Intel Historical CPU Benchmark Charts

SAFCasper

SAFCasper

Well-Known Member
Reaction score
804
Location
United Kingdom
Some data I was messing around with and thought others might find interesting

CPU benchmark numbers taken from Passmark for generation 2 through 14. These aren't flagship models I have purposely selected desktop i5's you would find in a mid-range OptiPlex/EliteDesk/ThinkCentre.

1711402191384.png

1711402200662.png


Some notes to go with...
- 5th gen was laptop only, hence the blank
- 8th gen changed to 6c/6t. Every generation previous was 4c/4t
- 10th gen enabled hyperthreading for i5 making it 6c/12t
- 13th gen introduced an additional 8 efficient cores hence the massive multi-threaded performance increase.
 
The raw numbers

ModelMT ScoreMT IncreaseST ScoreST Increase
i5-25004,127--1,707--
i5-34704,67713.33%1,94113.71%
i5-45905,36814.77%2,0877.52%
i5-65005,6334.94%2,1111.15%
i5-75006,0256.96%2,2657.30%
i5-85009,58159.02%2,4628.70%
i5-95009,8142.43%2,5764.63%
i5-1050012,73329.74%2,7647.30%
i5-1150017,39736.63%3,13713.49%
i5-1250019,92614.54%3,69917.92%
i5-1350032,05660.88%3,8925.22%
i5-1450032,5471.53%4,0052.90%


Can't say I'll be rushing out for any 14th Gen devices based on those numbers.

Will be interesting to compare 15th gen once it releases. It's rumoured to bring huge single-threaded improvements but at the cost of removing hyper-threading.
 
Last edited:
13th and 14th generation don't have much of a gap, and you won't see much in the way of performance gains between them. There is a moderate power savings gain... but mostly we're still waiting on the OS's to get the scheduling right on the e-Cores.
 
Would be nice to have the "real" progress: performance per core per GHz... and maybe per Watt.
 
Philippe said:
Would be nice to have the "real" progress: performance per core per GHz... and maybe per Watt.
Click to expand...

Turbo boost complicates this without knowing exactly how the benchmarks were performed. Most have a base wattage around 65W however can draw 150W+ during turbo boost. It gets even more complex once E-Cores are added to the mix.
 
I've been using the Passmark scores from cpubenchmark.net for years as a guide to CPU performance. I get they're not 'real' but when there's a big difference between numbers it's meaningful to some extent.

What else have we got to go on? In the early days, the clockspeed in GHz was pretty much all we needed. Some old-school customers still ask about GHz but these are now meaningless for comparison purposes (except between, say, two i5's from the same generation).

I sell refurbs and new computers to residential and very small business customers. I quote the Passmark scores as a guide for comparing performance between offerings, and sometimes compare it with their current laptop's pathetic score (e.g. AMD A4 or Celeron). This is far more relevant performance information than anything provided in a retail chain store.
 
fincoder said:
I get they're not 'real' but when there's a big difference between numbers it's meaningful to some extent.
Click to expand...

I doubt that anyone would disagree with the assertion that benchmarks can be used as guidelines that are definitely meaningful to a certain extent.

But they have always used carefully contrived and controlled conditions, much as virus testing does. The results from both give useful insights, but neither is close to the equivalent of "real world" conditions.

The problem is that some take these results as being somehow objectively connected to real world conditions, that is, that they are "real" results. They never have been, and never will be. There's been more than one occasion where, under real world conditions and running typical software, processors with lower benchmarks best ones that look better on paper.

We've also long gone past the day where what most people, and yes, I do mean most home and business users, use computers for does not come close to exploiting the computing power available. Just like it makes no sense to have 128 GB of RAM when your max usage is around 14 GB, it makes no appreciable difference if your processor can fly to the moon and back in two seconds if all you're doing is working spreadsheets and reading email.
 
Also as I was saying earlier turbo-boost can massively skew benchmark numbers. An i5-14500 will turbo up to 5GHz resulting in fantastic looking benchmark scores. However in the real world that lasts maybe 2 minutes before getting too hot and dropping back to the 3GHz range. The listed base frequency is actually 2.6GHz

I still believe benchmarks are useful and provide a good indicator of expected performance. Just need to understand they are not definitive.
 
Philippe said:
Come on! I meant: if we leave the number of GHz and the number of cores aside, does the inner working of the cpu (one core) have really increased its performance...
Click to expand...

I get what you are looking for. If someone could theoretically lock each CPU to a single core fixed at 1GHz. Benchmark this and also record the wattage drawn by the CPU during. It's only part of the picture but I'd definitely love to see those numbers.

Closest I can do is extrapolate from single-treaded scores using listed turbo frequencies to determine in theory what each CPU would score with a single core at 1Ghz.

To be taken with many pinches of salt.
1711661928665.png
 
Thanks! Taken with caution, but very interesting!
It helps separate marketing from "real" cpu design / technological breakthrough.

Lost of core performance :
- from i5-4590 to i5-7500
- between i5-8500 and i5-9500
- between i5-13500 and i5-14500
and sharp increases:
- between i5-2500 and i5-4590
- between i5-10500 and i5-12500
 
Philippe said:
Come on! I meant: if we leave the number of GHz and the number of cores aside, does the inner working of the cpu (one core) have really increased its performance...
Click to expand...
Yes it does, the clockspeed measurement is a terrible indicator of performance, which is why way back in the Barton core era AMD started selling model numbers, and Intel followed suit later.

That's true even if you're measuring the performance of a single core!
 
You must log in or register to reply here.
Back
Top