i9 7900X vs. i7 5960X - P3D V4 performance results 85% FPS increase

[WebMaximus 372]

It has been true for many, many years and I think it's still very true you won't get much more in terms of performance by upgrading your hardware if you already have a fairly descent setup.

Sure...you might be able to increase a couple of settings a few notches and run at a higher resolution etc but in order for us to see a true improvement the application needs to change fundamentally.

That we now finally have 64-bit is a good first step in that direction and let's keep our fingers crossed we will have flight simulation software in x years that will utilize our hardware in ways we can only dream of today.

[Guest]

5 hours ago, Cruachan said:

And look at the price of the thing?!! As I said, they're havin' a laugh at our expense! 

More cores = more profit per cm of wafer space.  It's all about profit and always will be, AMD are no different as they are going the same route.  Higher frequency CPUs requires more wafer space as they need to deal with things like electromigration.  More wafer space and less cores = less profit.

5 hours ago, Vali said:

As Threadripper started this approach with more cores (at a good price) and Intel follows it, I expect that the developers will start that too now..

So for now, it comes more to the apps and game developers.

 

Multi-core CPUs have been around for a long long time, not all paradigm (aka flight simulation) fits into a threaded application so it's NOT just a matter of developers coding the app. 

4 hours ago, Potroh said:

I get the EXACT same FPS as I used to with the former setup.
 

I'd like to see your data and test scenarios on that.  I provided all my data at the start of this thread for others to examine and presented the results, but I have no experience with the 4xxx CPUs.

3 hours ago, Potroh said:

I think Rob's original enthusiasm was a bit exaggerated, although I never had his previous CPU, so can't really tell.

Data is all there for everyone to examine, no exaggeration.

2 hours ago, Tony P said:

Exactly,  we should all hope that LM is working on spreading out the work on ALL cores and we can stop seeing the first core pegged at 100%.  I mean the whole point behind multi-threads is to actually take advantage of the technology and spread the load over all cores.  ( ie X-Plane)

X-Plane makes less use cores than P3D in the testing I've done ... P3D is using all cores:

The paradigm of flight simulators will always (without exception) require one core to synchronize all the other threads or else your vitural world would be a jumble of objects not making an "physical" sense and arrive and finish at different times.  Hence one core does more work than the others. I keep repeating this, but I guess folks just don't wanna believe ;)

Xplane 11 core usage (but keep in mind I'm not doing much in the Xplane video):

 

1 hour ago, WebMaximus said:

Sure...you might be able to increase a couple of settings a few notches and run at a higher resolution etc

But that's exactly why I upgrade my hardware, those few notches and higher resolution make a world of difference visually.

Cheers, Rob.

[WebMaximus 372]

21 minutes ago, Rob Ainscough said:

But that's exactly why I upgrade my hardware, those few notches and higher resolution make a world of difference visually.

Sure, increasing settings and especially a few of them can make quite a difference visually as you say but question is if that difference relates to the cost upgrading your hardware.

All I'm trying to say is that IMO we are still waiting for a new type of software when it comes to flight simulation that will use our hardware in a much more efficient way. Look at the number of cores for instance and HT - how much of a difference does that really make with our current type of flight sim software...?

[Guest]

2 hours ago, WebMaximus said:

new type of software when it comes to flight simulation that will use our hardware in a much more efficient way

The only "new" type of computing I'm aware of is quantum computing and I'll be honest, wrapping my head around that is difficult.  Quantum computing is a reality, but I'm not sure if it'll make it into flight simulators nor if it's actually well suited to work the flight simulator paradigm.

DX12 EMA SFR is the only current tech that I'm aware of that would introduce significant improvements but requires considered code changes to implement:

Quote

Then there's EMA, which has two modes: linked or unlinked mode. Linked mode requires GPUs to be close to the same hardware, while unlinked—which is what Ashes uses—allows any mix of GPUs to be used. The whole point of this, and why this works at all under DX12, is to make use of Split Frame Rendering (SFR). This breaks down each frame of a game into several tiles, which are then rendered in parallel by the GPUs. This is different to the Alternate Frame Rendering (AFR) used in DX12, where each GPU renders an entire frame each, duplicating data across each GPU.

In theory, with EMA and SFR, performance should go way up. Plus, users should benefit from pooling graphics memory (i.e. using two 4GB GPUs would actually result in 8GB of usable graphics memory).

I'm not aware of ANYONE (development team) doing DX12 EMA SFR ... but that's where you'll see significant FPS gains.

Cheers, Rob.

[WebMaximus 372]

OK, well...let's see what the future has in store for all of us flight simmers.

What I meant was that the way it is now spending lots of $ on new hardware to replace your CPU and GPU with the latest models only to be able to gain maybe a couple of FPS and/or to be able to increase some of the sliders a notch or two might not be worth it.

This is course is very much a personal thing and what you find worth the money. In my case for instance I'm running a 6700K @ 4.5 GHz and a 980Ti. Sure I could replace my rig with one of the latest CPUs and replace my 980Ti with a 1080Ti but IMO the performance benefit doing so wouldn't even be close to worth the money that would set me back.

[Guest]

2 minutes ago, WebMaximus said:

gain maybe a couple of FPS and/or to be able to increase some of the sliders a notch or two might not be worth it.

This is course is very much a personal thing and what you find worth the money.

I tend to avoid discussions of "worth" and/or "value", because like you suggest, it's a "personal thing" and "worth/value" can really only be define by oneself, not others.

Cheers, Rob.

[Cruachan 705]

"The paradigm of flight simulators will always (without exception) require one core to synchronize all the other threads or else your vitural world would be a jumble of objects not making an "physical" sense and arrive and finish at different times.  Hence one core does more work than the others. I keep repeating this, but I guess folks just don't wanna believe ;)"

I believe, Rob! Also it makes perfect sense.

However, I do wonder how this downtrodden single core will cope with the ever increasing number of threads charging towards it (simultaneously?) each demanding preferential attention. With Flight Simulation I'm speculating that this might restrict the overclock or risk chip failure due to uncontrollable heat and throttling back before 'digital burnout' occurs?

Again it's just speculation, but we might reach a point, if it hasn't been reached already, where we have to acknowledge a sweet spot for the core count that ensures optimum efficiency of operation.

Mike

[Guest]

23 minutes ago, Cruachan said:

but we might reach a point, if it hasn't been reached already

For gaming/simulation I would prefer a 6Ghz 4 core CPU over my 5Ghz 10 core CPU.  I'd also like a 4 Core CPU with MUCH Larger L2 and L3 Cache.  Intel and AMD could produce such CPUs, but like I said, AMD/Intel seem to believe there is no "good" profit in such a venture.  I'm actually a little disappointed AMD appear to be going the "Intel" route with "more cores" instead of higher frequency.

I'm not really sure why this is the case, perhaps AMD/Intel think most gamers/simmers use Consoles?  It's as if AMD/Intel think that PCs are only good for video rendering and/or 3D animation ... if you look at the benchmark products they tout like Cinebench and Adobe CC as a performance mark, it's even more clear AMD/Intel believe PCs aren't for gaming.  From the data I've seen, PC Gaming since 2012 has been the larger and more dominant gaming platform with projections that PC gaming is going to increase market share:

Cheers, Rob.

[sfm48 5]

I'm running a 4930k six core i7 with a gtx1080 ti and find it runs best on all cores using an affinity mask of 1365 to use the physical cores (excluding the hyperthreading portion) although I leave hyper-threading on in bios. Seems if core 0 (zero) is not used then I get blurries. The same seemed to be true on my old 3820 four core in that core 0 had to be in the mix to avoid blurries (i.e AF setting = 85). Not sure if this is to to the way p3d is coded or if this is architectural due to the way each core accesses other cores data / memory... Maybe the new sky lake processors uses better method of cross core communication to avoid latency??? Maybe that is why the new 7900x is performing great ( in addition to single core performance).

[Potroh 14]

3 hours ago, sfm48 said:

Maybe that is why the new 7900x is performing great ( in addition to single core performance).

The 7900x performs great because it has 10 cores and its inbuilt "turbo speed" speed is of 4.3GHz, way ahead of the earlier series.
The 7900X has 10MB of L2 and 13.75MB of L3 cache for a 23.75MB total.

One can disable the cores in Bios one by one and I did that test, with only 4 cores enabled you more or less get the known 4xxx performance.

In P3D it made no difference here if I enabled HT, the result was only I had to reduce overclocking due to the excess heat.
But the 7900x performs fabulously, I've never experienced such a smooth and stutter-free P3D performance ever...

[Cruachan 705]

1 hour ago, Potroh said:

But the 7900x performs fabulously, I've never experienced such a smooth and stutter-free P3D performance ever...

Perhaps you didn't have the opportunity of trying the i-7 5960X. It too continues to be a great performer. Of all my builds this one stands out as being the one that leaves me utterly content. I imagine many others will feel the same with their builds. These days, and thankfully, the incentive to constantly upgrade as we chase performance nirvana is far less compelling.

Mike 🤓

[Vali 83]

Yesterday I started additional tests with my system.

First, I used Cinebench to see if my values for i7-4790K (running at 4.4 Ghz, this is the max turbo frequency) match with those on the internet and they do (when HT is ON, otherwise they are lower but I suppose this is how they do it also).

Then I used settings from Rob but I mixed a little bit the autogen and shadows (between commercial day and commercial night settings, I took the commercial day for autogen and commercial night for the shadows). Also as DL was ON, I choosen only 2xSSAA instead of 4xSSAA. The idea is that I do not always want to switch the profile depending by the time of day, even if it is not such a big matter but when I fly on Vatsim and is an evening flight, I do not want to risk to change the settings during flight.

Test was done with the PMDG 747 (all the lights ON) at EDDF Professional at night time (as this stressed the most the system).

At the gate, my 1080TI GPU is clearly the most stressed component, 100% load. Is this also on your system Rob? (or who has also the Pascal,  1080TI paired with 7900x). Frames are between 24-26, I saw even less sometimes and I have stutters. Monitor at 30Hz. I noticed if I increase the GPU core freq from 5000 Mhz to 5200 Mhz experience seems a little bit smoother (temp is 72 degrees Celsius).

As soon as I take off then CPU load and autogen come into place and CPU goes often to 100% for 3 cores, and the first core from time to time. Fluctuation is quite significant. Frames are around 30 fps, this time, going to 26 sometimes.

So on my side, GPU being at 100% is I think an issue which LM should improve, with DL. Other than this, it has no issues.

With the CPU though.. in order to sustain the autogen settings and the rest, I probably need an update (I can try also to overclock it but.. temperature goes higher and I do not want a custom water cooling, using only Kraken X41 for the moment). On a new processor I will take the X62.

Also the DDR3 memory is a minus (G.Skill F3-1600C7D-16GTX, 32 GB in total).

 

And this is what CPU benchmark on a system predicts:

http://www.userbenchmark.com/PCBuilder/Custom/S59466-M233971.251565.19728.102222vsS62720-M325982.251565.19728.102222?tab=RAM

 

[Cruachan 705]

Hi Valentin,

Have you tried locking your frame rates? I have mine locked in-sim at 31fps with VSync OFF and rarely see my 1080Ti usage run much higher than 70%. If I run unlimited then, yes, the GPU quickly maxes out, and generates more heat, as it attempts to pump out as many frames as is achievable for a particular scenario...and that's when I start to see some distracting stuttering. BTW I continue to use 4xSSAA in-sim with 4XSGSSAA in NI and the 1080Ti handles this combo very nicely.

Regards,

Mike

[MarkW 1,491]

Anyone have a guess as to how the Coffee Lake chips that will be released in October will stack up.  6 cores / 12 threads.  4.3 Ghz and 12MB Cache ?  Apparently the top chip, i7 8700K will sell for around $350 USD.

[Vali 83]

I disabled today one core of the 4790K. Instead of 4 I was using 3. Who thinks it made a difference?

Well, I had 2 fps more as average... (both tests were with 1080p for 1080TI).