As far as HT ON/OFF and OOMs ... more cores = more threads doing work = more memory used and the more work the main thread will have to do to manage/sync the additional threads.

 

This might explain why I have never had an OOM with any version of P3D v2 with my 4670k; even after making multiple flights over the FTX suite (including regional scenery) plus changing planes between flights.  Very interesting...  

 

Ryan

An AM is required with HT=ON to control heat and prevent bundling on LPs.

 

 

An AM is required with HT=ON to control heat

 

Steve I don't understand this. I always thought that the CPU with all other things being equal would be hotter with HT=on than it would be with HT=off?

 

I guess that With HT=on you mean with an affinity mask that just uses every other LP?

 

Why is it that in the graph above performance is better with HT=off?

My system is now cold enough for 5Ghz. The ambient temp is 5c and the CPU idle temp is 6c.

The good news for anyone at 4.8ghz dreaming of 5Ghz is that the extra 200mhz doesn't really make that much of a

difference with my benchmark flight. Here is the graph:

 

Almost finished here just want to try this OC with reducing my 2400nhz CL9 mem to CL8

Its take a huge voltage increase to do it but I expect it to be cold enough to tomorrow to try it.

It won't increase FPS but I anticipate seeing less stutters. It will be a purely subjective observation since I don't

know how to or want to know how to measure short frames:-)

 

Also I will try a hefty GPU OC just to see what I get a 5Ghz but I doubt it will make a difference.

Hate to leave anything unfinished so here is my final post on the subject:

 

The default Bios is self explanatory CPU is at 3.5ghz. HT is off and and the only addition to the .cgf is FFTF=0.01

 

CPU overclocked to 4.8ghz The improvement is obvious

 

CPU overclocked to 5ghz. There is a slight improvement over 4.8ghz but it takes 1.53vcore to get 5ghz with my CPU.

Couldn't do that without ambient temperature less than about 10c

 

Getting the Case Latency value of the DRAM down is very difficult and requires a large and dangerous voltage increase.

In this case 1.88v and that I would not do it without an ambient temp of less than 5c. It really made no difference to the FPS

but my purely subjective observation is that it made a difference to stutters and an obvious one at that.

 

Next the GTX 780 has 200mhz added to the core and memory. Overclocking the GPU made no real difference at all to performance

except over the new P3DV2 ocean where there is as much as a 30% increase in FPS. But over land it made no difference at all.

Don't know what that says about FPS increases with SLI when and if there is ever a profile but I will believe it when I see it!

 

And Finally DSR, Factor 2.00 and Smoothness 50%

Can't say they are optimal values but the performance hit is I think OK.

But I did notice the it completely wiped out the FPS gain from the OC'ed GPU over

water and I thought that was strange?

 

All the changes are cumulative in the order shown. Sorry the graph is so cluttered

but here is another with the lines staggered just ignore the values down the X axis.

The FPS's at shown on the lines.

 

 

My opinion is that as long as it is cold

enough 5Ghz and CL8 are keepers but the GPU OC is not worth it.

I compared SD and HD textures and there was no performance difference

but HD (2048) looked quite a bit better and did not produce an OOM or

the Ping of death during the very short flight. It did in every example of the graph produce an OOM

at the end of the flight when I clicked reset in the P3D GUI that appears when escape is pressed.

Not so with 1028 testures

 

 

Now to FLY CS 727 from Flightbeam KPHX to KSFO

And all this just to play a game...   Dont get me wrong, I like what Im seeing.

My opinion is that as long as it is cold enough 5Ghz and CL8 are keepers but the GPU OC is not worth it.

Same as for FSX.

Conclusion? There's still a long way to go for P3D despite people claiming a better use of GPU. :ph34r:

 

 

And all this just to play a game...

 

All of that to play a game with 45% better performance than you would get without doing any of it.

 

Same as for FSX.

Conclusion? There's still a long way to go for P3D despite people claiming a better use of GPU. :ph34r:

 

Agreed! Overclocking the GPU only makes a difference if you are flying over the new water P3DV2 but I wounder are they working on a complete replacement for Land textures and Autogen too? The ocean is 2/3 the surface of the earth and P3D still works in the middle of the ocean. I mean the atmosphere are Aircraft are not effected by the fact that there is an entirely new ocean in the sim.

So how do you OC the GPU then? 

So how do you OC the GPU then? 

 

MSI afterburner.

But for FSX or P3D V2 its a waste of effort.

Hey Dave,

 

Did you post all 5 pages of your P3D settings anywhere and what add-ons are part of this testing?

 

GPU usage will vary based on those settings ... and if you use SGSS AA the GPU OC would indeed help as it might pending other settings (resolution, shadows, tessellation, etc.).

 

I'd hate to see folks walking away from your work thinking the GPU has no benefit similar to FSX -- which is just not correct.

 

Cheers, Rob.

Did you post all 5 pages of your P3D settings anywhere and what add-ons are part of this testing?

I can upload the videos of the flights for 5ghz with CL8 and 5Ghz with CL8 and the CPU 200mhz/200mhz OC so You can all see for yourself.

I agree that the GPU OC Had a huge effect on tessellation but unless I am mistaken I think tessellation affects water only in P3D V2. The setting where the dame of all the benchmarks except for DSR I set the resolution to 2715x1527. For the rest it was 1080p I am not complaining I got superb results and am very happy with P3D V2 and now its time to fly. Here are the setting. Not to shabby Ah!

 

I'll post the links for the videos when the upload is complete. Basically what I did is record the flight with the P3D recorder play it back with VC view and record it with shadow play with the FPS counter on. Then at 15 second intervals in the playback of the recording plot the FPS. Everything else is subjective observation.

Here is the benchmark for 5ghz and Case Latency of 8:

http://youtu.be/wdlIJCNUxfE

 

And here is the benchmark with the GPU Overclocked by 200mhz core and mem. Everthing else is exactly the same as above.

http://youtu.be/hmEdgOKDpn4

 

Setting are as indicated in the above post.

 

The addons used:

Milviz 407

Orbx PNW

FSDreamTeam CYVR

ASN

REX4

WOAI

Thanks for the details Dave, what were your GPU loads at the time of testing?

 

Typically if you are hitting 90%-99% GPU load, then a GPU OC will help that out.  SGSS AA is a good example of where GPU OC helps.

 

Tessellation is applied to all terrain which includes land and water ... that's why when nVidia broke tessellation in one of their driver versions you could see all the blue triangles covering both land and water.

 

Cheers, Rob.

what were your GPU loads at the time of testing?

 

Rob, I recorded another playback of the benchmark in a window with task manager and Afterburner running to show whats going on. The CPU is a 5ghz and DRAM at CL8. The GPU is not Overclocked.  You can see that when the aircraft is over open water the GPU is at 100% explaing why the GPU OC really increased the FPS over the water but then when it gets over land it is virtually always around the 50 to 70% mark while explaining why the GPU OC did nothing given that core 0 of the CPU is at 100% all the time:

https://www.youtube.com/watch?v=CrCwNjVlx3w&feature=youtu.be

 

I don't use NI anymore and never really got my head around AA but I was thinking that I might try an affinity mask and split the main thread work over all cores. I remember its not recommended but on my system if I can move some of the Core0 work load to 1,2 and 3 it might make a big difference given my heavily overclocked DRAM. I will try AF=15 for a laugh

 

One other thing I noted above that DSR seemed to wipe out the gain on FPS over water with the GPU OC. I think the simplest explanation for that is that Although I thought It was, The GPU was perhaps not OC'ed at all during that benchmark.