On Jan 3 I had knee surgery--patellofemoral, not a total replacement. I'm doing well with that, and I get the medical staples out tomorrow.

I'm tapering off the pain medications. Either I don't need them as much, or ... whatever. There is only so much television, so much gaming, so much mindless stuff one can do before the brain rebels. Either that, or goes completely asleep. Since it's easier to exercise the brain than to wake it up from complete dormancy, I thought I would work a bit on a project; someone at work would like their data analyzed using unsupervised methods, and the laptop I'm currently using (mid 2015 MBP, Retina, 16GB) has been giving me problems since I brought it home. It's used, and I suspect it was abused a bit prior to it going to the reseller. It has been restarting during the night when no one is using it; the fans start blazing away for no apparent reason (as it's been sitting unused for hours); it occasionally freezes.

I have been jonesing for a M1 16" MBP, M1Pro (not Max), and 32 MB integrated RAM. I can find the 16 MB version locally (I can get one today from Worst Buy), so the question becomes, do I need the increase in RAM? Have I chosen an appropriate machine for what I'm wanting to do? For me to order a new M1 with these specifications will mean the the machine will take about 6 weeks to get to me. I chose the 16" because screen real estate is precious, and I can take it with me wherever I go.

A bit more information may be needed: unsupervised methods of analysis on large data sets uses algorithms to find data that is similar to one another; such data will cluster together, as in this one is positive for this attribute or attributes, but negative for these others and it is similar to this other data point. It will point out hidden structure found in the data. At work, the data the researchers generate is normally looked at via one parameter plotted against another, and the data that fits the "preconceived" criteria is selected, and the process continues with another plot of one parameter against another using the selected data, etc. I say "preconceived" because the researcher will say "I'm looking at the data that fits this criteria and ignore the rest." This manual method is highly susceptible to bias, particularly as the number of parameters grow. The instrument I normally operate will detect up to about 16 parameters, including time, and the newest instrument in the lab will detect (I think) up to 43 parameters (including time).

For the algorithms to do what they need to do, the data needs to be cleaned up (known extraneous and confounding data need to be removed from the data set. As I will be working with data generated from blood samples as applicable to particular pathologies, for instance, dead cells and cellular debris can be eliminated), the data can then be dimensionally arranged and the algorithms let loose to do their work. Dimensionally arranged, you say? Well, think of three dimensional space, with an x, y, and z directions. Add time to that model, to get you to four dimensional. The brain begins to boggle at this idea, and one really cannot imagine it clearly. Now add more dimensions, one for each thing that is being measured. Now you have multidimensional space, and the complexity of the space (its dimensionality) increases with increasing numbers of parameters being measured.

The only time that a visual is usually needed is at the end, to see what the algorithm(s) have found. Thus, it is similar to graphical work only in that both will use large amounts of data.

This is what I'll be working with.

I think 16gb would be enough, but what do I know…
What kind of software will be doing the analysis? What do the developers say?
Glad you’re getting better, keep it up!



Edited 1 time(s). Last edit at 01/17/2022 03:54PM by rich in distress.

How well does your old machine work for what you do?

Its really hard, from your description, to tell what your hardware requirements are. I can tell you that because of the faster bus additional ram isn't as important as its been in the past.

The Max chip has 2x the bus speed. Its only $200 more when comparing 32GB ram configs.

---

30 day money back garauntee... but think you would be totally fine.

---

Edited 999 time(s). Last edit at 12:08PM by jdc.



Edited 1 time(s). Last edit at 01/18/2022 02:31PM by jdc.

Not sure about your software. The Max will give you 2x the memory bandwidth, which sounds like it would be useful.

What software are you using.

Quote
mattkime
How well does your old machine work for what you do?

Its really hard, from your description, to tell what your hardware requirements are. I can tell you that because of the faster bus additional ram isn't as important as its been in the past.

The Max chip has 2x the bus speed. Its only $200 more when comparing 32GB ram configs.

Diana, it's actually pretty simple to determine if 16GB will be enough in an M1Pro: is it enough in the 2015 MBP you're using now?

If yes, then because of the way the M1 chips are designed, it will be more than enough (a rule of thumb is an M1 behaves similarly to an Intel machine with double the RAM).

It might be more difficult to determine if the problems you're having with the 2015 are RAM related. Activity Monitor will be helpful. But bottom line is... those problems you describe--do they only occur when your data analysis software is running?

---

What makes this difficult to determine are the following:

1. The manufacturer of the software I will be using doesn’t really say what the minimum specifications are for their software. Yeah, not real good nor real fun. When trying to see if the latest iteration would run on my current setup, the only thing I could tell is that some folks have trouble using it with Big Sur, and they could only say definitively for use of the next OS down. There are no minimum RAM numbers; don’t know why but can guess that it all depends on how intensive your needs. It would run best in the old Intel world by feeding it as much RAM as you could manage. No guarantees though, because if it ran slowly (or choked) it could be the RAM, could be the processor … Yeah, a bit of talking in circles. The software was recently acquired by a big name company, and I’m hoping that this isn’t the beginning of “Welcome to the Windows World!” FWIW, my current setup will run the basic software, but … what I am proposing is a lot more than basic.

2. The algorithms have been optimized better and thus they should run. The last time I tried it, the data needed to be about 5,000-10,000 clean events and even then it would take 15-20 minutes to churn through it. The data sets I have will be larger than this. Data is data—it shouldn’t matter if this was photography, film, or whatever. The same algorithms that are used to smooth noise, recognize shapes and/ or colors, etc. will work. Granted, there may be slight modifications because I don’t want to see the result of each manipulation of each data point, the basic algorithms are the same. I was hoping that folks could give me indications of what they have done in the past, and what was suitable (from a purely data standpoint) and try to apply that to the new hardware. Which leads me to the next point.

3. As far as the problems I have with the current setup, it tends to choke and stall when I try to do much labor intensive work. I shut everything else down, and it still has problems. I have taken to chopping the job up. Like I said previously, I have put it to sleep for the night only to come back to it the next day and find that it panicked and rebooted; or put it to sleep, only to hear the fans start ramping up a few hours later. Investigation shows that the system is maxed out, both in cpu cycles and available RAM. Nothing is actively running that I know of. I’ve not had problems like this before. And I have not yet tried it using these larger data sets.

4. Some software packages are cloud based. I have several issues with this, but the biggest one is that the data doesn’t belong to me but rather either to the researcher that heads the project, or perhaps even to the Foundation where I work. Since it’s most likely from human as opposed to mouse or some other model, there is a whole other layer on top of it (HIIPAA, anyone?) So no, I cannot freely upload the data to some server and have it crunch away.

I hope this helps. The acquisition of the 2015 MBP was a stop gap measure, as the previous machine to it started showing the dreaded video issues. By going as old as the 2015 I should have skipped the keyboard issues that plagued the newer ones. Seems I just traded one set of problems for another.

I'm about as far from an expert as there is. But, I'm very familiar with running very big data sets that slow down pc's. I used to go to a computer lab at night and run different sets on a dozen machines at a time, moving back and forth among the computers. I had a much more powerful machine in my office that would run the sets but it was ultimately faster to go to the lab.

Is there a remote possibility of borrowing what you're thinking of to just run a set and seeing how it goes? And, of course jdc's point of a 30 day return is worth considering. If neither of those work, I'd wait the 6 weeks and get the increased RAM. But, again, this is from somebody who doesn't really know...

Its very possible that the hardware requirements are determined by your data set. As Michael mentioned, the best way to know what you need is to run the software on it.

You might find a local Mac seller that would let you try before you buy.

---

I’d rather begin now instead of 6 weeks. Your knee will keep you with low mobility anyway (how’s that going?).
The RAM paradigm is very different with M1.
Yes, less space may be available, but it will move in and out much faster.
That leaves the ball on the processor/software match.

Some data science discussion and links in a macrumors thread.

Best wishes on the knee.

---

“There comes a point where we need to stop just pulling people out of the river. We need to go upstream and find out why they’re falling in."

You could use Activity Monitor on the older MBP to gauge how much RAM your program(s) require, and go from there. As was said, if 16GB was enough then it will be more than enough on the M1, however if you expect to be handling larger data sets going forward then I would lean 32GB on the M1.

I just did a quick check and while the 32GB version appears to be back ordered, if you have money to burn then B&H Photo has the 64GB version in stock.

Edit: I also see a 32GB model in stock at Adorama, so check around and maybe you can find what you want at a reseller. Also, maybe a local Apple Store has what you want in stock.

---

Edited 1 time(s). Last edit at 01/18/2022 01:30PM by AllGold.

Quote
Michael
I'm about as far from an expert as there is. But, I'm very familiar with running very big data sets that slow down pc's. I used to go to a computer lab at night and run different sets on a dozen machines at a time, moving back and forth among the computers. I had a much more powerful machine in my office that would run the sets but it was ultimately faster to go to the lab.

Is there a remote possibility of borrowing what you're thinking of to just run a set and seeing how it goes? And, of course jdc's point of a 30 day return is worth considering. If neither of those work, I'd wait the 6 weeks and get the increased RAM. But, again, this is from somebody who doesn't really know...

Yeah, I used to do something similar in grad school: get one set of data ready to go to the next step, then run the next step on a different computer as I was continuing the first step with another part of the data, and when step 2 was finished on the first part of the data, go to the next step forward with a third computer. The real problem was if I got side tracked for any reason trying to remember where I was in the data. It wasn’t true multitasking, but what is? Otherwise it took forever to get through everything.

I keep checking various sites as far as availability. The local AS has been out of stock for everything but the most basic systems for a while now. Of course, I would like to do business somewhere I could get some kind of discount, as $3K+ for any kind of system rather takes my breath. Sales tax here is between 8 and 9%. The sig other squawks about prices, so he’s promoting waiting until they show up on the refurbished Apple site. By the time that happens, I will be completely back to work and this becomes a moot point, as I won’t have time to do it.

Update on the knee: doing well. The staples are now out. I have a script for PT, so that can get started. I tried to get PT before now to come out to the house, but only heard crickets. As it’s the right knee, driving is out of the question for now, but getting in and out of the car is getting easier.

Memory is shared between the CPU and GPU on an ARM Mac.

If your software is the slightest bit optimized to use a GPU ("similar to graphical work") then unhesitatingly go for more RAM, and given that the Max has more memory-bandwidth you should strongly consider the Max. (Note that the 24-core custom build Max is a little cheaper than the 32-core default Max.)

If your software is a kludge that barely runs and is seldom updated/patched then the Pro will probably run it just as well as the Max under Rosetta2.

Getting a bit off-topic: Have you heard of any problems working with this software and those databases on solid state media? Some of my scientist friends and clients have to work from hard drives because manipulating their large data-sets involves massive write-operations that wear out solid state drives very rapidly. (SSDs have limited write-operations usually measured in Petabytes, so this is not a concern for the average user.) Might not be a big deal in your case, but your description rang some bells.

---

Price comparisons:

[prices.appleinsider.com]#

Speed tests:

[browser.geekbench.com]

---

Saint Cloud, Minnesota, where the weather is wonderful even when it isn't.