Are more CPU cores better for huge Excel data sets ?

[Racepro]

I am building a high spec computer as below.
64GB 3200MHz DDR4 RAM
11GB GTX 1080 Ti GPU
1TB SSD PRO SM 961 M.2 Hard Drive
I am in a dilemma which CPU to install
My worksheets consist of up to 800,000 rows
and around 75 columns, most containing formulas.
size around 500-800MB.
No VBA is used only simple to complex formulas.
My initial reaction was to install an i9 14 core CPU
in keeping with the other high spec hardware.
However, from reports I have read, it appears that single
thread performance is almost as important as having multiple cores.
Can anyone shed some light on this for me. At what point does it
become an overkill ?
Thanks
Ian

[David (PassMark)]

See also this old post about the best CPU for Excel
https://www.passmark.com/forum/pc-ha...excel-workbook
Some spreadsheets are limited to single threaded. The only sure way to know is to do some testing with your particular Excel workbook.

I assume you are flexible on the motherboard and power supply, and also don't care about the cost?

For single threaded performance, get the Intel Core i7-8700K.

For multi- threaded performance some of the options would be,
1 x Intel Core i9-7980XE
1 x Intel Xeon Gold 6154
2 x AMD EPYC 7551
2 x Intel Xeon Platinum 8168
4 x Intel Xeon Platinum 8180
Lots of choice really, if the budget, motherboard, power, cooling and noise aren't a constraint.

[Racepro]

Thanks David. From the specs on those CPU's you listed they are massive.
If 10 cores would do the job satisfactorily then no point in going for 16 cores.
I agree with you I really need to test my data. But how ?

[David (PassMark)]

Correct, there is no point buying more cores than you need.
Just have a look at the current CPU usage on an existing machine.

[Racepro]

I've carried out the tests using my 10 year old Dell T7500 Workstation 24GB RAM and two X5570 CPU's and all cores are maxed out
when just moving from one cell to another. Using files sizes of 250,000+ rows.

[Racepro]

Is there a way I can have the spreadsheet calculated by a source that has multiple core solutions ?
I'm happy to send the WS to someone who can put it through the paces.on anything between an i7 8700x
and an i9 7980x

[David (PassMark)]

The X5570 was quad core with hyperthreading. So two of them would be 8 physical cores (16 virtual).
So you can be pretty sure the spreadsheet isn't single threaded. So more than 4 cores would be a good thing in this case.

Also my experience has been that if Excel is too slow you could rewrite the spreadsheet as an application (as a C++ application, not VB). Might be a fair amount of work, but you might get a 10x speed improvement.

[Racepro]

Aah, that's shed some light on it. I will buy the most cores I can afford. Probably 12 or 14.
I have long thought of having the Excel program written in another language. I have written the Excel
program with the help of a few formulas from a forum. I like using Excel as I developed the program
and if anything goes wrong I can source the problem and fix it. I am not familiar with code but at some time
soon it may pay me to move in that direction.
Thanks for your advice.
Ian

[sarahi]

First, thank you for your time and advice. I am an investment analyst, not a computer programmer. I am among the best in the world at building spreadsheets to digest large amounts of financial data to make better investment decisions. I am going to work for a hedge fund and I need a bad ass PC.Please save the comments that I should be building my tools, models, and algorithms using other programs and languages. I am a finance guy, and I develop in Excel 2010. My spreadsheets can approach 1 GB in size, but yes they still work Bluestacks TextNow Photomath

[David (PassMark)]

I develop in Excel 2010

You should at least switch to Excel 2017. I am sure they have been improving the multithreading over the last 10 years. Office 2010 is a also security risk now. Plus if you are using 2010, then likely you are also stuck on the 32bit version. Which is limited to 2GB of RAM usage. (but in practice a lot less than 2GB can be used due to fragmentation of the small 32bit address space).