Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Why do you think a Fortran compiler is required to accomplish a native build of Mathematica? Wolfram has had the Wolfram Player running on iOS/iPadOS since 2017. That was a port to the very same instruction set that Apple is running on the M1 processor. How did they get that to run, and why wouldn't they do exactly the same thing to build for macOS?
This is getting outside my wheelhouse but, as I understand it, Mathematica makes use of numerical libraries written in Fortran, including BLAS and LAPACK (see https://mathematica.stackexchange.c...entation-used-internally-for-numerical-matrix). Currently, they use the Intel MKL implementation, but that (presumably) won't work on AS.

Thus I concluded that, in producing a native AS build of MMA, WRI would need to to compile those Fortran libraries, which means they would need an AS Fortran compiler. But if I'm incorrect about this, I'd be happy to hear from someone that understands this better than I do.
 
Currently, they use the Intel MKL implementation, but that (presumably) won't work on AS.

Did you understand the question I asked? Theo Grey demonstrated a CDF player on the iPad way back in 2102:


And the full-blown Wolfram Player has been available for iOS since 2017.

There are no limitations of computable Wolfram expressions noted for this product. It clearly runs on Apple's iPhone/iPod Touch/iPad silicon implementations -- the same instruction set implemented on the M1 chip. An executable to provide that LAPCK functionality has clearly been available on those platforms for years. Why couldn't that same code base be used for their Universal 2 build?

The Stackexchange link you provided does not answer that question. It's an old inactive discussion.
 
Did you understand the question I asked? Theo Grey demonstrated a CDF player on the iPad way back in 2102:


And the full-blown Wolfram Player has been available for iOS since 2017.

There are no limitations of computable Wolfram expressions noted for this product. It clearly runs on Apple's iPhone/iPod Touch/iPad silicon implementations -- the same instruction set implemented on the M1 chip. An executable to provide that LAPCK functionality has clearly been available on those platforms for years. Why couldn't that same code base be used for their Universal 2 build?

The Stackexchange link you provided does not answer that question. It's an old inactive discussion.
I thought I understood your question, and I answered it the best I could. And while the last activity on that link is 2 years old, the information in it—that Mathematica uses BLAS and LAPACK—should still be current.

As for Wolfram Player, I didn't think it was relevant, because while it has many of the the computational abilities of Mathematica, it doesn't have all of them. As installed on my Mac, Wolfram Player is 3 GB, while MMA 12.1.1 is 12 GB. According to the Wolfram website, Player contains ~3,000 functions, while Mathematica contains ~5,000.

Indeed, if Wolfram Player contained all of MMA, then one could ask why, if WRI could provide Wolfram Player for iOS, why has it not provided MMA for iOS? I believe the answer is that there are signficant differences between the two.
 
  • Like
Reactions: Spectrum
And the full-blown Wolfram Player has been available for iOS since 2017.

There are no limitations of computable Wolfram expressions noted for this product.
That is incorrect. Wolfram Player is not the same as Mathematica, it only works with existing CDF, it does not support the creation of new content. Both Mathematica and MATLAB use Intel MKL. Mathworks plans to update their product to run via Rosetta 2 in the next month or so, but not natively. It will be interesting to see if the performance hit will be inline with other apps.
 
Apple does not need large cooling fans in their Mac. They are happy just to let it burn slowly at a 100 degrees.

Gosh, reminds me of the "early adopter" times of the first Multi-Processor Mac machine. My PowerMac 9500/180MP crashed 20 times a day using Adobe's Photoshop MP-Plugin. Over 20 times a day! Since that time, I know. Never buy the most expensive early adopter hi-tech stuff from Apple. It's not worth it.




Encoding audio and video can heat up your processors a lot.

Regards,

BLACK BARON
Design / Photo / Art / Music
 
  • Haha
Reactions: Babygotfont
I thought I understood your question, and I answered it the best I could. And while the last activity on that link is 2 years old, the information in it—that Mathematica uses BLAS and LAPACK—should still be current.
The Stackexchange discussion you referenced said that Eigensystem[] was part of the Wolfram code that relied on the FORTRAN code. In other words, if Eigensystem[] runs on the iOS CDF Player, then Wolfram already has implemented that FORTRAN code to run on Apple's Axx processors.

This is something easy to check. Have you tested out if your Stackexchange FORTRAN theory makes sense by creating a simple test CDF file?

As for Wolfram Player, I didn't think it was relevant, because while it has many of the the computational abilities of Mathematica, it doesn't have all of them. As installed on my Mac, Wolfram Player is 3 GB, while MMA 12.1.1 is 12 GB. According to the Wolfram website, Player contains ~3,000 functions, while Mathematica contains ~5,000.
Reference? Where are you getting this 3000/5000 functions from? Where on the Wolfram website is the chart? When does it tell you? I have written many CDF files in Mathematica; I've never seen a compatibility warning.

Note: The Wolfram FAQ about CDF says, "Almost all Wolfram Language functions can be incorporated into CDFs". There is absolutely nothing about 3000/5000 functions available. Please don't make up stuff.
Indeed, if Wolfram Player contained all of MMA, then one could ask why, if WRI could provide Wolfram Player for iOS, why has it not provided MMA for iOS? I believe the answer is that there are signficant differences between the two.
What significant [technical] differences? Do you have any evidence to support this? Mathematica 12.1 runs on a dinky little $35 Raspberry Pi; why in heaven's name do you doubt it would run on a far more capable A12-or-better Apple processor?

I think it's a marketing/financial decision: Wolfram doubts they can get ~$1500/year for a tablet version of Mathematica -- as they can for businesses laptop access.
That is incorrect. Wolfram Player is not the same as Mathematica, it only works with existing CDF, it does not support the creation of new content.
Nonsense. Read the CDF FAQ. This is a marketing decision rather than a technical one. The dinky little $35 Raspberry Pi can be used to generate CDF files.

@theorist9: note that somebody at WRI figured out how to compile the FORTRAN code to the RISC Raspberry Pi platform.
 
Last edited:
Note: The Wolfram FAQ about CDF says, "Almost all Wolfram Language functions can be incorporated into CDFs". There is absolutely nothing about 3000/5000 functions available. Please don't make up stuff.
Hey what happened? We were having a civil, collegial discusion, and suddenly you've gone off the deep end and are hurling a nasty insult. Specifically, for some bizarre reason you've decided to falsely accuse me of deception ("don't make up stuff"). What really frosts me about your mischaracterization is that I was very careful in responding to your post—I took the time to double-check the figures on Wolfram's website before posting.

Anyways, if that's the way you want to play it, fine. I'm not going to be civil with someone who chooses to behave in that manner. You asked for it, so I'm taking the gloves off:

You are completely clueless about what the Wolfram site does or doesn't say about the number of functions in Wolfram Player and Wolfram Mathematica:

Wolfram Player (source: https://www.wolfram.com/cdf/faq/ )

1606180648517.png


Wolfram Mathematica (source: https://www.wolfram.com/mathematica/)
1606180070206.png

Yet, reveling in the glory of your ignorance, rather than recognizing your cluelessness, you proceed to insult people who know far more than you. I suppose you also think NASA faked the moon landing, and accuse the rest of us of making stuff up about that as well.

And BTW, you could have easily found both of these by Googling:

site:wolfram.com mathematica 5000
and
site:wolfram.com player 3000

But then again, had you to competence to do that, you would have known enough not to pen your last post.

I'm done. Please don't interact with me any more. And please think twice before pulling this crap on the next guy you interact with.
 
Last edited:
  • Wow
Reactions: FloatingBones
Nonsense. Read the CDF FAQ. This is a marketing decision rather than a technical one. The dinky little $35 Raspberry Pi can be used to generate CDF files.
We are talking about Wolfram Player vs Mathematica. Clearly, you do not know the difference. From Wolfram Player Pro FAQ:

Can I create new content using Player Pro? No. If you're making new content, you need Mathematica or other Wolfram Language-based products.
 
Hey what happened?
You started spreading disinformation about the Wofram products. Wolfram's CDF Player uses exactly the same engine -- and the same functions -- that Mathematica uses. Virtually anything you can do in Mathematica you can publish in a Computable Document Format file. This is a great gift -- a great way to spread computations and visualizations. Wolfram gave us another great gift: the ability to both run Mathematica and generate those CDF files on Raspberry Pi computers.

You think there's a 3000-out-of-5000 limitation in that CDF functionality, but that is fiction.
We were having a civil, collegial discusion
We still are. And, as a matter of fact, you made up something about the CDF Player. You assumed that the current CDF Player only has ~3,000 functions.

You are completely clueless about what the Wolfram site does or doesn't say about the number of functions in Wolfram Player and Wolfram Mathematica:

Wolfram Player (source: https://www.wolfram.com/cdf/faq/ ) [SNIP]

You failed to notice: that 3,000 functions answer in the FAQ page goes back to October 24, 2011. If you had looked up the FAQ page in the Internet Archive, you would have seen it. And, for the record, the FAQ says "more than 3,000 functions". In 8/2011, Wolfram 8.0 was the current version, and there were far fewer functions. Look:

FunctionGrowthfixed.gif

I bet that Wolfram Research would be astonished what you did with that nine year old answer from their FAQ!
I'm done.
I hope not! I hope you acknowledge your error and continue the conversation.

CDF and Mathematica are in lock-step with each other. Virtually anything you code in a Mathematica you can publish to be run in the computational viewer. If there were a 3000/5000 discrepancy, that beauty and simplicity of CDF publishing would be impossible. That's how anyone familiar with CDF knew your "only 3000 functions [in current CDF Player]" claim was wrong. Do you now understand?
 
You started spreading disinformation about the Wofram products. [..] that is fiction.

[.. you made up something about the CDF Player. You assumed that the current CDF Player only has ~3,000 functions.

You failed to notice: [..] If you had looked up the FAQ page in the Internet Archive, you would have seen it.

I bet that Wolfram Research would be astonished what you did with that nine year old answer from their FAQ!

[..] I hope you acknowledge your error and continue the conversation.

[..] Do you now understand?
Do you always talk like this, or only on Tuesdays?
 
  • Haha
Reactions: FloatingBones
Do you always talk like this, or only on Tuesdays?
Gosh. I don't know. This situation hasn't happened before.

The whole point of CDF is that the runtime environment be 100% compatible with the Mathematica engine. Write it in Mathematica, run it on any platform running the same version of CDF. I believe Wolfram chose the name "CDF" to imply a similarity to PDF for publishing documents.

When @theorist9 claimed that only 3000 of 5000 Mathematica functions were available in [the current] CDF distributions, I asked how he came to that conclusion. He cited the CDF FAQ, which says "CDF Player provides essentially all of Mathematica—more than 3,000 built-in functions [....]". He/she failed to notice that statement from the FAQ was published back in October 2011 (or sooner). Theorist9 claimed:
And BTW, you could have easily found both of these by Googling:

site:wolfram.com mathematica 5000
and
site:wolfram.com player 3000

Both of what??? This is the google equivalent of the practice of using dunking to determine if someone is a witch. Makes no sense at all.

I regularly use Mathematica to generate CDF files, and I run them on a variety of platforms. It's a fantastic way to publish computational visualizations. Having someone claim that what I literally do all the time strikes me as bizarre. The whole point of this tool is WORA; it would make no sense for 2000 of the 5000 functions to be unavailable. I'd fondly hope anyone using these tools would already understand that. So, no, I can't answer your question. :)

As an aside, I think the Raspberry Pi 400 is one humdinger of an learning platform. It's ready to plug and play, and you can configure it as a VNC client to access on the network. There are all sorts of visual-oriented education software, and the GPIO panel facilitates all sorts of electronic experiments. Wolfram Mathematica is part of the NOOBS distribution, and you can cut CDF files to run on PCs, Macs, iPhones, and iPads. They even improved the thermals on this device to crank up the clock speed of the original Raspberry Pi 4. I highly recommend it as a gift.

kit-lg@2x-5e3ad81dda9d89e594ad356da33f9bac.jpg
 
This is getting outside my wheelhouse but, as I understand it, Mathematica makes use of numerical libraries written in Fortran, including BLAS and LAPACK (see https://mathematica.stackexchange.c...entation-used-internally-for-numerical-matrix). Currently, they use the Intel MKL implementation, but that (presumably) won't work on AS.

Thus I concluded that, in producing a native AS build of MMA, WRI would need to to compile those Fortran libraries, which means they would need an AS Fortran compiler. But if I'm incorrect about this, I'd be happy to hear from someone that understands this better than I do.
 
No mention at all yesterday of NVIDIA blowing the competition away with their new Ampere RTX 3000 series lineup, while Apple is about to transition to in-house graphics across the line.

I have no doubt Apple can compete on CPUs in the desktop space - but common, them dropping AMD graphics from Apple Silicon macs and not working with Nvidia due to a past feud is just going to cripple them in the graphics space. Even the new PS5 and Xbox Series X look pretty pathetic compared to these new chips.
graphics on Apple silicon and neural engine and shared ram are great. If you are talking gaming, we’ll Apple has not really entered that market. EGPU support would be great, is that software enabled?
 
Process comparisons based on the name only make sense when comparing processes from the same vendor (i.e. Samsung, TSMC, Intel). Intel could have called this process "1nm" if they wanted to, instead they called it 10+++ (or something like this)
Die size. I literally do not understand your meaning. Intel is far behind on nodes, hence the high power requirements. You make them sound like they are leading by years not lagging by years
 
Die size. I literally do not understand your meaning. Intel is far behind on nodes, hence the high power requirements. You make them sound like they are leading by years not lagging by years
Die (chip) size has nothing to do with the process. More advanced processes have better device/metal density so at the same die size they may fit more devices. With that in mind, take into account that density wise Intel 7nm process is believed to be between TSMC 3nm and 5nm (however, while TSMC 5nm is in production, Intel 7nm is not yet). So the names mean nothing. For more detailed information, read this very good article.
 
Die (chip) size has nothing to do with the process. More advanced processes have better device/metal density so at the same die size they may fit more devices. With that in mind, take into account that density wise Intel 7nm process is believed to be between TSMC 3nm and 5nm (however, while TSMC 5nm is in production, Intel 7nm is not yet). So the names mean nothing. For more detailed information, read this very good article.
Except density is meaningless to cpu designers. We *never* have a goal of squeezing as many transistors as we can within a given area, except for static RAMs. We care about minimum feature size and minimum spacing. And since you mentioned metal, we don’t care just about min width and pitch, but about aspect ratio (height to width).

And comparing the design rules, Intel’s naming convention puts each intel process at almost exactly where TSMC’s next smaller node is. I.E., Intel 10nm = TSMC 7nm, Intel 7nm = TSMC 5nm, etc.
 
  • Like
Reactions: Zdigital2015
Die (chip) size has nothing to do with the process. More advanced processes have better device/metal density so at the same die size they may fit more devices. With that in mind, take into account that density wise Intel 7nm process is believed to be between TSMC 3nm and 5nm (however, while TSMC 5nm is in production, Intel 7nm is not yet). So the names mean nothing. For more detailed information, read this very good article.
TSMC has been in production of 5nm for close to a year now, while Intel PR keeps on with these bottom of the barrel marketing campaigns, like some jilted prom date. I give them a 50/50 chance of getting their house in order. Perhaps, the fates will be kind and they won’t be able to buy their way out of hole they put themselves in over the past 6 years.
 
Except density is meaningless to cpu designers. We *never* have a goal of squeezing as many transistors as we can within a given area, except for static RAMs. We care about minimum feature size and minimum spacing. And since you mentioned metal, we don’t care just about min width and pitch, but about aspect ratio (height to width).

And comparing the design rules, Intel’s naming convention puts each intel process at almost exactly where TSMC’s next smaller node is. I.E., Intel 10nm = TSMC 7nm, Intel 7nm = TSMC 5nm, etc.
As far as process comparisons are concerned I am going to trust Scotten Jones expertise not yours. You are not a process guy. You listed the obvious important process metrics which have not had anything to do with process name for a very long time (since Intel introduced finfet). Many if them are related to density (at least indirectly). And obviously, ability to squeeze more transistors into the same die is important not just for RAM. Having, say, 12 cores instead of 6 is rather important for some. GPU cores also use a lot of silicon.
TSMC has been in production of 5nm for close to a year now, while Intel PR keeps on with these bottom of the barrel marketing campaigns, like some jilted prom date. I give them a 50/50 chance of getting their house in order. Perhaps, the fates will be kind and they won’t be able to buy their way out of hole they put themselves in over the past 6 years.
We are talking about process naming not about Intel or TSMC PR. If Intel 7nm fits between TSMC 3 and 5nm then that's where it fits regardless of when the process is production ready. Nobody (including Intel) is denying ghat right now TSMC is ahead of Intel just maybe not as far ahead as the meaningless process names might suggest (and some people happen to believe those names perhaps because in the past those names did have very clear relation to the process properties)
 

What's that info. got to do with my post? It only says that MMA 12 can run on AS under Big Sur. It doesn't say it can run natively on AS (it couldn't last time I checked), which is the issue at hand.
 
Last edited:
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.