Comparing virtex2 to spartan6

[Thomas Heller]

I know this is a question which is difficult to be answered,
but does anyone have an estimate which Spartan-6 fpga has
comparable 'size' as a xc2v1000 or a xc2v3000 fpga?
With 'size' I don't mean chip dimensions, I mean logic
functionality.

A very rough estimate would be enough...

Thanks,
Thomas

 

[Martin Thompson]

Thomas Heller <theller@ctypes.org> writes:

I know this is a question which is difficult to be answered,
but does anyone have an estimate which Spartan-6 fpga has
comparable 'size' as a xc2v1000 or a xc2v3000 fpga?
With 'size' I don't mean chip dimensions, I mean logic
functionality.

xc2v1000 has about 5000 slices, so about 10k (4-input) luts and
flipflops. It also has 40 blockrams and 40 multipliers.

It depends if you are LUT or flipflop limited (or memory/multiplier).
Note that Spartan-6 LUTs are 6-input and there re 2 flops per LUT.

A Spartan 6 LX16 has about 9k LUTs (hence 18k flipflops). The next
one down might suit (LX9) - ~5700 LUTs, >10k flipflops.

Both have 32 BRAMs, 16 DSP blocks (multiplier + extras) in the LX9, 32
in the LX16.

You can do the comparisons from here:

http://www.xilinx.com/support/documentation/data_sheets/ds031.pdf
http://www.xilinx.com/support/documentation/data_sheets/ds160.pdf

A very rough estimate would be enough...

Given the change in CLB architecture, that's all it'll ever be unless you
port your code across...

Cheers,
Martin


--
martin.j.thompson@trw.com
TRW Conekt - Consultancy in Engineering, Knowledge and Technology
http://www.conekt.co.uk/capabilities/39-electronic-hardware

 

[Thomas Heller]

Am 10.05.2012 11:32, schrieb Martin Thompson:

Thomas Heller<theller@ctypes.org> writes:

I know this is a question which is difficult to be answered,
but does anyone have an estimate which Spartan-6 fpga has
comparable 'size' as a xc2v1000 or a xc2v3000 fpga?
With 'size' I don't mean chip dimensions, I mean logic
functionality.


xc2v1000 has about 5000 slices, so about 10k (4-input) luts and
flipflops. It also has 40 blockrams and 40 multipliers.

It depends if you are LUT or flipflop limited (or memory/multiplier).
Note that Spartan-6 LUTs are 6-input and there re 2 flops per LUT.

A Spartan 6 LX16 has about 9k LUTs (hence 18k flipflops). The next
one down might suit (LX9) - ~5700 LUTs,>10k flipflops.

Both have 32 BRAMs, 16 DSP blocks (multiplier + extras) in the LX9, 32
in the LX16.

You can do the comparisons from here:

http://www.xilinx.com/support/documentation/data_sheets/ds031.pdf
http://www.xilinx.com/support/documentation/data_sheets/ds160.pdf

A very rough estimate would be enough...


Given the change in CLB architecture, that's all it'll ever be unless you
port your code across...

Well, this confirms the metrics that I used myself somewhat.
Does the virtex 2 or the spartan 6 have a higher operating frequency,
in practice?

Thanks,
Thomas

 

[Gabor]

On May 10, 1:54 pm, Thomas Heller <thel...@ctypes.org> wrote:

Am 10.05.2012 11:32, schrieb Martin Thompson:









Thomas Heller<thel...@ctypes.org>  writes:

I know this is a question which is difficult to be answered,
but does anyone have an estimate which Spartan-6 fpga has
comparable 'size' as a xc2v1000 or a xc2v3000 fpga?
With 'size' I don't mean chip dimensions, I mean logic
functionality.

xc2v1000 has about 5000 slices, so about 10k (4-input) luts and
flipflops. It also has 40 blockrams and 40 multipliers.

It depends if you are LUT or flipflop limited (or memory/multiplier).
Note that Spartan-6 LUTs are 6-input and there re 2 flops per LUT.

A Spartan 6 LX16 has about 9k LUTs (hence 18k flipflops).  The next
one down might suit (LX9) - ~5700 LUTs,>10k flipflops.

Both have 32 BRAMs, 16 DSP blocks (multiplier + extras) in the LX9, 32
in the LX16.

You can do the comparisons from here:

http://www.xilinx.com/support/documentation/data_sheets/ds031.pdf
http://www.xilinx.com/support/documentation/data_sheets/ds160.pdf

A very rough estimate would be enough...

Given the change in CLB architecture, that's all it'll ever be unless you
port your code across...

Well, this confirms the metrics that I used myself somewhat.
Does the virtex 2 or the spartan 6 have a higher operating frequency,
in practice?

Thanks,
Thomas

I've used both V2 and S6 (among others) and find S6 is easier to meet
the
same frequencies with. Also S6 has much better I/O performance and
a lot of features that were not available in the V2 days like SERDES
on
general IO pins, variable delay elements, PLL's and hard memory
control
blocks that allow the S6 to easily interface with DDR2 memory at
300+ MHz. even though the fabric clock rate is not much better than
V2's. The 6-input LUT's help to reduce logic levels, so depending on
your design requirements you could gain more speed. Also the S6
has the newer DSP48 structures instead of the simple multipliers
from the V2. V2 had more DCM's and block RAM's for the same
size part, but you can afford a larger S6 so you save a lot of money
even if you need to move to a part with more LUTs.

-- Gabor

 

[Gabor]

Thomas Heller wrote:

Am 10.05.2012 11:32, schrieb Martin Thompson:
[snip]

Well, this confirms the metrics that I used myself somewhat.
Does the virtex 2 or the spartan 6 have a higher operating frequency,
in practice?

Thanks,
Thomas

In my experience, Spartan 6 runs faster than Virtex 2. In addition,
you can reduce the number of logic levels because of larger LUTs.
Spartan 6 also has many features that weren't introduced before
Virtex 4, like DSP48's, SERDES on general I/O, variable IO delays,
PLL's...

Although the Spartan 6 fabric is not a whole lot faster than V2,
the hard memory controllers allow you to use faster external
memory, which can reduce pin count and power over using wider
slower memory. DDR2 can easily do 312 MHz (400 MHz with tweaked
VCCint). Also S6 uses much less power than V2.

As noted, S6 doesn't have as much block RAM for the same fabric
size, however the much lower price and lower power make it a
win even if you need to use a larger S6 part to get the memory
you need.

-- Gabor