Why is 3.7GHz 4-thread G5400 have slower benchmark than 3.4GHz i5-3570k?

[AveryFreeman]

I am a little confused about this.

Okay, so the G5400 is newer. It's a celeron, but it has 2 cores with hyperthreading, so the effective core count is the same as the 4-core i5-3570K without hyperthreading...

The G5400 has a base clock speed of 3.7GHz, while the i5-3570K is 3.4GHz... i5-3570K has turbo but it only puts it to 3.8GHz so not a huge increase over G5400's base speed...

the G5400's memory bandwidth is almost 50% higher - 37.5 GB/s vs 25.6 GB/s...

i5-3570K has AVX, while G5400 does not.

Anyway, most these points make the G5400 appear that it would be faster than the I5-3570K, yet passmark has the G5400 listed as 5270 in their benchmarks, while the i5-3570K is 7167. That's approx 50% "faster" according to the benchmark.

Can anyone explain to me why this discrepancy is so great when these processors seem so similar in so many respects?

 

[SticKx911]

First, virtual cores and real cores are not the same.

I havn't pulled up the specs, but I'd bet the i5 has more on die cashe.

I always thought the celerons were intentionally built to be worse on the ipc front. The architecture is different, and designed more for cost and power savings over its performance designed brethren.

 

[BinarySynapse]

Two hyper-threaded cores is still only two cores. Hyperthreading allows a second thread to utilize a core's execution units that aren't being used by the currently running thread to achieve some performance enhancement, but it will never be able to match the performance of those same two threads running on true cores.

 

[AveryFreeman]

First, virtual cores and real cores are not the same.

I havn't pulled up the specs, but I'd bet the i5 has more on die cashe.

I always thought the celerons were intentionally built to be worse on the ipc front. The architecture is different, and designed more for cost and power savings over its performance designed brethren.

Good point, I forgot to look at the L2 and L3 caches.

How is the architecture different? I thought the pentiums were just purposefully (maliciously? opportunistically?) restricted versions of the same thing...

 

[defaultluser]

I always thought the celerons were intentionally built to be worse on the ipc front. The architecture is different, and designed more for cost and power savings over its performance designed brethren.

No, you're thinking of the Silver Celeron/Pentium cores. The Gold ones are the same architecture as the i5/i7.

In fact, the Pentium G5400 us the same exact core configuration as the old Kaby Lake Core i3 7100.

The only difference between the Pentium Gold and the more expensive chips is AVX. That will only show up in canned benchmarks (games are not really effected).

AveryFreeman, the reason for the performance difference is they're adding extra processing threads to each core, to improve efficiency (execution units that are idle can be put to use by the second process). This is cheaper to add than two more actual cores, and as you might expect it is much slower.

Celeron Gold with 2 cores and 2 threads Passmark (3600 when you scale to match clock speed with the other two):

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Celeron+G4900+@+3.10GHz&id=3275

Pentium Gold with 2 cores and 4 threads = 5200.

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Pentium+G4620+@+3.70GHz&id=2948

Core i3 8100 with 4 cores and 4 threads = 8075

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i3-8100+@+3.60GHz&id=3103

So, The Pentium with two cores / four threads is 40% faster than the Celeron with two cores/ two threads.

The Core i3 with four real cores is twice as fast as the Celeron with two cores. (that > 2 difference is because of AVX).

And Passmark is a benchmark built for simple architectural comparisons between cores. In most real-wold use cases you don't get that good a speedup from Hyperthreading (closer to 25%) or doubling your cores (closer to 80%) or adding AVX (closer to 10%).

Passmark is a good indicator of optimal performance, but don't take it's scores as gospel