Processors
Intel® Processors, Tools, and Utilities
14405 Discussions

Puzzled by processors performance

BChua5
Beginner
1,383 Views

Hi

i am testing a program on 3 computers so that i decide the spec of a new computer i should be getting.

Comp 1 : Intel i5-4570 @3.20GHz, 3.20 GHz, 32 G RAM

Comp 2 : Intel i5-7500, 3.40 GHz, 3.41 GHz, 8 G RAM

Comp 3 : Intel Xeon Gold 6130 2.1 GHz, 2.10 GHz, 96 G RAM

i have a program.

Comp 1 : took 23 hrs to complete

Comp 2 : took 4.5 hrs to complete

Comp 3 : took 4.5 hrs to complete

I am puzzled and i do not know how to explain. I was expecting the performance of 1 and 2 to be similar and comp 3 will be more superior.

but the results showed otherwise. A normal PC (Comp 2) can perform as good as a Xeon (comp 3) but why Comp 1, which has similar processor as comp 2 took so significantly longer to complete the job ?

I wish to seek some help to explain this so that i can make a purchase decision ?

thank you

best regards

Boon Yiang

0 Kudos
5 Replies
AlHill
Super User
500 Views

https://ark.intel.com/compare/120492,97123,75043 Intel® Product Specification Comparison

Doc

0 Kudos
BChua5
Beginner
500 Views

Hi Al

Thanks for the comparison .

Can you help me please. I have read the section on performance but I do not know what I could use to interpret my test results.

I am expecting Comp 1 and 2 to complete almost same time . But it took me by surprise that Comp 2 actually performs similarly to a server grade Xeon.

I am worried that if I were to recommend Comp 2 but it may not be good enough . If I were to recommend Comp 3, i may be Over buying.

Comp 1 : Intel i5-4570 @3.20GHz, 3.20 GHz, 32 G RAM

Comp 2 : Intel i5-7500, 3.40 GHz, 3.41 GHz, 8 G RAM

Comp 3 : Intel Xeon Gold 6130 2.1 GHz, 2.10 GHz, 96 G RAM

Comp 1 : took 23 hrs to complete

Comp 2 : took 4.5 hrs to complete

Comp 3 : took 4.5 hrs to complete

Hope to seek your advice please

Thank you

Boon Yiang

0 Kudos
n_scott_pearson
Super User
500 Views

What does your test program do with threading? If it is a single-threaded program, then the response of the two latter processors is going to be similar (you will get no advantage from the extra memory controller nor from its (many) additional cores and threads).

What are your criteria for this new PC?

...S

0 Kudos
BChua5
Beginner
500 Views

Hi Scott

Thank you very much for your reply.

I am running a program called Mplus, it's a kind of factor analysis application, likely written as a single-threaded program.

I am trying to buy a new computer that could help my folks speed up running such program.

They have the impression that more CPU/CORE/Thread etc will be faster.

We are even thinking of subscribing to Azure/AWS. However, we do not know is the slowness due to Hardware or Software limitation.

Hence, do i gather that based on this, i should be getting a CPU that has "fastest threading" performance ?

That is, i do not need to get a Xeon since i am just looking for a computer that would be running a single analysis as fast as possible, at any one time, that's the situation most often.

Thank you

Best Regards

Boon Yiang

0 Kudos
n_scott_pearson
Super User
500 Views

If you are going to stick with a single-threaded analysis model, then you will want to look for processors that have higher Turbo clock speeds (higher base clock speeds and perhaps even larger L2/L3 cache sizes can help as well). There is the distinct possibility that, since you do not need the additional cores, you can get away with a Core i5 (or perhaps even a Core i3, depending upon how cache sizes affect execution), instead of a Core i7, if this is the case.

If the analysis lends itself to being parallelized, then rewriting it to take advantage of this can significantly reduce execution time as you spread the analysis across more and more cores. Then, depending upon how many parallel paths the analysis can take advantage of, you can determine the optimal number of cores that are necessary. Note that the higher the number of cores utilized, the lower the Turbo clock speeds that can be attained. In this case, higher base clock speed becomes a more-important factor.

I would add that you should also be looking at how I/O bound your analysis is. That is, if large amounts of data must be read from storage (and especially if (then) written to storage), then you should be looking at where you store - or perhaps cache - this data. Obviously, the faster the storage you use, the less the I/O bounding.

This is all I can say (abstractly) without knowing a lot more about the analysis (details I likely don't need or want to hear).

...S

0 Kudos
Reply