nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

The Ultimate JMP Machine

nick_shelton

Hi All,

I have a customer who wants to build the "Ultimate JMP Machine". He's specifically using the nonlinear platform to fit functions which can sometimes take time to converge (..and yes I've already shared with him all the methods and techniques in JMP to speed up this process).. My initial recommendation was to just tell him to buy a desktop machine with a 64-bit processor and maxout the installed RAM but I'm now I'm wondering if there's some known hardware configurations that will work better. Will the performance differ between PC and MAC? There's different kinds of RAM, do we know if a particular brand, speed and type is a role? Processor type? Does GPU play a factor?

Any input is greatly appreciated..

Accepted answers

MikeD_Anderson

If someone asks me this I generally tell them that they should think “gaming machine.” Lots of RAM (server or high-performance gaming) and lots of cores on the processor. There are some platforms in JMP that take advantage of multiple threads (the one that I use as an example is Bootstrap Forest). I’m not sure if nonlinear is in that camp. If it is then CPU cores are helpful. Maybe Dan V will chime in, but I don’t think we do any significant GPU multithreading at this point (but I’m not 100% sure on that point). The other thing that would be helpful is SSD’s, but that’s more for opening and saving big files quickly.

Both Mac and PC can hit these points on hardware, but on this point PC would win strictly on bang-for-buck. The Mac Pro’s are really slick, but they are pricy…

Best,

M

All comments

MikeD_Anderson

brady_brady

Nonlinear does support multithreading:

https://www.jmp.com/support/help/Platforms_That_Support_Multithreading.shtml

Byron_JMP

I like a deckedout lenovo workstation. Integration and durability are pretty good.

http://www3.lenovo.com/us/en/landingpage/workstations/?menu-id=Workstations

or if I was going to go completly nuts, I'd add every option to one of these

https://www.apple.com/shop/buy-mac/mac-pro?product=MQGG2LL/A&step=config#

there are some 8th gen processors coming out around Christmas time, and the i9s soon after that. I guess that would be nice too.

Fast memory, fast SSD, and lots of fast cores all make things smoother and likety split fast. At the end of the day a giant virtual machine might still be better?

nick_shelton

Thank you all for the valuable feedback. I'll share all of those options with the user.

Chris_Kirchberg

I use these types of comptures:

https://www.computerhope.com/jargon/e/eniac.htm

Then I take a break here to let it finish computing:

https://tahititourisme.com/en-us/islands/bora-bora/

When I get a call on my phone, I come home to view the results:

https://en.wikipedia.org/wiki/Mobile_phone#/media/File:DynaTAC8000X.jpg

Have a good weekend all.

MikeD_Anderson

@Chris_Kirchberg:

well played, sir!

Jeff_Perkinson

I got a similar question through Tech Support. Here's a response I gave a few years ago that still applies.

The most important factor in determining how much data JMP can work with is memory and more is always better.

Beyond that CPU cores will help for those portions of JMP that are multithreaded but not everything in JMP is (or can be) multithreaded. So, more cores are better than fewer.

Finally, even though JMP holds all the data in memory while we're processing, we do have to read it from disk to get it into memory and to speed up that process you want a fast I/O subsystem. That could be an SSD, RAID or other system. If there are lots of different large datasets to read you'll want to make sure the disk subsystem is as capable as possible.

We do everything through the operating system. We don't address hardware directly so the questions about disk arrays and CPU architectures aren't relevant. A fast Windows machine should run JMP well.

I can point out that the Windows graphical subsystem that we use (WPF) does take advantage of the GPU and more memory there is always better as well but to a lesser extent than the items mentioned above, and again, this is all done through the operating system. We don't access the GPU directly.

Here are a few more points to round this out.

As you know, we can't estimate the size of a dataset based on the number of rows and columns alone. A character column full of single character strings will take less memory than a standard numeric column, and a character column with lots of long strings will take more. JMP offers the Cols->Compress Selected Columns command to try to find the smallest storage format for columns. This can help make JMP faster since there will be smaller memory chunks to work through and move around.

So, in short, the priorities would be to maximize memory, processor speed and cores, and then i/o subsystems, in that order.

MikeD_Anderson

Just to add to @Jeff_Perkinson's last point about compressing numeric columns. JMP 14 added a feature called "Labels to Codes" (a pretty major feature IMO) under Cols > Utilities. This one does the job of John Sall's "Compress Labels to Coded" add-in. You can shrink file size significantly using this utility.