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

[dasa 20]

Back again after doing a cpu vs ram comparison in 2013 with a 2600k
Hope some of you find this of interest

Test System
3770k -- 6700k
Asus Maximus V Gene Z77-- Asus Maximus VIII Gene Z170
Samsung DDR3-1600c11 2x4GB 1.35v dual sided -- G.Skill DDR4-3200c14 2x8GB single sided
EVGA GTX1070 SC (OC=+75 core +750 mem +112 powertune use on all tests)
64g Samsung 830, 1tb Sandisk ultra II, 3tb WD green
Seasonic XP 1050 < tad overkill for the current system which has topped out at 306w from the wall
Silverstone TJ08-EW
Swiftech Apogee drive II
200mm Phobya rad + Silverstone Air Penetrator AP181
EK VGA Supreme HF

 

Windows 10

Nvidia driver 372.70

Arma Detail settings Ultra 1080p aa off + increased view distance ~5000-6000 object ~3500 sorry cant remember exactly should have screenshot them

Stratis benchmark_0.51

3770k was benched about a month before 6700k and there was some small updates that didnt seem to affect performance

 

Also posted performance tests with Fallout 4, R6 Siege and AIDA64, XTU over here

http://www.overclock.net/t/1611359/3770k-vs-6700k-in-fallout-4-arma-3-rainbow-six-siege-with-core-cache-memory-scaling-ddr3-1600c11-2133c9-ddr4-2133c15-3000c12-4000c17

 

Been playing on a liberation apex server with the new system and its sure running much nicer mostly 60fps+ but it has dropped to ~39fps a few times

[domokun 515]

Fascinating and jolly well done!

Particularly interesting for a bloke running a 3570k @ 4.4 & GTX 970 @ 980.

This will come in really useful when trying to justify that Skylake upgrade to myself and my better half ;)

I only wish that more chaps were more nerdy diligent when upgrading.

BTW is the difference between the 6700k @ stock with single channel (47.5) and the bar above (59.1), really just another stick of RAM? i.e. single versus dual channel?

[Valken 622]

Awesome. I just upgraded my ram recently and what benchmark and settings are you using to test?

[froggyluv 2136]

 This is all above my intellectual paygrade but im interested in whether just speedy ram is worth the upgrade? Im running 4.5 OC on a 4690k with just 8 gigs 1600 ram. Wondering if getting some really speedy ram will give me any worthwhile gains

[St. Jimmy 272]

What is "Stratis time demo"? Nice results.

 

Just wondering what kind of benchmark mission you used? Scale isn't likely that big in something like YAAB

 

Thanks.

[ceeeb 147]

Interesting results, fast RAM > overclock. Would be good to see the results from YAAB (which is typical of SP gameplay, with AI bound performance).

 

My system results in YAAB:

6700k @ 4.2GHz, 32GB DDR4 @ 2133 CL15 Default ->  41 FPS
6700k @ 4.5GHz, 32GB DDR4 @ 3200 CL14 XMP  ->  50 FPS

[St. Jimmy 272]

Interesting results, fast RAM > overclock.

Yep this was also shown in the older topic:

 

 

 

 

533MHz in the RAM or 600MHz in the CPU is pretty much the same, if the timings are kept somewhat tight. (it's pretty even in this situation).

 

In this topic benchmark it follows the same line. 666MHz in RAM or 500MHz in the CPU, the performance is pretty much the same, if the timings are kept somewhat tight. (RAM wins marginally because of higher overclock).

[dasa 20]

BTW is the difference between the 6700k @ stock with single channel (47.5) and the bar above (59.1), really just another stick of RAM? i.e. single versus dual channel?

Yes it is

If you click the link you will see that in aida64 it makes a huge difference to bandwidth and it seems that amra\skylake benefit from both bandwidth and latency

 

Awesome. I just upgraded my ram recently and what benchmark and settings are you using to test?

Sorry should have screenshot it

Ultra 1080p aa off i think view distance was around 3k object was probably ~1.5k cant remember exactly im afraid

 

 This is all above my intellectual paygrade but im interested in whether just speedy ram is worth the upgrade? Im running 4.5 OC on a 4690k with just 8 gigs 1600 ram. Wondering if getting some really speedy ram will give me any worthwhile gains

Depends a bit if your ram is 1600c8 then its fairly good as the old tests show and probably wont make a noticeable difference if you upgrade

But if you have a slower 1600mhz kit then 2400c10 could bring a noticeable improvement

As for weather or not this is worth the price to upgrade it i cant say

What is "Stratis time demo"? Nice results.

 

Just wondering what kind of benchmark mission you used? Scale isn't likely that big in something like YAAB

 

Thanks.

benchmark_0.51

https://forums.bistudio.com/topic/142875-arma3mark-benchmark-your-arma-3/

 

Interesting results, fast RAM > overclock. Would be good to see the results from YAAB (which is typical of SP gameplay, with AI bound performance).

 

My system results in YAAB:

6700k @ 4.2GHz, 32GB DDR4 @ 2133 CL15 Default ->  41 FPS

6700k @ 4.5GHz, 32GB DDR4 @ 3200 CL14 XMP  ->  50 FPS

I will take a look at YAAB nice results

[infiltrator_2k 29]

Hmm... I have a 5820K which scores higher than the 6700K in CPU Benchmark, but it seems the 6700K is quicker in single core tests.  So I guess this means Arma runs better with a 6700K.

[brightcandle 114]

Hmm... I have a 5820K which scores higher than the 6700K in CPU Benchmark, but it seems the 6700K is quicker in single core tests.  So I guess this means Arma runs better with a 6700K.

 

http://www.hardware.fr/getgraphimg.php?id=223&n=14

 

Significantly better. The enterprise chips actually get hurt rather than helped and I think that is because they have a different design for the memory controller designed to spread out memory bandwidth across cores whereas Arma is heavily dependent on its main core thread.

[dasa 20]

http://www.hardware.fr/getgraphimg.php?id=223&n=14

 

Significantly better. The enterprise chips actually get hurt rather than helped and I think that is because they have a different design for the memory controller designed to spread out memory bandwidth across cores whereas Arma is heavily dependent on its main core thread.

the main difference there is clock speed and ipc

quad channel memory does result in higher latency though

once overclocked to similar speeds haswell-e is almost on par with skylake but skylake does usually oc a little better

http://translate.googleusercontent.com/translate_c?depth=7&hl=en&rurl=translate.google.com&sl=pl&tl=en&u=http://pclab.pl/art69780-16.html&usg=ALkJrhi7wXIJK6YXYeAmrBYtTKRZnxzRVQ

 

infiltrator_2k i would love to see you compare single vs dual vs quad channel ram performance in arma

[Dwarden 1125]

what i miss is what OS is used, some setups of the OS, drivers used, settings in arma 3 used etc. so it's more information rich

[dasa 20]

what i miss is what OS is used, some setups of the OS, drivers used, settings in arma 3 used etc. so it's more information rich

some setups of the os?

added extra detail to the op with os win10 latest nvidia driver that was used and basic arma settings

 

i should have mentioned in the op that the 3770k vs 6700k isn't a perfect comparison as i did the first tests about a month before the 6700k tests due to the mb being listed as in stock when it wasnt then getting left at a startrack depo for two weeks by the time the lazy sods got around to delivering it i was on holidays

there was some updates released in that time although they didnt seem to have much affect on performance

the benchmark was extremely cpu limited with the gpu only running ~40% load at times while downclocking itself to ~1000mhz

i can post some of the extra data that was recorded if you would like

for example this is from fallout 4

[Valken 622]

I upgraded from DDR3-1333/1600 @ 1333 CL9 to DDR3-2400 CL10 and it gave me at least 10 FPS on the bottom end.

 

4790K @ 4.4 GHZ all cores, DDR3-2400 CL10, SSD, AMD 6950 unlocked CFX:

 

Altis 47 FPS

Stratis 58 FPS

YAAB 34-36 FPS

 

 

Single AMD 6950 unlocked with DDR3-2400 CL10

 

Altis 37 FPS

Stratis 48 FPS

YAAB 29-31 FPS

 

I am looking to upgrade to a GTX 1070 or faster (waiting on AMD Vega) then will decide. 

 

I did have a chance to pickup a 5775C for really cheap but my cooling is only a Corsair H50 (2nd version) so its pretty old. Didn't feel like spending 100 USD just to push the CPU up by 200-400 MHZ!

 

RAM was a superbuy for 32 GB at 130 USD.

[froggyluv 2136]

@Dasa: Thanks. Yes its dd3 1600 8cl but looking at Valken's post it looks like it may be worth the upgrade. Recently bumped my CPU overclock to 4.6 and while it seems to hold steady with decent temps,  I've had a few crashes with some sort of memory error (arma only) that ive not seen before (flashes too quick to capture) -wondering if the CPU OC is too much for the mem...?

 

 @Valken - wow nice! Think Im gonna go that route as well and ive also got either a 1060 or 70 inbound today as well :292: :627:

[dasa 20]

first up sorry i believe my memory was way out with the view distance used in those tests i think it was closer to 5000\3000 after running it again

 

ok gave yaab a quick run at 1080p ultra aa off 5000\3000 view distance

stock clocks 44fps

stable oc i settled on 4.7\4.6 3866c16-16-16-2t 62fps with sub timings tweaked its performing as good or better than the settings i used in the first lot of tests

 

 

 

@Dasa: Thanks. Yes its dd3 1600 8cl but looking at Valken's post it looks like it may be worth the upgrade. Recently bumped my CPU overclock to 4.6 and while it seems to hold steady with decent temps,  I've had a few crashes with some sort of memory error (arma only) that ive not seen before (flashes too quick to capture) -wondering if the CPU OC is too much for the mem...?

valken was running 1333 CL9

thats a hell of a lot slower than 1600c8 which matches 2133c11

im not saying dont do it as the gains may be noticeable but they will only be half what valken is seeing

maybe try overclocking your current ram first who knows maybe it will do 2133c9 which is a bees dick away from 2400c10

 

as for testing stability testing i have just been going through this

http://www.overclock.net/t/1569364/official-skylake-haswell-e-broadwell-e-24-7-ddr4-memory-stability-thread

 

Google stressapp test via Linux Mint does a excellent job of finding stability problems with ram in a fraction of the time other tests take

 

HCI does a very good job of stressing the link from ram to cpu cache

you have to run one instance for every cpu thread and want to use most your available ram

 

Realbench is a good cpu stability test

[Valken 622]

@ dasa - good advice. Should really OC the ram to see how far it goes. My old ram cannot OC at all so I had to lower it down to 1333 to match the 2x 8 GB sets.

 

But I will try to OC my new RAM further. I can honestly say, in non GPU situations, the HASWELL CPU LOVES FAST RAM. Everything is snappier.

 

RAM was the biggest upgrade to ARMA for my system at the below settings which is what I play online in MP games.

 

Also, I am super GPU bound due to 2 GB VRAM and older architecture with no new driver support.

 

For everyone's reference:

 

Win 7 Pro 64, AMD 15.6 drivers (yes, over a year old but BEST for my GPUs).

 

My settings were 1080P, all Ultra except for:

 

Objects set to LOW

PIP set to LOW

Clouds set to LOW

3800 draw distance

3800 object distance

HW AA OFF

FXAA Ultra

No BLUR

No DOF

200 Sharpness

 

Otherwise, everything was from Ultra PRESET.

 

Yes, I traded object, skies and pip quality for draw distance but I can advise my PIP is SHARPER than setting it to ULTRA! PIP LOW + TEXTURES ULTRA is actually near perfect.

 

When I get the new GPU, I will up everything back to ULTRA and try to max out SHADOW, TERRAIN and OBJECT DISTANCE while keeping ~ 60 FPS. I am so itching to just buy that 1070 GTX right now! 

[froggyluv 2136]

 Might want to hold off on the VC upgrade Valkan -just threw a new 1060 6gb in and benchies are the same with a couple seemingly even smaller ...oh YOU Arma... :wub:

 

 Thats an upgrade from a 760 4gb, maybe ill have less texture drop with tons of effects mods running -need to justify it somehow..

[dasa 20]

apex is a fair bit heavier on the gpu

the gtx1070 is working fairly hard some of the time at 2560x1440 8xaa on everything

[Valken 622]

@ Froggy - did you OC your ram yet? I think the 1060 6 GB should allow you to EXTEND the draw distance for terrain and objects with next to no performance hit. 

[jumpinghubert 49]

Hi dasa,

 

thanks for this detailed comparison. There is something wrong with the 3770k results because I get by far higher results.

 

 

Windows 10

Arma Detail settings Ultra PRESET 1080p aa off

Stratis benchmark_0.51

hardware specs in my signature (4.8GHZ instead of 4.9 @ cpu)

 

result is 79fps.

[dasa 20]

thanks for the feedback

nothing wrong with the tests as far as i know just my crapy mistake with not taking a screenshot of the settings

i upped the view distance a lot to ~5000-6000 and objects ~3000-3500 going by performance of the current system

don't think it affects performance much but i also play with blur & dof off sharpness increased these may have been changed in the settings to

6700k system at 24\7 oc gets 102fps at ultra preset 1080p aa off no other changes

http://cdn.overclock.net/f/f7/f77dcbef_hci1000aida64.jpeg

 

thats a nice clock speed you got our of your 3770k and ram i had to deli the 3770k just to get it stable at 4.5ghz it only did ~4.1ghz before that

the ram at 2400 c10 was stable if it posted but it often failed to post over 2133

[froggyluv 2136]

@ Froggy - did you OC your ram yet? I think the 1060 6 GB should allow you to EXTEND the draw distance for terrain and objects with next to no performance hit. 

 

 Yeah well I was wrong, apparently my ram while being sold at 1600, is like 1333 and can be overclocked to 1600 ...sigh..so no, it wont budge above that. As far as the 1060, yeah im definietly noticing stronger FPS while running all my Franken-mods with less blocky smoke textures. Was able to bump the draw distance as well.

 

 Im happy.

[dasa 20]

ddr3 will default to 1333 till you load its xmp profile for 1600c8

a 2400c10 kit will also default to 1333c9

[froggyluv 2136]

 Ah I see. I did always have XMP enabaled and knew it was running at 1600 -but if you look at my System Information page etc it states 1333.