Horsepower On Tap

R

Rob Gaddi

Guest
I was waiting for Quartus to finish crunching my latest build, and poking
around idly on NewEgg trying to see what it would cost to get a machine
with a little more juice to it. I started thinking what a shame it was to
have to keep upgrading to the latest and greatest machine in order to
squeeze out some more clock cycles for big builds. And then I started
thinking about Amazon EC2.

The idea behind EC2 is that Amazon runs virtual machines for you, and you
pay them by the hour for their use. Use of an "Extra Large High-Memory
Instance" running RHEL would run $0.63 an hour. It's the new spin on the
old "timeshare the supercomputer" concept. This seems like a handy way to
get a beast of a computer when I need to do big builds and/or long
simulation runs without having to keep upgrading my core machine.

Has anyone tried this out to see if it works in practice? http://
moxielogic.org/blog/?p=450 says he had serious trouble with it, but that's
from a year ago. Ideally, anyone with experience trying this with Quartus,
but ISE, Modelsim, Rivera, etc would all be interesting.

--
Rob Gaddi, Highland Technology -- www.highlandtechnology.com
Email address domain is currently out of order. See above to fix.
 
On 12/07/2011 11:54 AM, Rob Gaddi wrote:
I was waiting for Quartus to finish crunching my latest build, and poking
around idly on NewEgg trying to see what it would cost to get a machine
with a little more juice to it. I started thinking what a shame it was to
have to keep upgrading to the latest and greatest machine in order to
squeeze out some more clock cycles for big builds. And then I started
thinking about Amazon EC2.

The idea behind EC2 is that Amazon runs virtual machines for you, and you
pay them by the hour for their use. Use of an "Extra Large High-Memory
Instance" running RHEL would run $0.63 an hour. It's the new spin on the
old "timeshare the supercomputer" concept. This seems like a handy way to
get a beast of a computer when I need to do big builds and/or long
simulation runs without having to keep upgrading my core machine.

Has anyone tried this out to see if it works in practice? http://
moxielogic.org/blog/?p=450 says he had serious trouble with it, but that's
from a year ago. Ideally, anyone with experience trying this with Quartus,
but ISE, Modelsim, Rivera, etc would all be interesting.
Click to expand...
There certainly could be some license issues with installing software
licensed to you on a machine that is not under your control.
If you do this on a regular basis, it seems it would make more sense to
just buy a bare-bones machine with huge memory rather than go through
all the remote setup hassles. but, if you only do this very rarely,
maybe it makes sense.

Jon
 
HI,

Rob Gaddi skrev 2011-12-07 18:54:
I was waiting for Quartus to finish crunching my latest build, and poking
around idly on NewEgg trying to see what it would cost to get a machine
with a little more juice to it. I started thinking what a shame it was to
have to keep upgrading to the latest and greatest machine in order to
squeeze out some more clock cycles for big builds. And then I started
thinking about Amazon EC2.

The idea behind EC2 is that Amazon runs virtual machines for you, and you
pay them by the hour for their use. Use of an "Extra Large High-Memory
Instance" running RHEL would run $0.63 an hour. It's the new spin on the
old "timeshare the supercomputer" concept. This seems like a handy way to
get a beast of a computer when I need to do big builds and/or long
simulation runs without having to keep upgrading my core machine.

Has anyone tried this out to see if it works in practice? http://
moxielogic.org/blog/?p=450 says he had serious trouble with it, but that's
from a year ago. Ideally, anyone with experience trying this with Quartus,
but ISE, Modelsim, Rivera, etc would all be interesting.

Sun tried that for a number of years ago in their huge build/simulation
Click to expand...
center, but I think it failed when you need licensed software.


/michael
 
On 12/7/2011 11:54 AM, Rob Gaddi wrote:
I was waiting for Quartus to finish crunching my latest build, and poking
around idly on NewEgg trying to see what it would cost to get a machine
with a little more juice to it. I started thinking what a shame it was to
have to keep upgrading to the latest and greatest machine in order to
squeeze out some more clock cycles for big builds. And then I started
thinking about Amazon EC2.

The idea behind EC2 is that Amazon runs virtual machines for you, and you
pay them by the hour for their use. Use of an "Extra Large High-Memory
Instance" running RHEL would run $0.63 an hour. It's the new spin on the
old "timeshare the supercomputer" concept. This seems like a handy way to
get a beast of a computer when I need to do big builds and/or long
simulation runs without having to keep upgrading my core machine.

Has anyone tried this out to see if it works in practice? http://
moxielogic.org/blog/?p=450 says he had serious trouble with it, but that's
from a year ago. Ideally, anyone with experience trying this with Quartus,
but ISE, Modelsim, Rivera, etc would all be interesting.
Click to expand...
You can provide some feedback here, http://bit.ly/rFONhB, that you would
like cloud FPGA P&R.
 
Hi,

Rob Gaddi skrev 2011-12-07 18:54:
I was waiting for Quartus to finish crunching my latest build, and poking
around idly on NewEgg trying to see what it would cost to get a machine
with a little more juice to it. I started thinking what a shame it was to
have to keep upgrading to the latest and greatest machine in order to
squeeze out some more clock cycles for big builds. And then I started
thinking about Amazon EC2.

The idea behind EC2 is that Amazon runs virtual machines for you, and you
pay them by the hour for their use. Use of an "Extra Large High-Memory
Instance" running RHEL would run $0.63 an hour. It's the new spin on the
old "timeshare the supercomputer" concept. This seems like a handy way to
get a beast of a computer when I need to do big builds and/or long
simulation runs without having to keep upgrading my core machine.

Has anyone tried this out to see if it works in practice? http://
moxielogic.org/blog/?p=450 says he had serious trouble with it, but that's
from a year ago. Ideally, anyone with experience trying this with Quartus,
but ISE, Modelsim, Rivera, etc would all be interesting.

One other drawback is that you will have to pay per P&R or simulation
Click to expand...
run, this was how it was in the minicomputer era(VAX and more) then came
Unix workstations suddenly you did not have to think about the cost for
a simulation run it was your own desktop workstation.

No, I think flatrate(you pay for your own box and run) is best, you will
not have flat rate in a elastic clound :)

You can pay on demand for licenses but that is not nice either!

Just my experiences

/michael
 
Wasn't there a short lived push by EDA companies about 10 years ago to
provide internet services in which the number crunching was done by their
server while you entered and submitted your design using their software? I
seem to remember that it was back in the day when we paid huge amounts of
money for complete design and build tools and did it all ourselves. It was
the pay as you go model. Perhaps the need has resurfaced. It seems to me if
you can get a build 10 times faster for a buck or two a run, it'll be cost
effective.

JJS

"Rob Gaddi" wrote in message news:jbo99g$652$1@dont-email.me...

I was waiting for Quartus to finish crunching my latest build, and poking
around idly on NewEgg trying to see what it would cost to get a machine
with a little more juice to it. I started thinking what a shame it was to
have to keep upgrading to the latest and greatest machine in order to
squeeze out some more clock cycles for big builds. And then I started
thinking about Amazon EC2.

The idea behind EC2 is that Amazon runs virtual machines for you, and you
pay them by the hour for their use. Use of an "Extra Large High-Memory
Instance" running RHEL would run $0.63 an hour. It's the new spin on the
old "timeshare the supercomputer" concept. This seems like a handy way to
get a beast of a computer when I need to do big builds and/or long
simulation runs without having to keep upgrading my core machine.

Has anyone tried this out to see if it works in practice? http://
moxielogic.org/blog/?p=450 says he had serious trouble with it, but that's
from a year ago. Ideally, anyone with experience trying this with Quartus,
but ISE, Modelsim, Rivera, etc would all be interesting.

--
Rob Gaddi, Highland Technology -- www.highlandtechnology.com
Email address domain is currently out of order. See above to fix.
 
Wasn't there a short lived push by EDA companies about 10 years ago to
provide internet services in which the number crunching was done by thei

server while you entered and submitted your design using their software?
I
seem to remember that it was back in the day when we paid huge amounts o

money for complete design and build tools and did it all ourselves. I
was
the pay as you go model. Perhaps the need has resurfaced. It seems to m
if
you can get a build 10 times faster for a buck or two a run, it'll be cos

effective.

JJS
Click to expand...
Yes, It ended when all of their customers told them that they would would
never let them see any of their IP.

The problem with running in the cloud is that they have to create a imag
and
run it with root access open to the world so that the customer can contro
it.

Well so can anyone else who can guess it's password. It's really nice i
root
has to be sitting by the box in order to get it.


John Eaton


---------------------------------------
Posted through http://www.FPGARelated.com
 
Hi,

jt_eaton skrev 2011-12-09 05:46:
Wasn't there a short lived push by EDA companies about 10 years ago to
provide internet services in which the number crunching was done by their

server while you entered and submitted your design using their software?
I
seem to remember that it was back in the day when we paid huge amounts of

money for complete design and build tools and did it all ourselves. It
was
the pay as you go model. Perhaps the need has resurfaced. It seems to me
if
you can get a build 10 times faster for a buck or two a run, it'll be cost

effective.

JJS



Yes, It ended when all of their customers told them that they would would
never let them see any of their IP.

The problem with running in the cloud is that they have to create a image
and
run it with root access open to the world so that the customer can control
it.

Well so can anyone else who can guess it's password. It's really nice if
root
has to be sitting by the box in order to get it.


John Eaton

That you dont have to do, why do you say so?
Click to expand...


/michael
 
You must log in or register to reply here.

Welcome to EDABoard.com

Sponsor

Back
Top