[Simbelmynë]

Hey,


So I have tried to figure out why my (overclocked) 7600k is slower than the 8700k, at similar core-counts (also, the 8700k does not improve when going from 4 cores to 6). The IPC of coffee lake does not seem to be significantly better than kaby lake.


So can this be related to the double L3 cache of the 8700k compared to the 7600k?


Here is my testfile (note that overclocking has been performed on all cores):

-----------------------------------------------------------

core i7 7600k standard turbo frequencies
1C 4.2 GHz
2C 4.1 GHz
3C 4.1 GHz
4C 4.0 GHz


core i7-7600K, DDR4 2133 MHz CL 15
# cores   Wall time (s):
------------------------
2 423.92
4 357.42


core i7-7600K, DDR4 2400 MHz CL 15
# cores   Wall time (s):
------------------------
2 389.9
4 322.65


core i7-7600K, DDR4 2400 MHz CL 13
# cores   Wall time (s):
------------------------
2 388.86
4 320.97


core i7-7600K @ 4500MHz, DDR4 2400 MHz CL 15
# cores   Wall time (s):
------------------------
2 385.99
4 319.42


core i7-7600K @ 4700MHz, DDR4 2400 MHz CL 15
# cores   Wall time (s):
------------------------
2 379.26
4 321.16

core i7-7600K @ 3000MHz, DDR4 2400 MHz CL 15
# cores   Wall time (s):
------------------------
2 441.56
4 334.5

core i7-7600K, DDR4 3200 MHz CL 15
# cores   Wall time (s):
------------------------
2 364.07
4 291.68

core i7-7600K @ 4400MHz, DDR4 3200 MHz CL 15
# cores   Wall time (s):
------------------------
2 341.53
4 278.77

core i7-7600K @ 4700MHz, DDR4 3200 MHz CL 15 *Probably throttling, more testing needed*
# cores   Wall time (s):
------------------------
2 344.83
4 287.32

core i7-7600K, DDR4 3466 MHz CL 15
# cores   Wall time (s):
------------------------
2 335.94
4 268.92

core i7-7600K @ 4500 MHz, DDR4 3466 MHz CL 15
# cores   Wall time (s):
------------------------
2 326.45
4 267.25

and as a reference the 8700k yields:

core i7-8700K, DDR4 3200 MHz 
# cores   Wall time (s):
------------------------
2 312.15
4 249.55

[flotus1]

It is possible that some of the difference is due to the difference in L3 cache size. Hard to find out. Did you control uncore frequency during your tests?
Maybe if you ran a stream benchmark on both system to compare the actual memory performance. It does not necessarily have to be identical just because the memory is clocked at the same frequency with identical CL.

[Simbelmynë]

Uncore (cache ratio on my Asus board) has no significant effect on the 7600k results. I will check the settings on the 8700k tomorrow, although they should be on "auto" since I have not tinkered with them (I haven't been able to dig up what the default uncore is for the 8700k. Some references state 44x and some state 34x )



Is this the benchmark that you recommend?


https://www.cs.virginia.edu/stream/

[flotus1]

Yes, that's it. The reason being that you can estimate memory bandwidth based on frequency and number of channels. But the actual memory bandwidth that you can get will always be lower than this estimate. By how much exactly depends on a lot of factors, including the CPU/platform.
The influence of uncore frequency may not seem high, but since you are looking for relatively small performance differences keeping it at the same frequency for both CPUs would be the right thing to do. I once had a CPU that ran with different uncore frequencies on different boards with default bios settings.

Btw, which exact memory are you using in both systems?

[Simbelmynë]

The 8700k has the following memory:


https://www.corsair.com/us/en/Catego...2GX4M2B3200C16


And the 7600k has the following:


http://www.kfa2.com/kfa2/ram/hof-ram...e-extreme.html

[flotus1]

In that case my money is on the memory. Your I5-7600k has single-rank memory, the I7-8700k has dual-rank. At the same frequency, this can account for up to ~10% performance difference.

[Simbelmynë]

So I tested the 2R corsair memory with my 7600k. Unfortunately it refused to post @ 3200 MHz, but @ 3000 MHz I got the following results:

# cores   Wall time (s):
------------------------
2 348.52
4 280.29

So similar to 3200 MHz with the 1R memory. The highest possible frequency I manage with the 1R memory is 3466 MHz, which yields better values compared to 2R 3000 MHz (but probably not compared to 2R 3200 MHz). I could probably test my 1R memory in the 8700k machine, but I suspect that this will just confirm your 10% projection of the benefit of 2R over 1R.


So what is the highest rated OC 2R memory kit sold (and is is supported by any motherboard?)

[flotus1]

There are some 16GB modules (aka 2R) rated for DDR4-4000+
https://geizhals.eu/?cat=ramddr3&xf=...M1%7E5831_DIMM

For the motherboard, your best bet is one that only has 2 DIMM slots. Routing for only one DIMM per channel is easier which is why these boards are slightly better suited for high memory speeds and better throughput at a given memory speed. So either a mini-ITX board or something crazy like the Asus ROG Maximus X Apex (hats off to whoever comes up with those names )https://dlcdnets.asus.com/pub/ASUS/m...0-180129-2.pdf

But success will still depend on the quality of the IMC in your specific CPU. Just like some CPU cores overclock better than others, there is also a variation for memory controllers.

Then again, I would question if this all makes sense. 2 DIMM slots limit the maximum RAM capacity and for the same price or even less you could simply use a quad-channel platform with moderately clocked components.

[Simbelmynë]

Yes, it makes very little sense if we are talking about workplace related simulations.


This is more of cool-your-cpu-with-liquid-nitrogen type of stuff that is just fun to test.


OC memory is generally horrible to work with, both my setups demanded that I had the memory sticks in a certain order in order to post! (Not talking about the a1 b1 or a2 b2 setup here).