On Oct 13, 11:12 pm, radarman <jsham...@gmail.com> wrote:
> On Oct 6, 12:06 am, rickman <gnu...@gmail.com> wrote:
>
>
>
> > On Oct 5, 5:50 pm, Ed McGettigan <ed.mcgetti...@xilinx.com> wrote:
>
> > > On Oct 5, 12:43 pm, rickman <gnu...@gmail.com> wrote:
>
> > > > I am listening to the Xilinx webinar on partial reconfiguration and
> > > > they say the Artix 7 will be the first "spartan" type device that will
> > > > be supported under the ISE software for partial reconfiguration.
> > > > Anyone heard when the Artix 7 devices are supposed to be in
> > > > production?  Are we talking about a practical time frame such as 6 to
> > > > 9 months or is it being stated as a year or more?  Or have they only
> > > > talked about when they will be shipping "samples"?  I know there can
> > > > be quite a difference.
>
> > > There is no announced date for Artix-7 device availability at this
> > > time.
>
> > Ok, so if I understand what was said in the presentation today, there
> > is no support for partial reconfiguration in any of the available
> > Spartan devices, right?  I remember looking into PR for quite a while
> > some years back and had been told that Xilinx "was committed" to
> > supporting PR in Spartan parts.  That was literally 8 years ago.  A
> > lot of the presentation talked to the cost savings that was possible
> > using PR.  But that only makes sense to me if it can be used with low
> > cost parts.  Reducing the cost of using a more expensive part by using
> > a complex process is a poor substitute to just using a part that costs
> > less.  But if I can use the less expensive part and cut my costs
> > further by using PR, that can make a number of projects practical that
> > otherwise wouldn't be.
>
> > I don't even want to do "dynamic" PR, I just want to have the
> > flexibility of configuring the modules when I configure the full part
> > initially.  Would that be "static" PR?  This makes a huge difference
> > in the exact scenario they described for a product with multiple
> > interface modules, but the number of possible modules larger than just
> > three.  Instead of a 100,000 LUT Spartan, I would be able to use a
> > 10,000 LUT device and have room to spare.  This could be such a
> > enabler of projects.
>
> > Rick
>
> You can sort of do what you describe now with a Spartan 3E and multi-
> boot. It's obviously not the same as partial reconfiguration, but with
> a bit of clever design, it can serve a similar purpose. It can also be
> entirely implemented in-chip.
> I have a little business-card sized demo board from an Avnet rep that
> does this. It boots up with a simple animation, but when you press a
> button, it turns into an electronic die (simulates rolling dice)
>
> While nifty, there are some real limitations, though. For one, you
> need external RAM if you want the two images to be able to pass data
> to each other - since all of the internal block RAM will get
> reinitialized. For another, there is a fairly significant delay where
> the FPGA is "dead" - meaning you need to be careful about when the
> switchover occurs.

Some eight years ago I built a DSP board which used small daughter
cards for I/O.  The idea was to build the minimum amount of hardware
possible and to do the bulk of the design in the FPGA.  A variety of
different module types would be supported by the CPU being able to
poll the daughter cards at boot up to determine their type and using
partial configuration to load in the appropriate module.  This didn't
even require that it be done while the FPGA was live, just that the
modules could be loaded independently.  Although I was told that the
Spartan could support this and that Xilinx was committed to supporting
Spartan PR in the tools, it never happened.  So instead I had to
configure a unique load for each customer on request.  That was not
very easy to support.  That is basically the "multi-boot" approach.
It's ok until the number of permutations gets too large.

Rick

On Oct 6, 12:06=A0am, rickman <gnu...@gmail.com> wrote:
> On Oct 5, 5:50 pm, Ed McGettigan <ed.mcgetti...@xilinx.com> wrote:
>
> > On Oct 5, 12:43 pm, rickman <gnu...@gmail.com> wrote:
>
> > > I am listening to the Xilinx webinar on partial reconfiguration and
> > > they say the Artix 7 will be the first "spartan" type device that wil=
l
> > > be supported under the ISE software for partial reconfiguration.
> > > Anyone heard when the Artix 7 devices are supposed to be in
> > > production? =A0Are we talking about a practical time frame such as 6 =
to
> > > 9 months or is it being stated as a year or more? =A0Or have they onl=
y
> > > talked about when they will be shipping "samples"? =A0I know there ca=
n
> > > be quite a difference.
>
> > There is no announced date for Artix-7 device availability at this
> > time.
>
> Ok, so if I understand what was said in the presentation today, there
> is no support for partial reconfiguration in any of the available
> Spartan devices, right? =A0I remember looking into PR for quite a while
> some years back and had been told that Xilinx "was committed" to
> supporting PR in Spartan parts. =A0That was literally 8 years ago. =A0A
> lot of the presentation talked to the cost savings that was possible
> using PR. =A0But that only makes sense to me if it can be used with low
> cost parts. =A0Reducing the cost of using a more expensive part by using
> a complex process is a poor substitute to just using a part that costs
> less. =A0But if I can use the less expensive part and cut my costs
> further by using PR, that can make a number of projects practical that
> otherwise wouldn't be.
>
> I don't even want to do "dynamic" PR, I just want to have the
> flexibility of configuring the modules when I configure the full part
> initially. =A0Would that be "static" PR? =A0This makes a huge difference
> in the exact scenario they described for a product with multiple
> interface modules, but the number of possible modules larger than just
> three. =A0Instead of a 100,000 LUT Spartan, I would be able to use a
> 10,000 LUT device and have room to spare. =A0This could be such a
> enabler of projects.
>
> Rick

You can sort of do what you describe now with a Spartan 3E and multi-
boot. It's obviously not the same as partial reconfiguration, but with
a bit of clever design, it can serve a similar purpose. It can also be
entirely implemented in-chip.
I have a little business-card sized demo board from an Avnet rep that
does this. It boots up with a simple animation, but when you press a
button, it turns into an electronic die (simulates rolling dice)

While nifty, there are some real limitations, though. For one, you
need external RAM if you want the two images to be able to pass data
to each other - since all of the internal block RAM will get
reinitialized. For another, there is a fairly significant delay where
the FPGA is "dead" - meaning you need to be careful about when the
switchover occurs.

On Oct 6, 3:56=A0am, Uwe Bonnes <b...@elektron.ikp.physik.tu-
darmstadt.de> wrote:
> Ed McGettigan <ed.mcgetti...@xilinx.com> wrote:
> > On Oct 5, 3:12=A0pm, Uwe Bonnes <b...@elektron.ikp.physik.tu-
> > darmstadt.de> wrote:
> > > Ed McGettigan <ed.mcgetti...@xilinx.com> wrote:
>
> > > ...
>
> > > > There is no announced date for Artix-7 device availability at this
> > > > time.
>
> > > Is there something announced for S6?
>
> > With the exception of the XC6SLX4 and XC6SLX9 all devices are in
> > production and in stock at Avnet for small quantities and 6-8 weeks
> > times for larger quantities.
> >http://avnetexpress.avnet.com/store/em/EMController?action=3Dproducts&N.=
..
>
> Any expected roll-out time for SLX4/SLX9? They are the only parts in QFP.=
..
>
> --
> Uwe Bonnes =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0b...@elektron.ikp.physik.tu-dar=
mstadt.de
>
> Institut fuer Kernphysik =A0Schlossgartenstrasse 9 =A064289 Darmstadt
> --------- Tel. 06151 162516 -------- Fax. 06151 164321 ----------- Hide q=
uoted text -
>
> - Show quoted text -

There isn't an announced date for the 6SLX4 and 6SLX9 devices at this
time, but it will be sooner rather than later.

Ed McGettigan
--
Xilinx Inc.

On Oct 6, 1:12 pm, Kolja Sulimma <ksuli...@googlemail.com> wrote:
> On 6 Okt., 07:06, rickman <gnu...@gmail.com> wrote:
>  A
>
> > lot of the presentation talked to the cost savings that was possible
> > using PR.  But that only makes sense to me if it can be used with low
> > cost parts.  Reducing the cost of using a more expensive part by using
> > a complex process is a poor substitute to just using a part that costs
> > less.  But if I can use the less expensive part and cut my costs
> > further by using PR, that can make a number of projects practical that
> > otherwise wouldn't be.
>
> I completely disagree.
> If you have a design that barely fits in a 6k$ part and can reduce the
> LUT
> count by a factor of 4 by using PR you save about 4k$ per chip
> This is what happens for DNA pattern matching machines.
>
> Try saving 4k$ with a spartan-3.
>
> The volume where saving 10$ per chip justifies the more complex PR
> design process
> is rather high.

Yes, the designs that use a $10 chip ARE high volume apps.  I've never
seen a part that costs $6000, but I did see one that used a $1500
chip.  The unit this was going in sold for $100,000 and they expected
to sell much less than 100 per year.  I remember that the initial
design for that chip used less than 20% of the LUTs.  Do you think
they cared at all about using a tool like PR?

I'm sure there are designs using Vertex chips (high cost) that would
benefit from the PR.  My point is that if PR were available for the
low cost Spartan line, it would result in significant additional
sales.  Xilinx doesn't want to sell lower costs chips unless it
somehow means more profit.  The only way PR makes profit for them is
if it sells chips they otherwise would not have sold.  That makes much
more sense at the low end than the high end.

The real problem is that Xilinx doesn't "get" the low end.  Yes, they
will brag about some product where they sold a million Spartan chips
without saying it required cutting their margin to under a buck,
picking an imaginary example.  But if PR makes a design feasible by
cutting the part size they could have sold the cheaper chips at the
same price with a two dollar margin yielding twice the total profit.
Or better yet, design variations of their chips for the truly low end
like Silicon Blue does.  Xilinx thinks they can't make any worthwhile
money at $2 a chip.  Maybe they can't, but others will.

Rick

On 6 Okt., 07:06, rickman <gnu...@gmail.com> wrote:
=A0A
> lot of the presentation talked to the cost savings that was possible
> using PR. =A0But that only makes sense to me if it can be used with low
> cost parts. =A0Reducing the cost of using a more expensive part by using
> a complex process is a poor substitute to just using a part that costs
> less. =A0But if I can use the less expensive part and cut my costs
> further by using PR, that can make a number of projects practical that
> otherwise wouldn't be.

I completely disagree.
If you have a design that barely fits in a 6k$ part and can reduce the
LUT
count by a factor of 4 by using PR you save about 4k$ per chip
This is what happens for DNA pattern matching machines.

Try saving 4k$ with a spartan-3.

The volume where saving 10$ per chip justifies the more complex PR
design process
is rather high.

Kolja

Ed McGettigan <ed.mcgettigan@xilinx.com> wrote:
> On Oct 5, 3:12&#4294967295;pm, Uwe Bonnes <b...@elektron.ikp.physik.tu-
> darmstadt.de> wrote:
> > Ed McGettigan <ed.mcgetti...@xilinx.com> wrote:
> >
> > ...
> >
> > > There is no announced date for Artix-7 device availability at this
> > > time.
> >
> > Is there something announced for S6?
> >

> With the exception of the XC6SLX4 and XC6SLX9 all devices are in
> production and in stock at Avnet for small quantities and 6-8 weeks
> times for larger quantities.

> http://avnetexpress.avnet.com/store/em/EMController?action=products&N=0&&term=XC6SLX

Any expected roll-out time for SLX4/SLX9? They are the only parts in QFP...

-- 
Uwe Bonnes                bon@elektron.ikp.physik.tu-darmstadt.de

Institut fuer Kernphysik  Schlossgartenstrasse 9  64289 Darmstadt
--------- Tel. 06151 162516 -------- Fax. 06151 164321 ----------

On Oct 5, 5:50 pm, Ed McGettigan <ed.mcgetti...@xilinx.com> wrote:
> On Oct 5, 12:43 pm, rickman <gnu...@gmail.com> wrote:
>
> > I am listening to the Xilinx webinar on partial reconfiguration and
> > they say the Artix 7 will be the first "spartan" type device that will
> > be supported under the ISE software for partial reconfiguration.
> > Anyone heard when the Artix 7 devices are supposed to be in
> > production?  Are we talking about a practical time frame such as 6 to
> > 9 months or is it being stated as a year or more?  Or have they only
> > talked about when they will be shipping "samples"?  I know there can
> > be quite a difference.
>
> There is no announced date for Artix-7 device availability at this
> time.

Ok, so if I understand what was said in the presentation today, there
is no support for partial reconfiguration in any of the available
Spartan devices, right?  I remember looking into PR for quite a while
some years back and had been told that Xilinx "was committed" to
supporting PR in Spartan parts.  That was literally 8 years ago.  A
lot of the presentation talked to the cost savings that was possible
using PR.  But that only makes sense to me if it can be used with low
cost parts.  Reducing the cost of using a more expensive part by using
a complex process is a poor substitute to just using a part that costs
less.  But if I can use the less expensive part and cut my costs
further by using PR, that can make a number of projects practical that
otherwise wouldn't be.

I don't even want to do "dynamic" PR, I just want to have the
flexibility of configuring the modules when I configure the full part
initially.  Would that be "static" PR?  This makes a huge difference
in the exact scenario they described for a product with multiple
interface modules, but the number of possible modules larger than just
three.  Instead of a 100,000 LUT Spartan, I would be able to use a
10,000 LUT device and have room to spare.  This could be such a
enabler of projects.

Rick

On Oct 5, 3:12=A0pm, Uwe Bonnes <b...@elektron.ikp.physik.tu-
darmstadt.de> wrote:
> Ed McGettigan <ed.mcgetti...@xilinx.com> wrote:
>
> ...
>
> > There is no announced date for Artix-7 device availability at this
> > time.
>
> Is there something announced for S6?
>
> --
> Uwe Bonnes =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0b...@elektron.ikp.physik.tu-dar=
mstadt.de
>
> Institut fuer Kernphysik =A0Schlossgartenstrasse 9 =A064289 Darmstadt
> --------- Tel. 06151 162516 -------- Fax. 06151 164321 ----------

With the exception of the XC6SLX4 and XC6SLX9 all devices are in
production and in stock at Avnet for small quantities and 6-8 weeks
times for larger quantities.

http://avnetexpress.avnet.com/store/em/EMController?action=3Dproducts&N=3D0=
&&term=3DXC6SLX

Ed McGettigan
--
Xilinx Inc.