaf at yahoo.com (Mohamed Elsawaf)
Date: Mon, 26 Jan 2009 06:28:50 -0800 (PST)
Subject: [Hotspot] blockTemperature is increase very fast
In-Reply-To: <web-6885691@guppy.mail.virginia.edu>
Message-ID: <323791.70722.qm@web53701.mail.re2.yahoo.com>

Hi Weithanks for the update, but why we select ratio 1.6 or 1.5 or 1.2?May be ratio is greater than this 2 or 4 or even 8How can i select correct ratio? do we have any guide line? Or range?Thank youyours M Elsawaf--- On Mon, 1/26/09, Wei Huang <wh6p at cms.mail.virginia.edu> wrote:
From: Wei Huang <wh6p at cms.mail.virginia.edu>
Subject: Re: [Hotspot] blockTemperature is increase very fast
To: "Mike Kadin" <michaelkadin at gmail.com>, mohamed_elsawaf at yahoo.com
Cc: hotspot at mail.cs.virginia.edu
Date: Monday, January 26, 2009, 6:56 AM

Hi, Mohamed and Mike,

I think the scaling from 90 to 45 should k^2, where k is the scaling factor.
Regarding power density scaling, it is not explicitly mentioned in any ITRS 
document. But you can approximate by Dynamic Power/Area. Whether it is 1.66 
or 1.5 or 1.2 is not important. The trend of increasing PD is what we care 
about.

-Wei

On Sun, 25 Jan 2009 08:16:07 -0500
  Mike Kadin <michaelkadin at gmail.com> wrote:
> Hi Mohamed,
> 
> Scaling from 90nm to 45nm is two technology generations:  90nm -> 65nm
and
> then 65nm->45nm.  So The power density should be multiplied by 1.66
TWICE
> not THREE times.
> 
> Wei, I too would be interested in the source for the 1.66x number.  When I
> was researching thermal management, I had a lot of trouble finding that
> information.
> 
> Mohamed, your area and real-time calculations look legitimate.
> 
> Hope that was helpful,
> Mike Kadin
> 
> Research Assistant
> Brown University Division of Engineering
> 201-463-3029
> MichaelKadin at gmail.com
> 
> 
> On Sun, Jan 25, 2009 at 3:02 AM, Mohamed Elsawaf
> <mohamed_elsawaf at yahoo.com>wrote:
> 
>> Hi Mike and Wei
>>
>> Thank you for your help, i am reviewing the scaling keep you updated
with
>> the progress.
>>
>> Thanks Mike for paper they are really useful
>>
>> Wei, can you please tell me ITRS document name? having this scaling
>> information: "Power DENSITY scales about 1.66x/generation
according to 
>>ITRS"
>>
>> so power density scaling from 90 nm to 45 nm should be: 3 * 1.66x =
4.98x
>>
>> i am correct?
>>
>> regarding the real time calculations a got confused my HotSpot
>> configurations are as follow:
>>
>> sampling_intvl 3.333e-06
>> base_proc_freq 3e+09
>>
>> real time should be: 18265 samples * 3.33*10^(-6) seconds/cycle =
>> 0.06082245 seconds
>>
>> also i scaled Width and height by 45/90 so area scale is (45/90)^2
>>
>>
>> your,
>> M. Elsawaf
>>
>>
>>

Wei Huang, PhD
Postdoctoral Researcher
Dept. of Computer Science
University of Virginia
Charlottesville, VA 22904



      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20090126/eb6385f2/attachment.html 

From skadron at cs.virginia.edu  Mon Jan 26 07:11:29 2009
From: skadron at cs.virginia.edu (Kevin Skadron)
Date: Mon, 26 Jan 2009 10:11:29 -0500
Subject: [Hotspot] blockTemperature is increase very fast
In-Reply-To: <323791.70722.qm@web53701.mail.re2.yahoo.com>
References: <323791.70722.qm@web53701.mail.re2.yahoo.com>
Message-ID: <497DD2A1.4070802@cs.virginia.edu>

If you take k as the feature size scaling per technology generation, and 
shrink a particular design according to Dennard scaling, per generation 
we should observe:

A goes down by k^2
C goes down by k (ideally, although parasitics mean that it is less)
f goes up by k (ideally, in practice probably less)
V goes down by k (but ITRS projects that V almost flatlines, scaling 
only about 2.5% per generation)

Ideally, according to PV^2f and neglecting leakage, P also goes down by 
k^2 so we have constant power density.  In reality, if we assume k=0.7, 
C scaling down by only 0.8, f going up by 1.1, and V going down only 
2.5%, we get P only going down by 0.84 (0.8 * 0.975^2 * 1.1, rounded).

So power density (P/A) is scaling 0.84/0.5 = ~1.7 per generation with 
these assumptions.  You can vary the assumptions a lot, and these 
calculations assume that leakage stays a constant fraction.  The need to 
control leakage is the main factor restricting reductions in Vdd.  This 
choice of 1.1 for f is somewhat arbitrary.  This rough calculation also 
assumes a constant core microarchitecture, which isn't necessarily 
realistic.

My perception is that it seems unlikely designers will allow leakage to 
grow much beyond its current fraction of total power.  So power density 
seems unlikely to scale much more than 2X per generation.  It is also 
possible to come up with reasonable scenarios where power density scales 
much less.  Another factor to keep in mind is that air cooling appears 
to have reached affordable limits, hence total air cooling capacity is 
also limited to a thermal design power (TDP) of about 200W (depending on 
whose numbers you believe).  This caps what your chip can do and this 
has some implications for your power density if you assume constant chip 
size.

In the end, you have to make a case for your assumptions and justify, 
perhaps with some sensitivity analysis.

Hope this helps,
--Kevin


Mohamed Elsawaf wrote:
> Hi Wei
> 
> thanks for the update, but why we select ratio 1.6 or 1.5 or 1.2?
> 
> May be ratio is greater than this 2 or 4 or even 8
> 
> How can i select correct ratio? do we have any guide line? Or range?
> 
> Thank you
> 
> yours
> 
> M Elsawaf
> 
> --- On *Mon, 1/26/09, Wei Huang /<wh6p at cms.mail.virginia.edu>/* wrote:
> 
>     From: Wei Huang <wh6p at cms.mail.virginia.edu>
>     Subject: Re: [Hotspot] blockTemperature is increase very fast
>     To: "Mike Kadin" <michaelkadin at gmail.com>, mohamed_elsawaf at yahoo.com
>     Cc: hotspot at mail.cs.virginia.edu
>     Date: Monday, January 26, 2009, 6:56 AM
> 
>     Hi, Mohamed and Mike,
> 
>     I think the scaling from 90 to 45 should k^2, where k is the scaling factor.
>     Regarding power density scaling, it is not explicitly mentioned in any ITRS 
>     document. But you can approximate by Dynamic Power/Area. Whether it is 1.66 
>     or 1.5 or 1.2 is not important. The trend of increasing PD is what we care 
>     about.
> 
>     -Wei
> 
>     On Sun, 25 Jan 2009 08:16:07 -0500
>       Mike Kadin <michaelkadin at gmail.com> wrote:
>     > Hi Mohamed,
>     > 
>     > Scaling from 90nm to 45nm is two technology generations:  90nm -> 65nm
>     and
>     > then 65nm->45nm.  So The power density should be multiplied by 1.66
>     TWICE
>     > not THREE times.
>     > 
>     > Wei, I too would be interested in the source for the 1.66x number.  When I
>     > was researching thermal management, I had a lot of trouble finding that
>     > information.
>     > 
>     > Mohamed, your area and real-time calculations look legitimate.
>     > 
>     > Hope that was helpful,
>     > Mike Kadin
>     > 
>     > Research Assistant
>     > Brown University Division of Engineering
>     > 201-463-3029
>     > MichaelKadin at gmail.com
>     > 
>     > 
>     > On Sun, Jan 25, 2009 at 3:02 AM, Mohamed Elsawaf
>     > <mohamed_elsawaf at yahoo.com>wrote:
>     > 
>     >> Hi Mike and Wei
>     >>
>     >> Thank you for your help, i am reviewing the scaling keep you updated
>     with
>     >> the progress.
>     >>
>     >> Thanks Mike for paper they are really useful
>     >>
>     >> Wei, can you please tell me ITRS document name? having this scaling
>     >> information: "Power DENSITY scales about 1.66x/generation
>     according to 
>     >>ITRS"
>     >>
>     >> so power density scaling from 90 nm to 45 nm should be: 3 * 1.66x =
>     4.98x
>     >>
>     >> i am correct?
>     >>
>     >> regarding the real time calculations a got confused my HotSpot
>     >> configurations are as follow:
>     >>
>     >> sampling_intvl 3.333e-06
>     >> base_proc_freq 3e+09
>     >>
>     >> real time should be: 18265 samples * 3.33*10^(-6) seconds/cycle =
>     >> 0.06082245 seconds
>     >>
>     >> also i scaled Width and height by 45/90 so area scale is (45/90)^2
>     >>
>     >>
>     >> your,
>     >> M. Elsawaf
>     >>
>     >>
>     >>
> 
>     Wei Huang, PhD
>     Postdoctoral Researcher
>     Dept. of Computer Science
>     University of Virginia
>     Charlottesville, VA 22904
> 
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> HotSpot mailing list
> HotSpot at mail.cs.virginia.edu
> http://www.cs.virginia.edu/mailman/listinfo/hotspot

-- 
Kevin Skadron
Associate Professor of Computer Science
University of Virginia
Charlottesville, VA 22904
skadron \at/ cs.virginia.edu
http://www.cs.virginia.edu/~skadron
voice: (434) 982-2042,   fax: (434) 982-2214