V2- Still no quad core or SLI optimization

[WarpD 824]

Also I guess there were something in the background as well that caused the other cores to work, but I repeated the process multiple times and always the same result one core maxed out th rest is below 40ish percent. In comparison fsx maxes out 3 cores at AM 14 and this is how it should be, the workload should be equally divided between the cores.

 

Sent from my GT-N7105 using Tapatalk

I highlighted one part of your post... which just simply isn't true.  That is what you may think "is how it should be" but there should really not be multiple cores maxed out... just shouldn't be.  Nothing about this sim could legitimately require maxing out multiple cores the entire time.  You keep forgetting that a great deal of the graphics processing was moved to the GPU which will reduce CPU load.

[EGLD 439]

I for one am very grateful of earlier adopters that are out spending money on P3D2 and video cards, running through configurations and reporting what works and doesn't work. For those of us on the sidelines observing, you are the pioneers that will allow the rest of us to make informative decisions when the time comes to adopt P3D2.

 

Indeed!

 

I'm hovering over the Buy button on a 4gb 770, but want to make sure it's going to be worthwhile...

 

 

• 1 posts
• Joined 11 Years, 8 Months and 17 Days Ago.

 

:lol:

[ntme9 3]

WarpD, on 26 Nov 2013 - 05:24 AM, said:

The CPU isn't the bottleneck, and just to be really clear... you're comparing simplistic games against a complex simulation.  It's not a valid approach at all.  Aside from displaying graphics, they have little else in common code-wise.  This is not a first-person-shooter and it's world is global, not a single game out there has the same size scenery by any stretch of the imagination.

 

I disagree. If gpu usage is not at 100% then the cpu, ram or vram is holding it back. In this case it is the cpu holding it back. There seems to always be at least 1 thread that is maxed out. I believe only 1 thread handles cpu to gpu data. The sim needs to be fixed to utilize more cpu power for cpu to gpu data.

 

Just because the sim is large doesn't mean much. It only loads a predetermined amount of area around you.

[sweoggy 1]

Intel's Core i7, i5, and i3 series can handle 2 threads per core*, so any i7 quad-core model could handle 8 threads.

*hyper-threading, but it's rather limited in its utility.

 

AMD's Athlon, Phenom, Sempron, and Opteron chips, on the other hand, are designed around a 1-thread-per-core model.

 

But, you know that already I'm quite sure.

 

EDIT: struck out incorrect information

 

Even that is incorrect. There's CPUs which belongs to the i5 and i3 series which has HT (hyper threading) as well. This is usually the case for i5 mobile CPUs found in laptops, 2 cores, 4 threads.

[WarpD 824]

I disagree. If gpu usage is not at 100% then the cpu, ram or vram is holding it back. In this case it is the cpu holding it back. There seems to always be at least 1 thread that is maxed out. I believe only 1 thread handles cpu to gpu data. The sim needs to be fixed to utilize more cpu power for cpu to gpu data.

 

Just because the sim is large doesn't mean much. It only loads a predetermined amount of area around you.

First, you edited out my final line, I don't know why... don't like what I stated regarding amount of video memory being the primary issue for performance???  Second, the sim loads far more scenery than any game you play, period.  If you set visibility so that it's completely clear, the sim is probably loading up hundreds of square miles worth of scenery imagery/textures.  Add to that the mesh, environment, clouds, etc, etc... it easily eclipses the size of absolutely any game you're trying to compare it with graphics wise.

[TrailBlazer 0]

I highlighted one part of your post... which just simply isn't true. That is what you may think "is how it should be" but there should really not be multiple cores maxed out... just shouldn't be. Nothing about this sim could legitimately require maxing out multiple cores the entire time. You keep forgetting that a great deal of the graphics processing was moved to the GPU which will reduce CPU load.

You have a point there BUT

If a program is multicore aware, whenever one core has no resources left it should start the take the extra resources from the second and then from the third etc. Until the affinity mask says that is all you can get. What happened here is it maxes out the the one cpu core and if you pushhing the sttings higher it starts to reduce fps due to bottlenecking and the load on other coees does not change at all.

 

In fsx it's different, it keeps loading up the cores until all three is 100% and then the frames started to drop.

 

These observations are based on my tests where every eyecandy that is gpu related and not present in fsx was turned off in p3d so that the comparison is better.

 

In summary to me p3dv2 has much worse multi core utilization than fsx. Correct me if my methidology is not good somewhere.

 

Sent from my GT-N7105 using Tapatalk

 

 

[ntme9 3]

WarpD, on 26 Nov 2013 - 06:32 AM, said:

First, you edited out my final line, I don't know why... don't like what I stated regarding amount of video memory being the primary issue for performance???  Second, the sim loads far more scenery than any game you play, period.  If you set visibility so that it's completely clear, the sim is probably loading up hundreds of square miles worth of scenery imagery/textures.  Add to that the mesh, environment, clouds, etc, etc... it easily eclipses the size of absolutely any game you're trying to compare it with graphics wise.

I apologize, I had to run and only had time to reply to that part.  

 

Yes, I agree with you that vram can be a issue. I have been seeing 2.25gb usage (1080p maxed). I think its a good idea to have at least 3gb for vram not to be a issue at 1080p.

 

Its not only the size that matters its how it looks (that's what she said). Games like BF4, GTA4, and Crysis 3 can use over 2gb of vram also. How? All of these sims, (bless them) (I do love them but) they are utter 1998 graphics equivalent xxxxx when compared to modern games. Let me explain, for example when you walk past plants and grass it moves as you brush by it! Compare that to the 4 sided green pixel sticks they call trees in P3D/FSX/XPLN. Again, I love these sims but don't act like they are the only ones possessing a bunch of information.

[ntme9 3]

Well this sim officially has me scratching my head. I am now consistently seeing 99% gpu usage.

 

I was testing all day yesterday and hardly ever saw it past 70% usually50-60%

 

It seems to have to do with clicking with the mouse anywhere on the inside of the window. Even if you go full screen (which is how I was testing) it runs at 70% UNTIL you click inside the window with the mouse then bam 99% 

 

So make sure you click the sky!  

 

EDIT: ok I found that the above is only true if terrain cast shadows are on! With them off gpu usage is back to 60s

[nhall72 0]

 

 

In summary to me p3dv2 has much worse multi core utilization than fsx. Correct me if my methidology is not good somewhere.

 

This is an oversimplification of multithreaded software behavior.  It's much more complex than you describe. 

 

You seem to be obsessed with the utilization of your CPU and/or GPU but there's a lot more to a computer than just these two components.  The bottleneck will always be somewhere...it's just a question of where.  At any given moment the software is doing a variety of different tasks included reading data from your hard drive, accessing RAM, rendering and moving data to the GPU over the system bus, etc.  Even this isn't a complete description but the point is that any of these things can and will cause variations in utilization of any one component at any given moment.

 

If you really want to get to the bottom of it then you need to ask yourself some more questions.  What is the transfer rate of your hard drive?  What is your system bus speed?  How fast is reading and writing to your RAM?  What kind of processor are you using and how big is it's L1 and L2 cache?  What is the transfer rate over the system bus from system RAM to GPU memory?  This is only a start.

 

While all of this is very interesting, I would say that instead of keeping my eyes glued to the processor utilization I'm going to take a look at my screen.  Is it smoother, or not?  On my screen...it is.

[jabloomf1230 2,445]

The bottleneck will always be somewhere...it's just a question of where. 

 

True. No matter how many times somebody makes this point, there are still people in the community that refuse the grasp this concept. It would be the most unusual app that will max out both one's GPU and one's CPU simultaneously at 100%. Even forgetting about RAM or the disk subsystem, no matter how tightly an app is coded, at some point the CPU needs to wait for the GPU or vice versa.

[styckx 1,871]

True. No matter how many times somebody makes this point, there are still people in the community that refuse the grasp this concept. It would be the most unusual app that will max out both one's GPU and one's CPU simultaneously at 100%. Even forgetting about RAM or the disk subsystem, no matter how tightly an app is coded, at some point the CPU needs to wait for the GPU or vice versa.

 

I knew before even P3D v2 was released not only would the GPU be the new bottleneck but specifically the VRAM on the GPU..  Gamers (I know that is a sin in this section but) knew this limitation was coming for awhile now hence 3-4GB GPU's are more common on the Nvidia 700 series. 600 series had them but they were EXPENSIVE in comparison to their 2GB brothers at the time.   Textures are growing in size, and the GPU is inherently more powerful than a CPU is at a lot of tasks and developers are leaning on it to do more work than ever..  Gotta look at it this way.. Where have CPU's advanced in the past 5 years? No where really.. They've practically plateaued.. GPU's on the other hand are still shooting for the stars...

 

Long story short.. Being GPU dependent is a hell of a lot better than CPU dependent..   B)

[nhall72 0]

Long story short.. Being GPU dependent is a hell of a lot better than CPU dependent..

 

Agreed.  By far, the best scenario is that the GPU is the bottleneck.  I need to spend more time with Prepar3D but I'm very encouraged.  I think the team at Lockheed have done a great job redesigning the pipeline to offload much more to the GPU.  At the very least it's a huge step forward and should give us much better scalability for the future.

[TheGrunt 44]

Some people seem to have strange ideas that mulithreading or multicore support is just a decision to make in software which does bunch of stuff on real time (ie. game, flight simulator etc.). There are very few games out there right now which utilize even four cores optimally. Those using more are even more rare. Some dual core Core i3 is still quite good gaming platform. We may see improvement on this area when the new consoles really kick in and programmers may have to think differently to get maximum out of those APU-processors which employ quite a lot less power than best of the desktop CPUs.

 

For something like games (and P3D/FSX too) benefits of multithreading can decrease quite rapidly. Why? Because most of the stuff that you can multithread has to be in sync all the time. When multithreading you always get some threads which are just waiting for the heavier threads to finish. At the same time those threads may be so light, that their overall CPU load is really low. Now you have possibly complicated your code a lot without benefitting in performance. Multithreading is not just a decision you make and boom - you have a software which utilizes your top of the line CPU computes with all its cores and everyone is happy. It depends on the software how you can utilize it.

 

Besides P3D V2 still is based on the old code and after a certain point you can just as well code a whole new platform. Also after those modifications we may lose most of the backwards compatibility because the changes would modify the core components so heavily that same type of solutions than before aren't even applicable anymore. I'm also pretty sure that for example AI in FSX is not heavy on CPU itself: the .bgl based AI routines seem quite simple. Heaviness comes from the model rendering that just loads CPU too much. P3D V2 could be a lot better in this area, but I haven't tested the AI yet to see the differences.

 

One more thing. I really would hope that people wouldn't see task manager and make assumptions about how many threads there are at a certain time or how many CPUs software utilizes effectively. It really isn't good at that all. For example, If you see two CPUs showing pretty much identical loads, it is very likely that there is one thread that scheduler bounces between those two cores. Update time of the task manager and the time it takes a thread to be computed is so small, that task manager doesn't keep up with it. Of course it gives roughly the picture how software loads your CPU, but it really isn't very precise.

[e345spd 0]

"those APU-processors which employ quite a lot less power than best of the desktop CPUs."

 

 

Not exactly sure exactly how much you know about semi-conductors and architectures as a whole, however you VASTLY over-estimate both the integer and floating-point compute performance of 'best of the desktop CPU' in comparison to the jaguar console APUs. The above is especially true in terms of real-time performance, since you don't have 8 gb ddr5 with a unified memory controller to reduce the time spent on consuming transfers and the desktop CPUs are hampered by poorly optimized APIs. 

 

There's no magic 'utilize more instructions per clock' with games when you have no additional execution resources....

 

 

 

 

No, hyper-threading will never scale in a game over 20% absolute maximum and even then would be indicative of code that cannot utilize the pipeline as effectively as possible on a per thread basis. Stop using task manager percentages, they are meaningless.  

 

 

 

 

 

 

 

 

Almost everyone (fantasy latest console build included) is GPU limited. Excellent work, LM, excellent work. 

 

I assume FSX was highly reliant on CPU FPU, a really ridiculous but common occurrence with older engines. Hell, maybe it was even running fully depreciated x87 code like the engine used for the 'skyrim' game.

 

 

Anyways, they have done a great job.

 

 

 

Really, the image quality (plus pretty impressive shadows) and smoothness speaks for itself. 99.99% sure FSX will never be launched again. 

 

I wonder how heavy the draw calls are....pretty heavy I would think....Possibly many of us are rather API limited.