Quick benchmark question

Toggle sidebar Toggle sidebar
Skylyne

Skylyne

Reputable
Sep 7, 2014
678
0
5,160
I've been using Passmark's benchmark numbers for quick references to performance, and I've also been using the CPU comparison tool for quick comparisons on the fly. I'm wondering if someone can explain what the "# of Samples" section means in the comparison charts. I assume it has to do with how many samples the hardware can handle (as it appears on both the CPU and GPU comparisons). Also, if anyone can shed a little light on what the significance of (what I assume is) the sample rate, that would be awesome.

I also know that synthetic benchmarks aren't really ideal for genuine comparisons to real-world performance; I just find their comparison charts to be a decent enough reference for hardware specs and certain types of performance. If you have another comparison site you would recommend for quick reference, I'm open to suggestions. However, an explanation of the sample rate would still be nice.
 
Sort by date Sort by votes
Skylyne

Skylyne

Reputable
Sep 7, 2014
678
0
5,160
That makes perfect sense. More than what I originally thought. Great, thank you.

Is there any other specs I should consider outside of processing power, core count, and single thread rate? Is it safe to assume the multi/hyper-thread rate of a CPU is effectively double the single thread rate?
 
Upvote 0 Downvote
Karsten75

Karsten75

Judicious
Oct 15, 2009
6,909
0
36,960



The benchmark numbers include the effect (benefit) of Hyperthreading. Hyperthreadind actually has a small overhead cost, so the effective single-core rate of a hyperthreaded CPU will be a little lower than the rate of a similar processor without hyperthreading.

I think you can test your hypothesis by dividing the benchmark number by the number of cores. that is what I would use for the throughput of a core with hyperthreading. Not that I see where it's useful, since anything that can multitask on a thread can multitask on a core.


As for what other numbers to consider - I don't really know. Benchmarks is one of the things that can most often be misinterpreted and misapplied. Along the lines of Lies, Damnded lies, Statistics and benchmarks. To paraphrase Mark Twain.
 
Upvote 0 Downvote
Skylyne

Skylyne

Reputable
Sep 7, 2014
678
0
5,160
So, it seems like a benchmark number should, in theory, give me a rough estimation of overall throughput per core with multi/hyper-threaded cores? That seems fair enough. I'm really just trying to figure out the best way to compare the difference between CPUs while on the fly; so, it doesn't have to be a perfect measure, as long as it is somewhat a representative of real-world performance during multi-threaded tasks. One of the things I'm trying to compare at times are single thread cores to hyper-threaded cores, and comparing an FX chip to certain i3/i5 cores to find out if there would be a noticeable difference in different performance scenarios. Looks like the only sure-fire way is to build rigs and test them myself! If only I had the money....

Thanks for your help, mate.
 
Upvote 0 Downvote
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom