P3D upgrade CPU from 4930k 4.4 OC'd hex core to 6700k 4.2 turbo or wait for 7700k kaby lake?

[theskyisthelimit 48]

Guys, did you use 2x4Gb or 4x4Gb sticks? Does 16Gb improves performance in P3D/FSX?

 

Thanks,

Dirk.

In my 4930k ivy testing it was 4x4gb ddr3 gskill while on the 6700k its 2x8gb.

[westman 257]

The 2666 was 2x4 kingston fury , G-Skill Ripjaws V 3200 C15 and TridentZ 3600 C15 2X8 G

Have not seen any benifit to run 2x4 or 2X8 Its the chip itself B-die thats faster both the G-Skill have the same chip, Kingston Fury 2X4 have Hynix.

[Noel 4,503]

It's amazing how processor development seems to have really stalled over the past 5 years.  I still am using 3930K which has always o'clocked to 4.42Ghz w/o no fanfare and temps in the high 50's to low 60's depending on ambient. No big changes on the horizon are there?

[SimonC 116]

Over the past 5 years, the CPUs and the technology have advanced considerably. I recently changed from 2600K to 6700K, and yes, I know this was also DDR3 -> DDR4, and that DDR4 at 3600 compared to 1866mhz, but my performance really jumped. I run the 6700k on the same 4.8Ghz clock as my 2600k, and this was without hassle OC. I can't say how much in percent exactly, as this is hardly precisely measurable, however now I can run 60fps in P3D quite well. Of course not in every situation obviously, but the smoothness is OK. Not even mentioning IOPS on the M2 or PCI-Express, which gives storage and performance another kick.

[james_2k 36]

no they havent. and no you havent. clock for clock, its marginal at best between generation. compare to gpu advances where you get in the order or 30-50% more perf than the previous gen. not the 5-10% in cpus.

 

memory bandwidth is very situational, and for most games makes no difference.

[Noel 4,503]

 

 

I can't say how much in percent exactly, as this is hardly precisely measurable,

 

Well, when you look for example at single-threaded performance comparison of 3930K v 6700K, both at 4.4Ghz, believe the results or not 3930K outperforms 6700K, and it's several generations apart now:

 

https://www.cpubenchmark.net/singleThread.html

 

There are other optimizations that we aren't talking but we've really hit a wall in terms of this metric which if I"m not mistaken is quite critical to flight simulation performance.  This is why I haven't upgraded now for going on 5 years.  My last build I decided to buy the best I could find at the time hoping it would last a while and it sure has:  3930K@4.42Ghz, DDR3@2400, GTX Titan.   The sim runs very smooth as long as I respect what I'm doing with it--we avoid for example FTX regional scenery when flying PMDG stuff into big terminals, but can use FTX Global into 3rd party terminals no problem.  I hope this improves at some point but glad I don't have to discard parts now as it's all working quite well after all of these years.

no they havent. and no you havent. clock for clock, its marginal at best between generation. compare to gpu advances where you get in the order or 30-50% more perf than the previous gen. not the 5-10% in cpus.

 

memory bandwidth is very situational, and for most games makes no difference.

That's what I read, and yes indeed w/ memory bandwidth.   Because of this I've hoped LM could tap more and more into the computational prowess of modern GPU and SLI to push performance forward.   Some improvements for certain in this domain but not near what it could be at least what it seems.  For example, I can be in the 737NGX taxiing at FB's KPHX HD.  Frame rate will be sitting there at 22 let's say, and when I open GPUeze and read GPU Utilization it might well be sitting there at 60%, and CPU at 100% on the main core, the other 10 virtual cores are bouncing around per terrain texture processing, but it's really stuck because of this.  There is so much unused processing power untapped in that 5y/o GTX Titan.  I think that is the avenue that makes sense to try improve explotation of, the GPU especially thru SLI etc.   If the core box can't really jump much seems somehow have to get at the GPU more.  Sure, you can overtax the GPU with stuff that arguably doesn't add nearly so much as better core processing needs to--IOW the stuff that squashes performance in this very scenario I just described, restricted to 22fps with all of the processing power sitting idle on the sidelines.

[james_2k 36]

i only see my gpu maxed in heavy clouds really.

[theskyisthelimit 48]

i only see my gpu maxed in heavy clouds really.

I always hit 99% everytime I load up, say the fsxmark07 situation paused.. puzzles me how you arent hitting 99%, are you tweaking settings on the gpu or in the sim to avoid hitting 99%? My current settings might have something to do with it i guess.

 

Just an update on my earlier results thread, since i cant edit it anymore..

 

I had an error on my fsxmark07 120sec test (stock 4.2 turbo, 2133 ram, etc).. the 32 avg must have been a fluke, its been averaging 28 fps on average with an initial paused screen of 35 fps (all ftx,gex, rex etc).  Since then, i removed all drivers from my system via uninstall and driversweeper.. now i have results of 30.30 fps and a paused value of 40.   Quite happy at this point..

 

Moving forward i've started attempting an overclock to 4.6, so far i'm already up to 1.30 volts at least doing the blend prime95.. and temps are getting too hot with the h100i v2, hitting 85c when it was at 4.5 ghz (havent tested yet at 4.6 and new volts).. i have the h100i set to intake mode.. i may try the reverse.. i also have the case side off too.

[james_2k 36]

no, just doesnt use much of the gpu without heavy clouds, in my experience,

 

ssgs will also use more gpu.

[theskyisthelimit 48]

no, just doesnt use much of the gpu without heavy clouds, in my experience,

 

ssgs will also use more gpu.

 

Interesting, I dont have any NI settings like SSGS turned on in my case.. still hits 99% at start of the test.. still curious with other folks too, on the 1080 at least.  Ill have to test some other less intense scenarios and see how the gpu use is.

[james_2k 36]

initial testing (game not benchmark) on my new 1080 show less usage even in clouds.

[Noel 4,503]

I always hit 99% everytime I load up, say the fsxmark07 situation paused.. puzzles me how you arent hitting 99%, are you tweaking settings on the gpu or in the sim to avoid hitting 99%? My current settings might have something to do with it i guess.

You're running frames at unlimited, which translates to asking your GPU to crank out as many frames as possible relative to all the configuration settings you have setup in NI plus how you have the various GPU-dependent sliders in P3D. Try limiting to 30fps (reduces load on your GPU) and see if your GPU utilization doesn't drop significantly. It may allow you to move in-game GPU-dependent sliders a little more towards max and still keep your GPU from maxing out at 100% which is good for smoothness I believe.  Just s guess but based on what I notice when running unlimited versus locked at 30.  I find overall smoothness is better w/ a frame lock set.

[downscc 1,327]

I get best results with frame rate lock set at 24, even at 30 my frame rate would drop drastically with AS16 overcast/undercast clouds but at 24 there is no drop at all... always rock solid and smooth.

 

I have 6700K OC 4.7GHz 3300MHz DDR4 and 980Ti with 4K TV at 30 Hz.

[theskyisthelimit 48]

I get best results with frame rate lock set at 24, even at 30 my frame rate would drop drastically with AS16 overcast/undercast clouds but at 24 there is no drop at all... always rock solid and smooth.

 

I have 6700K OC 4.7GHz 3300MHz DDR4 and 980Ti with 4K TV at 30 Hz.

 

Ill definitely be trying locked frames to reduce GPU overhead and probably reduce the artifacts i only seem to see since moving from the 980ti to 1080 (flashes of pixels)..

 

I cant get a solid overclock with the new hardware.. 6700k + asus a170-a + vengeance lpx 3200 ram (at 2133 and 1.36 volts set in bios).

 

No matter what i do, from 1.29 through 1.40 and various LLC attempts.. it always ends with either errors in P95 26.6 blend or realbench 15 minute test errors or intel burn test after say 20 rounds erroring..

 

I'm thinking its a ram issue perhaps.. unsure how to check.. the old memtest v4.0 opening 8 of them and dividing up the ram comes to mind.. .. been trying that for about 1 hour at 30% so far no errors.  Maybe needing to Tune VCCIO and System Agent voltages (1.15 to 1.25) and maybe dram to 1.40, havent tried this yet.

 

My current test set has no xmp, vcore at 1.38, LLC at 1 (i've had the vcore lower and llc up to 5 in some tests).. ram set to 1.36 and 2133.    I have the new gskill cas15 coming friday as well.

When i'd test with this setup, prime95 26.6 would crash after 20 minutes.. realbench reported an error after its 15 minutes

[downscc 1,327]

My favorite basic cpu and memory test utility is the Intel XTU.  In my experience, it's stress and heat are a better representation of my usage whereas others seem to hit it as hard as they can.

 

Available from Intel here: https://downloadcenter.intel.com/download/24075/Intel-Extreme-Tuning-Utility-Intel-XTU-