Intel® Fortran Compiler
Build applications that can scale for the future with optimized code designed for Intel® Xeon® and compatible processors.
Announcements
FPGA community forums and blogs on community.intel.com are migrating to the new Altera Community and are read-only. For urgent support needs during this transition, please visit the FPGA Design Resources page or contact an Altera Authorized Distributor.

IVF vs CVF (speed)

lklawrie1
Beginner
3,510 Views
I've been comparing the execution speed of IVF vs CVF (6.6C or B).
It looks like IVF is nearly twice as slow as CVF for the same set of test files. (IVF -- 2 hours, CVF 1 hour).
Is this what others are seeing?
Linda
0 Kudos
33 Replies
lklawrie1
Beginner
2,226 Views
I forgot to mention -- both are compiled "optimized for speed".
Linda
0 Kudos
TimP
Honored Contributor III
2,226 Views
Several other things you "forgot to mention:" which CPU are you compiling for, and executing on, with each compiler. Which compiler version, what type of code. I might believe such a result, if you are running on PII or P-Pro, as ifort 8 has dropped the ability to optimize for them.
0 Kudos
lklawrie1
Beginner
2,226 Views
CVF - "blend", IVF - Pentium 4 (default "platforms" for both compilers)
Both are being executed on same platform, obviously. A Pentium 4 Mobile processor.
Most other compiler options have been matched as much as possible (IVF project was built from scratch, not from translating from CVF) though CVF has "debug - partial" and IVF has "debug - full" at the moment. I may try the IVF "maximize across project" though not sure that will help.
Size of exes: CVF ~ 12MB, IVF - ~60MB.
Linda
0 Kudos
TimP
Honored Contributor III
2,226 Views
You said "optimize for speed," now you have debug options on. IVF doesn't pretend to give debug runtime performance competitive with CVF partial debug. It's not really very good with any options which prevent vectorization.
0 Kudos
lklawrie1
Beginner
2,226 Views
So, are there any debugging options that can be used with optimizations in IVF? When the inevitable crash happens, it's nice to have some idea though usually we bring those back to the development environment, when they occur for the users.
Linda
0 Kudos
Steven_L_Intel1
Employee
2,226 Views
To answer your original question - no, that's NOT what most people see. Independent benchmarks for a variety of programs show that IVF tends to generate code 5 to 25% faster than CVF. Results vary, though - some tests IVF is twice is fast, others it is somewhat slower.

Half the speed is a very unusual case and I'd appreciate it if you would send us an example at Intel Premier Support. Please ask that it be assigned to Steve Lionel.

I'm not saying it's impossible, but it seems rather unlikely given what I, and our customers, have seen.
0 Kudos
lklawrie1
Beginner
2,226 Views
Thanks, Steve, for the offer. Last time I tried to submit my whole project to the Developer Services (to find a bug in IVF), it was too big for them.And it's grown since them -- very difficult to "trim".
How I am "timing" these is to use the built-in "Elapsed Time" F90 procedure. As an example, I have one test file thatbuilt with CVF, is
4 min 16 sec.
the same test file, with IVF (max speed, debug "lines") (.exe size ~33MB) is
5 min 45 sec.
Conditions during the run (i.e. other programs running, machine being used) were the same (no other programs running, machine essentially dedicated to running that item).
I think you would need the full projects, yes? Not just the exe and inputs?
Linda
0 Kudos
lklawrie1
Beginner
2,226 Views
I should have added with IVF (max speed but full debug) that same file was 9 min 10 sec.
LKL
0 Kudos
Steven_L_Intel1
Employee
2,226 Views
Yes, we would want to see the whole project. How big is it zipped?

If it is too big to attach, say so in the report and we can tell you how to put it in an FTP drop area
0 Kudos
lklawrie1
Beginner
2,226 Views
Do you want both source and object in the zip file or just source, project files, and files to run with? It will probably be around 4MB zipped, maybe less, with the source files. Probably quite a bit more if I included the object folder.
Linda
0 Kudos
Steven_L_Intel1
Employee
2,226 Views
Just what is needed to build and any data files - we can build it. 4MB should be no problem to attach to a case.
0 Kudos
lklawrie1
Beginner
2,226 Views
I've put in the support request (258976) but I cannot get the zip file uploaded (it's only about 3MB as it turns out). I've tried about 4 times and will try again in a moment.
LKL
0 Kudos
Steven_L_Intel1
Employee
2,226 Views
Linda,

If you don't manage to get it uploaded, I'll contact you tomorrow with an alternate method.

What happens when you try to upload through Premier Support?
0 Kudos
lklawrie1
Beginner
2,226 Views
The little popup window appears (the one that says it will close when finished). At the bottom of the main web page (turned to the issue number), the bar progresses during the upload but at the end, nothing happens.
Sooner or later, the system times out I think.
Linda
0 Kudos
peter_cool14
Beginner
2,226 Views
Quoting - lklawrie
I've been comparing the execution speed of IVF vs CVF (6.6C or B).
It looks like IVF is nearly twice as slow as CVF for the same set of test files. (IVF -- 2 hours, CVF 1 hour).
Is this what others are seeing?
Linda

0 Kudos
peter_cool14
Beginner
2,226 Views
Quoting - pgalimutti
I have a problem something similar, maybe little serious. I've been using CVF in my Toshiba laptop which is a Intel P4 processor(2 gb ram, 80 gb hard disk and 32 bit). Now I've got a new HP desktop which is more advanced(8 gb ram, 500 gb hard disk and 64 bit). I am not sure if it is P4 , but its Intel Core2 Quad. Coming to the point, I use a massiveCFD codewhich has6568 lines . I ran the code in my laptop and the speed ispretty decent. Since CVF doesnt support 64 bit I've to buy IVF. As an initial test I ran asample which took more than 5 mins in my desktopwhereas ittakes less than a minute in my laptop. Thats just unbeleivable, I was shocked,it got on my nerves and very disappointing. My goal is to develop a more complicated CFD code on my desktop computerand get faster results (Iam aware of my limitations for not using parallel processors or super computers since my code will be just upto the point). Unfortunately this project can not be sent because this is a private code. At the most I can try sending a .exe file. Can somebody have an answer for this??


0 Kudos
peter_cool14
Beginner
2,226 Views
Quoting - lklawrie
I've been comparing the execution speed of IVF vs CVF (6.6C or B).
It looks like IVF is nearly twice as slow as CVF for the same set of test files. (IVF -- 2 hours, CVF 1 hour).
Is this what others are seeing?
Linda

0 Kudos
peter_cool14
Beginner
2,226 Views
Quoting - pgalimutti
I have a problem something similar, maybe little serious. I've been using CVF in my Toshiba laptop which is a Intel P4 processor(2 gb ram, 80 gb hard disk and 32 bit). Now I've got a new HP desktop which is more advanced(8 gb ram, 500 gb hard disk and 64 bit). I am not sure if it is P4 , but its Intel Core2 Quad. Coming to the point, I use a massive CFD code which ha about 6550 lines. It normally takes a minute or two to run on my laptop which has CVF 6.6. Since CVF doesnot support 64 bit I have to buy IVF 11.0 in my new desktop. Initially as a test I ran the same file and to my shock it took 5 mins to run the same code. That is really disappointing. This code was written in fortran 77 version. Is there a way to get around this?
Unfotuantely this is a private code and can not be transfered.

Peter


0 Kudos
Steven_L_Intel1
Employee
2,226 Views
What compiler options are you using? Go to the Command Line property page and copy the options shown there then paste into a reply.

You are building a release configuration, yes?

It may well be that the performance is limited by something other than compiler optimization. If you have a dual-core system, look into the parallel processing options.
0 Kudos
peter_cool14
Beginner
2,169 Views
What compiler options are you using? Go to the Command Line property page and copy the options shown there then paste into a reply.

You are building a release configuration, yes?

It may well be that the performance is limited by something other than compiler optimization. If you have a dual-core system, look into the parallel processing options.

0 Kudos
Reply