Just ordered the following for P3D3.3

[Robert McDonald 167]

 

Hi guys,

 

I just ordered the following from newegg, I'll see what this can do and then I'll order the other three to complete the setup.

 

1 - Antec Twelve Hundred V3 Black Steel ATX Full Tower Case

 

1 - MSI X99A GodLike Gaming Carbon LGA 2011-v3 Intel X99   HDMI SATA 6Gb/s USB 3.1 Extended ATX Intel Motherboard

 

1 - Intel Core i7-6950X 25M Broadwell-E 10-Core 3.0 GHz LGA 2011-v3 140W BX80671I76950X Desktop Processor

 

1 - G.SKILL TridentZ Series 64GB (4x16GB) 288-Pin DDR4 SDRAM DDR4 3400 (PC4 27200) Intel Z170 Platform Desktop Memory Model F4-3400C16Q-64GTZ

 

1 - SAMSUNG 850 PRO 2.5" 1TB SATA III 3-D Vertical Internal Solid State Drive (SSD) MZ-7KE1T0BW

 

1 - CORSAIR AX1500i CP-9020057-NA 1500W ATX12V / EPS12V 80 PLUS TITANIUM Certified Full Modular Active PFC 80 Plus Titanium TÜV Modular ATX12V & EPS12V Power Supply Power Supply

 

3 - Super clocked EVGA Titan Xs (until I setup the projectors and other computers, then its back to one)

1 - Noctua NH-D15 SSO2 D-Type Premium CPU Cooler, NF-A15 x 2 PWM Fans

 

1 - ASUS DVD Burner 24X DVD+R 8X DVD+RW 12X DVD+R DL 24X DVD-R 6X DVD-RW 16X DVD-ROM 48X CD-R 24X CD-RW 48X CD-ROM Black SATA Model DRW-24B3ST/BLK/G/AS

 

 

My dream system, at least for the "Server" in P3dv3.3.5  Once I had the 3 client PCs, I would pull the TitanX from the server (wasted GPU) and likely a 770GTX 4GB Msi Gaming card in the server, 3 Titans (1 each) in the 3 clients.  I likely would build Skylake clients OC to about 5ghz.

[SteveW 1,792]

Reports are suggesting the single threaded performance is what you pay for, better than all other CPUs: 5960x@4.3GHz is beaten easily by 6950x@3.9GHz at over 60% faster.

[westman 257]

Steve check your sources please.

 

Single tread performance at same freq

 

Base 3970x

5960x 12%

6950x 20%

6700k 26%

 

Broadwell is Broadwell check the 4core 1150 for example.

Happen to have test a 6950x on a asus V evelution10

 

The benefit is more cores and slightly better IPC but not a mindblowing 70-80% at same freq

[SteveW 1,792]

PC Pro August edition.

[Guest]

Not sure why the focus on single thread performance, P3D is multi-threaded?  Along with P3D's own threading system, add-ons like SODE, Virtuali, ASN, GoFlight, ShadowPlay, MJC Q400, etc. will all run on separate threads/EXE ... as will the OS spread it's ongoing work loads across threads.

 

Keep in mind 5960X and 6950X are all quad channel with 68GB/s vs. Skylake which is 34GB/s memory bandwidth and the L3 Cache is relevant 20-25MB vs 8MB Skylake.  And finally the Skylake has a chunk of it's core setup as a graphics processing unit which is not useful for P3D at all.

 

Is there a point of diminishing returns when it comes to number of cores for P3D, maybe, I haven't tested a 6950X ... but even if there is a point of diminishing returns for cores, one has the option to disable cores on 6950X BUT still keep that 25MB L3 Cache fully operational.

 

Many end users have reported their results of the 6700K under P3D, they are reporting lower numbers in terms of FPS performance (and long frames) under P3D than I'm getting with my 5960X (HT OFF) -- given very similar settings and add-ons and add-on configuration.

 

As far as value, that's really a different topic.  Price to performance ratio, the Skylake is a good choice.  For a multi-channel build it's a good chunk of savings.

 

Cheers, Rob.

[SteveW 1,792]

Not sure why the focus on single thread performance

I presume otherwise we would all run P3D at 3GHz and stop messing with 3.9 to 4.5 overclocks? P3D is multi-threaded yes, very, in fact it will consume every Logical Processor it can lay it's grubby bits on. That, and the problem of HT mode can be tamed with an affinity mask. There's some merit in reducing that ten by a couple of cores with respect to cache I expect.

[SteveW 1,792]

 

Is there an optimum number of cores for a system with P3D? It's easy to makes some tests to see what happens as more cores are utilised by the sim. As cores are added, the scenario loading time shortens, up to a point, then there's no gain, so no point going over that.

[aerostar 42]

I'm wondering if this is a viable method of getting a rough idea of the Single core Overclock performance of CPUs...

 

Take a CPU in this list and divide it's 'score' by the number of cores to get a score per core...  then compare this score with the others to find the best....   looks like from this angle that the 6700K is the winner.

 

https://gyazo.com/7681af7c4caa3574697e079a0c0dabff

 

from Passmark website...

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

 

My 2600k is currently scoring (in my system overclock) around 12400 /4 cores = 3100 per core score

 

I have read in these Forums somewhere that more than 4 cores and FSX/P3d don't like it ...  they tend to generate duplicate threads giving the CPU extra work that is discarded...

 

When I monitor my CPU usage in 'Process Lasso' there is always a core/thread that does most of the work while the others rise and fall when loading scenery etc...  when that core gets swamped then stutters start.  I run with HT on and allocate that P3D main thread to my CPU's core/thread 2 and keep everything else off of core/thread 2 and 3 ( a complete uninterrupted core of the CPU dedicated to the MAIN P3D thread ) and never allocated to core/thread 0 where there are too many interruptions from other processes..

 

I use an 'Affinity Mask' of 84 to get P3D to make it's main working thread on core/thread 2.

 

https://gyazo.com/bf05ff6ac7af9645c478ebe4b1fa1405

 

Process Lasso website...

https://bitsum.com/

 

At the moment most of my stutters when they occur are due to the GPU getting swamped.....

GTX980ti @ ~1450Mhz

 

https://gyazo.com/f47a98d2cedd82b72b8d52c5919e91e9

[Mike_CFII_MEL 788]

Hi,

 

Time will tell, just working out arraignments with Jetline to get the 6950X to them... Maybe even let them set it up water cooled like the other three computers they are building for me.

 

We will see.

[SteveW 1,792]

The single threaded performance talked about is really a layman's term. I simply threw that in as it's a matter of interest with respect to this new ten core CPU.

 

What engineers are discussing is the number of instructions that can be completed in a given time on a core. They are not discussing a single threaded application.

 

When a core can run at 3.9GHz and complete more instructions than an other design of core running at 4.3GHz, this is said to be a faster CPU. Also this more efficient core requires less energy and makes less heat to complete the same instruction count.

 

What I have not inferred is that single threaded performance is a requirement or not with regard to running P3D.

 

Apps like P3D can utilise more than one logical processor so that other tasks are carried out on other processors. This is to take load off of the main thread which can require high speeds. Many tasks across many processors can eventually become a burden for these type of apps.

 

Another kind of app may use many processors deliberately to brute force work something out that can be split into many similar tasks. For example a cipher might be broken by applying many copies of little apps across many processors all trying different codes. These type of applications do scale well with additional cores and are designed to do well that way.

 

When we enable hyperthreading those tasks become two per core unless an affinity mask is used to mask off one LP per core. We have to be careful because we may not want to double the tasks per core. This is especially true with the first task on the first logical processor encountered by P3D. In this case we can mask off one of the LPs of the first core encountered.

[aerostar 42]

I have the opinion that FSX / P3D predominately use one core for the main running of processes related to the core calculations of the sim...  all the forces acting on the aircraft , environmental factors, NAV system calc , etc etc..  whereas they have manged to kind of separate the scenery loading and a few other things over other cores it would disjoint processes if they did that to the main core engine.

 

 To split a string of tasks coming into the engine between cores and then reprocess them to feed to the GPU is probably more work and load on the CPU and would cause all sorts of 'Interrupts' for access to the CPU causing stutters etc.

 

The results in the Passmark utility are showing the capability of the CPU of processing data.  With FSX / P3D that data needs to be processed FAST , not wider bandwidth and slower (More slower cores).

 

To me, my basic approach with the graphs seems to show that a fast single core speed CPU is best for FSX /P3D when flying.

 

The number of instructions processed by a single core are surely represented by the score... the fastest core with the higher score does the most processing?

[SteveW 1,792]

Quite right. When working out the physics, the sim says my current fps is 30, for the next frame where will the terrain be in 1/30s? Then calculates for that location and if the frame is not displayed on that 1/30s period, the terrain is in the wrong place. The Fixed fps slider setting ensures that each period between frames is the same, so long as it can be maintained. However with that a series of look-ahead frames are calculated. Mostly this setting will continue to make next frames never easing up unless the look ahead buffer is filled. With the Unlimited setting P3D will settle the fps on the monitor refresh period if that can be maintained, or a limit set in NI can act like a monitor refresh period with the VSync setting. This VSync is the sync between the renderer and the GPU, not to be confused with the old vsync of the flyback trace with cathode ray tube monitors. When the VSync in P3D is enabled, we also have access to the Triple Buffer, this enables the sim to keep rendering, never letting up unless held back by the refresh period and the triple buffer full. Therefor many will find the Unlimited setting works well with a 30Hz monitor. Any of these situations can pull down the requirement for the sim to be rendering all the time and can show as less than 100% on the first sim job. A different machine may always show 100% for that thread since it never gets to sleep. Even a tiny hint of sleep will reduce the graph to 60%, does not mean it's going idle.

[gboz 157]

Problem.. plenty of theory but not enough bench marking.

We really need the tech gurus to perhaps officially covert the old FSXIIMark bench mark over to P3D

to get some hard data on what makes a real world difference here.

Perhaps add time frame analysis to the current FPS only.

 

gb.

[SteveW 1,792]

 

P3D comes as a professional package. Unlike FSX and FSX-SE, the AffinityMask setting with P3D provides full control of how the app will utilise the cores, since P3D will make a job on each and every LP if not contained. With FSX the logical processors unmasked in the AM are loaded with the main sim jobs, then the affinity is expanded to utilise all LPS. With FSX-SE the AM is ignored it does it's own thing but can be steered by an external app. If we enable hyperthreading and do not use an affinity mask with P3D the first two cores will contain the first four jobs of P3D. More than four P3D jobs simply pull in scenery faster, up to a point. So it can be seen that the entire sim is being run on only two cores, or rather four logical processors. The sim works well even like that until things get heavy. We need to ensure that these sim jobs are getting the bandwidth they desire, especially the first one as aerostar pointed out.

 

The current best practice is to ensure the main sim job, the renderer, works on a core to itself, may as well be core zero, and that addon exe's start above core zero as in Robs list, Q400 etc. The sim jobs other than the main job can mingle with other processes like these exe's because in a way these background jobs of the sim take seconds to complete, whereas the renderer has to keep going whatever state of play the scenery loaders are at.

 

The sim can be tested with a stopwatch to see how long it takes to load a scenario, and a package like fraps can look at the stability within the fps trace. Or a professionally written software tool can artificially load the sim with scenery and other objects, and create simconnect clients and measure the throughputs of cores and trace the fps stability at the same time for a real understanding of the performance.

[darth_damian_000 8]

I would say the ASUS DVD burner is the most important component, not only due to its importance to the nature of modern file transfer and storage but also the fact that you have listed every possible specification of the product. tehehe

 

seriously though, good build.