OneAPI 2024.0

[JohnNichols]

The latest update is available on the download site and it will show up on the programs icon on the Control Panel in Windows.  

The tea leaves were wrong and as usual Intel slips it out quietly. 

[JohnNichols]

The base kit install time is 5 hours.  Be prepared to wait. 

[andrew_4619]

Thanks for the tip. Now downloaded an installed over the top of my existing OneAPI install.  Your install time seems totally crazy something must be seriously foobahed.

Download Offline OneAPi Base (2.7Gb)  4minutes

Download HPC Kit offline (1.2GB) 1.5 minutes

Install Oneapi  Base (integration with VS2022) 24 minutes

Install HPC 6 minutes

Total download and install 36 minutes.

Which bits take up the time out of the 5 hours?

The Base kit installer seem a bit smaller the previous versions range between 3.6 and 3.8GB and this one is 2.7

[JohnNichols]

I used the Control Panel and selected change.  The base kit took 4 hours to do the 10 minute integration.   I told you at 10:23 and had been going for 2 hours, I finally went and did a lot of typing, it finished at 1:23 pm.  

I was going to stop it, but it finally worked.  

You need the C++ update, have a look at your installed programs, and now in Control Panel, IFORT shows up as a separate program that did not update with HPC it required a separate update.  

I was surprised, but I would follow your method next time.  

[andrew_4619]

"Settings Apps" only shows the top level base  and hpc kits as being installed not the components on my system. All the parts seem installed and working. 

[mochongli]

true, it installs slowly in vs2022 17.8, previous versions were fine.

[Ron_Green]

Intel has to validate all the downloads, 100s of documentation changes, etc and all online to make sure it's ready to announce.  So things start to appear online before the release announcement.  And not just for the Fortran compiler - it's the entire suite of tools these days.  Not like the old IVF days.  

The announcement will come after all repos, IRC, docs, blogs, etc. are checked live online.  It's one thing to check these before going live, but there is no substitute for putting it live and validating it on the live Internet.  The announcement is the official release.  Just wait, it is coming soon.  We'll announce it here.

And in the past, sometimes they do find last minute problems in one package or another, or a repository does not have the right download, and then we have to pull things offline, rebuild or redocument, and try again.  To quote the title of one of my favorite books, To Engineer is Human. 

My recommendation is to wait for the official release so you do not get a package that is not officially announced. 

IF ( announcement == .TRUE. ) then

   download = supported

   safe_to_download = .TRUE.

ELSE

  download=unsupported       !.. or more precisely, is Schrödinger's cat

  !safe_to_download is undefined, unallocated, unassociated

END IF

[andrew_4619]

I guess announcement attracts a volume of downloads which if the release has some problems the problem gets more extensive. However, if someone wants to download OneAPI they go to the downloads page and the only choice they get is to download and install the one choice that is presented so whilst I understand the reasoning presented it is moot because there is not  an announced/unannounced status just a single download link not a range of downloads available with an attached status such as  "release/ beta/ pre-release etc.

Anyway so far no issues other than a compile error  where there is some new check on the legality of the code. I worked around that in 10 minutes but will check that as I am not 100% sure I  agree with the compiler. I will post a simple example if I decide I am still unsure after further reflection.

[Ron_Green]

Integrations:  the interesting thing is that for our tools, we make a call to the VS API interface to integrate our package.  It's that one call that is taking all the time.  if the VS installation is out of date, or incomplete, the MS VS function call goes back to the mothership to upgrade everything, download missing packages, etc.  And Registry updates, and so while it seems like a simple thing, and under Intel's control, it's really quite out of our hands.  When you think of the billions of operations per second a modern computer can do, it's absolutely mind boggling that this can take hours.  But it's networking, Registry update, and so on.  Very similar to the old days when we waited on drum storage or mag tapes to be loaded. How many of us remember our JCL asking for a tape mount and waiting minutes or hours for the mount, and finally calling the data center to get a poor work study or undergrad to go to the tape library to pull and mount our tape?  I was that work study student, hence I know.  Actually I was a lot faster than an integration call.  Thankfully there were no cell phones and social media to distract me.  

[andrew_4619]

Thanks for that info it explains a lot! I have experienced this long delay in the past! I would imaging it does something via the sledge hammer (lazy) method with longs lists of things being checked against long lists of things with restarts after changes..... 

[JohnNichols]

1.  I could say I am sorry for blowing your secrets, but half the fun in life is outwitting the Man, and Intel is a big Man.  

2. I could stop hunting out odd things, but it is the inquiry of one's mind that solves challenging problems, not that it was challenging, just an amusing minute in a day of tedium of coding.  

3. I like having the latest and also it gives you a check that it works with two friendly characters of dubious origin and but clear motives, who will tell you the truth immediately. 

4. Your secret is safe with me, I will whisper it down a well and just tell one or two select friends on the Intel website.  

5. At this point I ponder how do you keep the countries who are not supposed to have this from having at it?  

6. I will blame Intel, one of your people sent me a note to say I should only use the latest, which started the hunt. 

7. God made me. 

8.  You could tell my mother and she would say something sharp if she was still alive, more likely she would have laughed.  

9. I told my daughter and she laughed and said I was a loser, with affection of course. 

10. There is no ten, aside from well done and splice the main brace. 

11. It works and integrates with the latest VS Preview.  

12. This is the real dozen, not a baker's dozen.

[Devorah_H_Intel]

@JohnNichols  Among so many things you have managed to catch (including the future date of the KB article I have updated), did you manage to notice the change in the HPC Toolkit name?