3770k to 6700k performance review with ram\cache scaling

[jumpinghubert 49]

Hi dasa,

 

that was my thought too. But I remember the view- and objectdistance is cut by the benchmark itself. It would be nice if you can make a run (stratis and yaab) with clean ultra preset (I copy&pasted all my arma3 settings before in a new folder (documents/arma3) to get the own settings in a fly after it again. I have still fred´s malloc running (if it is still working..)

 

If you have aida64 it would be nice to have bandwich run with your 6700er system. A 6700er with 4.7 Ghz...just wow, really nice clockspeed :)

 

My i3570 is watercooled on the blanc die without heatspreader so I can reach the 4.9 with a ton of volts without going above 58°C.

 

edit:

oh I´ve overseen your link with all the aida64 results :P

56000+ mb/seconds read....pffff thats a lot...I get only 38000+. I think arma3 likes that a lot.

[dasa 20]

that was my thought too. But I remember the view- and objectdistance is cut by the benchmark itself.

 

if so but it must be at a level well above ultra as at the ultra preset it scores 102fps and with 12k\12k view distance it drops to 70fps while ~5k\3k it was 84fps

i can record a few more tests on the 6700k with the ultra preset tomorrow if you like

just strat and yaab at the 24\7 oc ultra preset or something else?

[jumpinghubert 49]

Only strat & yaab with standard ultra preset/0 aa/1080, that would be wonderful :)

[dasa 20]

ok

stratis 103fps

yaab 58fps first run while recording data 62fps second run when i wasnt i will reboot and run it a few more times

cpu usage could be interesting but with 8 threads bouncing all over graphed its not exactly readable

maybe i should try with ht off

 

edit:

with ht off cpu usage bounces around 40-80% over 4 cores

performance may drop a few fps with ht off not much

there does seem to be ~4fps variance in the yaab benchmark not sure if its just from the ai doing things different like destroying apc vs missing and maybe part of it is a performance hit from recording fps\usage ect.

[ceeeb 147]

Yes, YAAB has about +-5% FPS range due to random nature of AI simulation. The only thing to do is to run multiple times and average the results.

That slow frames around 75 seconds are also a common feature, maybe loading new assets or maybe an engine issue.

[Valken 622]

@ dasa - I was curious how your performance would be if you turned HT off? Just curious if ARMA makes efficient use of it.

[jumpinghubert 49]

Thanks dasa for this very professional documentation. I get 44fps in yaab with ultra preset in 1080 and aa off ( with a 3570k @ 4.8Ghz and 2666mhz ram). Thats a big difference to your 62fps! With the newest intel generation arma3 seems to benefit a lot!

[dasa 20]

@ dasa - I was curious how your performance would be if you turned HT off? Just curious if ARMA makes efficient use of it.

no ht doesn't help

this is a very old test so things may have improved a bit with core scaling but as far as ht goes it still doesn't really help and this is true with most games

even those dx11 games where it does help its mostly towards the max fps rather than the minimum and only ~5-7% gain

dx12 may change this a bit and bring that up to ~15% in some situations but its still to early to know

there is an exception to this rule and that is in games that are only optimized for more cores than your system has for example a dual core i3 cpu running a game that only supports a quad core cpu then ht can bring massive gains to average fps although you may still get some stuttering you wouldn't get with a real quad core

[froggyluv 2136]

 Why is DDR4 cheaper than DDR3? Im looking at 16gb (2x8) 3000mhz and while DDR4 hovers between $150 and $200 the DDR3 versions are closer to a $300?

[dunedain 48]

Might because you're comparing standard spec mass produced DDR4 to high end DDR3.

[Valken 622]

@ Froggy - Lack of supply. DDR3 is running out and DDR4 is in full production for current and newer systems!

 

I got a set of G.skill 2400 CL10 Trident X DDR3 32 GB if you were wondering. 16 GB would work as well.