From BulaPalm at gmail.com  Wed Jan 13 08:37:21 2010
From: BulaPalm at gmail.com (BulaPalm at gmail.com)
Date: Wed, 13 Jan 2010 16:37:21 +0000
Subject: [Hotspot] ptrace unit
Message-ID: <0016e64cc984759b9d047d0e6267@google.com>

Dear,

Are ptrace unit and HotSpot calling interval the same and what kind of unit  
of calling interval(ms or ns)?
Thank you!

Bula
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20100113/a29b4935/attachment.html 

From bernauer at informatik.uni-tuebingen.de  Thu Jan 14 00:35:28 2010
From: bernauer at informatik.uni-tuebingen.de (Andreas Bernauer)
Date: Thu, 14 Jan 2010 09:35:28 +0100
Subject: [Hotspot] ptrace unit
In-Reply-To: <0016e64cc984759b9d047d0e6267@google.com>
References: <0016e64cc984759b9d047d0e6267@google.com>
Message-ID: <c92980111001140035t74eeb573ubb5d0e7b87eeb4f1@mail.gmail.com>

On Wed, Jan 13, 2010 at 17:37, <BulaPalm at gmail.com> wrote:

> Dear,
>
> Are ptrace unit and HotSpot calling interval the same and what kind of unit
> of calling interval(ms or ns)?
>

You can set the sampling interval of HotSpot in its configuration file
hotspot.config.

I don't know about ptrace.

Cheers,

-- 
Andreas Bernauer <bernauer at informatik.uni-tuebingen.de>
WSI/TI, Sand 13, B202, 72076 T?bingen, +49 70 71 29 75 940
http://www.ti.uni-tuebingen.de/Andreas_Bernauer.227.0.html
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20100114/fd0c0d22/attachment.html 

From fsamie at gmail.com  Thu Jan 14 12:45:51 2010
From: fsamie at gmail.com (farzad samie)
Date: Fri, 15 Jan 2010 00:15:51 +0330
Subject: [Hotspot] ptrace unit
Message-ID: <ffd054461001141245t2ef9bfc8qc880788dd646702c@mail.gmail.com>

Hello Bula,
I think that ptrace unit is Watt and sampling interval corresponds to
10,000 cycle of your processor's clock frequency. by default, sampling
interval is 3.333e-6 second.
hope it helps you.

On Thu, Jan 14, 2010 at 11:31 PM, <hotspot-request at cs.virginia.edu> wrote:

> Send HotSpot mailing list submissions to
>        hotspot at mail.cs.virginia.edu
>
> To subscribe or unsubscribe via the World Wide Web, visit
>        http://www.cs.virginia.edu/mailman/listinfo/hotspot
> or, via email, send a message with subject or body 'help' to
>        hotspot-request at mail.cs.virginia.edu
>
> You can reach the person managing the list at
>        hotspot-owner at mail.cs.virginia.edu
>
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of HotSpot digest..."
>
>
> Today's Topics:
>
>   1. Re: ptrace unit (Andreas Bernauer)
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Thu, 14 Jan 2010 09:35:28 +0100
> From: Andreas Bernauer <bernauer at informatik.uni-tuebingen.de>
> Subject: Re: [Hotspot] ptrace unit
> To: BulaPalm at gmail.com
> Cc: hotspot at mail.cs.virginia.edu
> Message-ID:
>        <c92980111001140035t74eeb573ubb5d0e7b87eeb4f1 at mail.gmail.com>
> Content-Type: text/plain; charset="iso-8859-1"
>
> On Wed, Jan 13, 2010 at 17:37, <BulaPalm at gmail.com> wrote:
>
> > Dear,
> >
> > Are ptrace unit and HotSpot calling interval the same and what kind of
> unit
> > of calling interval(ms or ns)?
> >
>
> You can set the sampling interval of HotSpot in its configuration file
> hotspot.config.
>
> I don't know about ptrace.
>
> Cheers,
>
> --
> Andreas Bernauer <bernauer at informatik.uni-tuebingen.de>
> WSI/TI, Sand 13, B202, 72076 T?bingen, +49 70 71 29 75 940
> http://www.ti.uni-tuebingen.de/Andreas_Bernauer.227.0.html
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL:
> http://www.cs.virginia.edu/pipermail/hotspot/attachments/20100114/fd0c0d22/attachment-0001.html
>
> ------------------------------
>
> _______________________________________________
> HotSpot mailing list
> HotSpot at mail.cs.virginia.edu
> http://www.cs.virginia.edu/mailman/listinfo/hotspot
>
>
> End of HotSpot Digest, Vol 31, Issue 2
> **************************************
>



-- 
Farzad Samie
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20100115/00bb6536/attachment.html 

From vsn at umail.ucsb.edu  Fri Jan 15 12:10:03 2010
From: vsn at umail.ucsb.edu (Vivek S Nandakumar)
Date: Fri, 15 Jan 2010 12:10:03 -0800
Subject: [Hotspot] resizing spreader and heatsink issue
Message-ID: <4B50CB9B.1080109@umail.ucsb.edu>

Hi,

I am using HotSpot for my research at UCSB and have the following 
questions:

1. I am trying to determine the steady state temperature of a 
microprocessor SRAM register block of width 0.1024mm and height 0.076mm. 
I have divided this block into four sub-blocks. I would like to run this 
block as a standalone in HotSpot (as in step 1). But I am not able to 
resize the spreader and the heatsink size in the config file (HotSpot 
generates some error if I do that). Because of the large size of the 
spreader and sink compared to my register block, my steady state 
temperature always reaches the ambient value (318K). However, if I 
replace this block in ev6 floorplan (as in step 2), I get reasonably 
high temperatures as I would expect (~350K). My question is, are there 
any guidelines to follow to size the spreader and the heatsink for the 
program to work correctly so that I can simulate the register block 
seperately?

Step 1.
./hotspot -c hotspot.config -f flp36_l -p power36_l

Step 2:
./hotspot -c hotspot.config -f ev6_viv.flp -p gcc_viv.ptrace -model_type 
grid

2.
 From the forum archives it seems that mostly people use Wattch to 
generate the power trace file. I am using HSPICE to estimate the worst 
case power of the most critical path in the SRAM array.  I calculate the 
average dynamic power consumption for a single read operation per cycle 
assign it to the blocks. The power values are of the order of ~uW which 
is fairly small. Can I use these power values directly in the ptrace 
file as I am using now?

Thanks,
Vivek
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: hotspot.config
Url: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20100115/efd8c5e5/attachment.cc 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: flp36_l
Url: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20100115/efd8c5e5/attachment-0001.cc 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: power36_l
Url: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20100115/efd8c5e5/attachment-0002.cc 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: ev6_viv.flp
Url: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20100115/efd8c5e5/attachment-0003.cc 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: gcc_viv.ptrace
Url: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20100115/efd8c5e5/attachment-0004.cc 

From sarom at ucla.edu  Fri Jan 15 13:23:12 2010
From: sarom at ucla.edu (Saro Meguerdichian)
Date: Fri, 15 Jan 2010 13:23:12 -0800
Subject: [Hotspot] ttrace temperatures changing too rapidly?
In-Reply-To: <14f2695d1001142051p5fb10415uff6c14797524eec9@mail.gmail.com>
References: <14f2695d1001142003p504e1fbej76a6ac72b64cdf68@mail.gmail.com>
	<14f2695d1001142051p5fb10415uff6c14797524eec9@mail.gmail.com>
Message-ID: <14f2695d1001151323i497a1016sf5b22cf7faa36965@mail.gmail.com>

Using the default HotSpot config, I'm seeing temperatures changing on
the order of a couple degrees up and down in time intervals on the
order of 100s of microseconds. Is this reasonable? Why could this be
happening? The power numbers in the ptrace file range from 0 to 2
Watts.

Thanks,
Saro

From wh6p at virginia.edu  Fri Jan 15 14:04:54 2010
From: wh6p at virginia.edu (Wei Huang)
Date: Fri, 15 Jan 2010 17:04:54 -0500
Subject: [Hotspot] resizing spreader and heatsink issue
In-Reply-To: <4B50CB9B.1080109@umail.ucsb.edu>
References: <4B50CB9B.1080109@umail.ucsb.edu>
Message-ID: <web-7937369@tetra.mail.virginia.edu>

Hi Vivek,

First, I wouldn't recommend run only the SRAM array as if it is a standalone 
chip, because this doesn't include the thermal impact of neighboring blocks 
that are present in actual chips. If you really want to do so, please make 
sure that the floorplan file is updated with the other blocks removed, and 
the coordination of the SRAM array's lower-left corner is set to (0,0). 
After that, as long as the spreader&sink have sizes greater than the SRAM 
block, you should be good to go.

As for the unit of power, you'll have to convert it to watts.

Hope this helps.
-Wei


On Fri, 15 Jan 2010 12:10:03 -0800
  Vivek S Nandakumar <vsn at umail.ucsb.edu> wrote:
> Hi,
> 
> I am using HotSpot for my research at UCSB and have the following 
>questions:
> 
> 1. I am trying to determine the steady state temperature of a 
>microprocessor SRAM register block of width 0.1024mm and height 0.076mm. I 
>have divided this block into four sub-blocks. I would like to run this 
>block as a standalone in HotSpot (as in step 1). But I am not able to 
>resize the spreader and the heatsink size in the config file (HotSpot 
>generates some error if I do that). Because of the large size of the 
>spreader and sink compared to my register block, my steady state 
>temperature always reaches the ambient value (318K). However, if I replace 
>this block in ev6 floorplan (as in step 2), I get reasonably high 
>temperatures as I would expect (~350K). My question is, are there any 
>guidelines to follow to size the spreader and the heatsink for the program 
>to work correctly so that I can simulate the register block seperately?
> 
> Step 1.
> ./hotspot -c hotspot.config -f flp36_l -p power36_l
> 
> Step 2:
> ./hotspot -c hotspot.config -f ev6_viv.flp -p gcc_viv.ptrace -model_type 
>grid
> 
> 2.
> From the forum archives it seems that mostly people use Wattch to generate 
>the power trace file. I am using HSPICE to estimate the worst case power of 
>the most critical path in the SRAM array.  I calculate the average dynamic 
>power consumption for a single read operation per cycle assign it to the 
>blocks. The power values are of the order of ~uW which is fairly small. Can 
>I use these power values directly in the ptrace file as I am using now?
> 
> Thanks,
> Vivek


From wh6p at virginia.edu  Fri Jan 15 14:08:48 2010
From: wh6p at virginia.edu (Wei Huang)
Date: Fri, 15 Jan 2010 17:08:48 -0500
Subject: [Hotspot] ttrace temperatures changing too rapidly?
In-Reply-To: <14f2695d1001151323i497a1016sf5b22cf7faa36965@mail.gmail.com>
References: <14f2695d1001142003p504e1fbej76a6ac72b64cdf68@mail.gmail.com>
	<14f2695d1001142051p5fb10415uff6c14797524eec9@mail.gmail.com>
	<14f2695d1001151323i497a1016sf5b22cf7faa36965@mail.gmail.com>
Message-ID: <web-7937411@tetra.mail.virginia.edu>

Hi Saro,

I would consider this is reasonable. Our first-order analysis shows that the 
"thermal time constant" of silicon is around milliseconds, so in tenths of 
ms, you should be able to see some moderate temperature changes.

Just curious, why would you think this change is too fast, do you have some 
source of references?

Regards,

-Wei


On Fri, 15 Jan 2010 13:23:12 -0800
  Saro Meguerdichian <sarom at ucla.edu> wrote:
> Using the default HotSpot config, I'm seeing temperatures changing on
> the order of a couple degrees up and down in time intervals on the
> order of 100s of microseconds. Is this reasonable? Why could this be
> happening? The power numbers in the ptrace file range from 0 to 2
> Watts.
> 
> Thanks,
> Saro
> _______________________________________________
> HotSpot mailing list
> HotSpot at mail.cs.virginia.edu
> http://www.cs.virginia.edu/mailman/listinfo/hotspot


From vsn at umail.ucsb.edu  Sun Jan 17 00:16:20 2010
From: vsn at umail.ucsb.edu (Vivek S Nandakumar)
Date: Sun, 17 Jan 2010 00:16:20 -0800
Subject: [Hotspot] resizing spreader and heatsink issue
In-Reply-To: <web-7937369@tetra.mail.virginia.edu>
References: <4B50CB9B.1080109@umail.ucsb.edu>
	<web-7937369@tetra.mail.virginia.edu>
Message-ID: <4B52C754.6040703@umail.ucsb.edu>

Hi Wei,

Thanks for your reply. I have a few more questions:

1. It seems like the layer.lcf file uses resistivity values instead of 
conductivity values as in config file. For the silicon layer, you have 
used the thermal resistivity of Si which is 0.01. What exactly is the 
interface layer made of and how did you calculate its thermal 
resistivity (0.25)? Does this layer represent air+interconnects?

2. I'm trying to use the layer.lcf file to model a single layer as 3 
sub-layers (bulk silicon, active silicon and metal) as shown in the 
paper "Thermal Analysis of a 3D Die-Stacked High Performance 
Microprocessor". In this paper the authors model the metal layer as a 
combination of resistivities of both copper and SiO2 (dielectric). Do 
you have any suggestions on how to differentiate the bulk from the 
active layer based on thermal resistivity?

Thanks,
Vivek



Wei Huang wrote:
> Hi Vivek,
>
> First, I wouldn't recommend run only the SRAM array as if it is a 
> standalone chip, because this doesn't include the thermal impact of 
> neighboring blocks that are present in actual chips. If you really 
> want to do so, please make sure that the floorplan file is updated 
> with the other blocks removed, and the coordination of the SRAM 
> array's lower-left corner is set to (0,0). After that, as long as the 
> spreader&sink have sizes greater than the SRAM block, you should be 
> good to go.
>
> As for the unit of power, you'll have to convert it to watts.
>
> Hope this helps.
> -Wei
>
>
> On Fri, 15 Jan 2010 12:10:03 -0800
>  Vivek S Nandakumar <vsn at umail.ucsb.edu> wrote:
>> Hi,
>>
>> I am using HotSpot for my research at UCSB and have the following 
>> questions:
>>
>> 1. I am trying to determine the steady state temperature of a 
>> microprocessor SRAM register block of width 0.1024mm and height 
>> 0.076mm. I have divided this block into four sub-blocks. I would like 
>> to run this block as a standalone in HotSpot (as in step 1). But I am 
>> not able to resize the spreader and the heatsink size in the config 
>> file (HotSpot generates some error if I do that). Because of the 
>> large size of the spreader and sink compared to my register block, my 
>> steady state temperature always reaches the ambient value (318K). 
>> However, if I replace this block in ev6 floorplan (as in step 2), I 
>> get reasonably high temperatures as I would expect (~350K). My 
>> question is, are there any guidelines to follow to size the spreader 
>> and the heatsink for the program to work correctly so that I can 
>> simulate the register block seperately?
>>
>> Step 1.
>> ./hotspot -c hotspot.config -f flp36_l -p power36_l
>>
>> Step 2:
>> ./hotspot -c hotspot.config -f ev6_viv.flp -p gcc_viv.ptrace 
>> -model_type grid
>>
>> 2.
>> From the forum archives it seems that mostly people use Wattch to 
>> generate the power trace file. I am using HSPICE to estimate the 
>> worst case power of the most critical path in the SRAM array.  I 
>> calculate the average dynamic power consumption for a single read 
>> operation per cycle assign it to the blocks. The power values are of 
>> the order of ~uW which is fairly small. Can I use these power values 
>> directly in the ptrace file as I am using now?
>>
>> Thanks,
>> Vivek
>

From wh6p at virginia.edu  Tue Jan 19 02:45:29 2010
From: wh6p at virginia.edu (Wei Huang)
Date: Tue, 19 Jan 2010 05:45:29 -0500
Subject: [Hotspot] resizing spreader and heatsink issue
In-Reply-To: <4B52C754.6040703@umail.ucsb.edu>
References: <4B50CB9B.1080109@umail.ucsb.edu>
	<web-7937369@tetra.mail.virginia.edu>
	<4B52C754.6040703@umail.ucsb.edu>
Message-ID: <web-7961419@neon.mail.virginia.edu>

> 1. It seems like the layer.lcf file uses resistivity values instead of 
>conductivity values as in config file. For the silicon layer, you have used 
>the thermal resistivity of Si which is 0.01. What exactly is the interface 
>layer made of and how did you calculate its thermal resistivity (0.25)? 
>Does this layer represent air+interconnects?

The TIM (thermal interface layer) is just a layer of thermal paste that 
glues heat spreader to silicon. There are varous types of thermal pastes (ie 
TIM), we just used a reasonable one. You can google for more TIM materials.
> 
> 2. I'm trying to use the layer.lcf file to model a single layer as 3 
>sub-layers (bulk silicon, active silicon and metal) as shown in the paper 
>"Thermal Analysis of a 3D Die-Stacked High Performance Microprocessor". In 
>this paper the authors model the metal layer as a combination of 
>resistivities of both copper and SiO2 (dielectric). Do you have any 
>suggestions on how to differentiate the bulk from the active layer based on 
>thermal resistivity?
> 
It is OK to model active Si, bulk Si and metals in different layers using 
.lcf config and the grid model. I don't have suggested values for thermal 
properties of active silicon and metal (and associated oxide). I remeber a 
paper by K. Banerjee from UCSB has some data on these:

Scaling Analysis of Multilevel Interconnect Temperatures in High Performance 
ICs. by Sungjun Im, Navin Srivastava, Kaustav Banerjee and Kenneth E. 
Goodson

Hope this helps.

-Wei