What more can I do?

There's more on my site in the forums, maybe something there can help you further. Plenty of comparisons made accurately showing HT on outperforms HT off. You should be able to show that HT enabled outperforms HT disabled just the same, if not then maybe there is something you missed. Let's face it, Intel make it obvious in their literature that HT enabled outperforms HT disabled, and it costs extra $$$. Surely, it's up to you to find out how to get that. I'm confident I've done enough hand-holding, and explaining at length how some apps need "steering" with HT on. But I'm afraid it is over to you now.

 

When I said five cores will never outperform four, there's a reason for it, and I've not come clean on that. It's why six cores work better than five, so I can leg-pull with that. Six cores always show better performance than five but not four, due to the way the sim works. It would be a good exercise for you to demonstrate that even with HT disabled you should see it. If you can do that, or prove the performance advantage of HT on, you would be getting somewhere.

 

Whether HT enabled or not, each core will run many threads on a Windows PC, and this means much time saved task switching.

 

The jobscheduler allocates a time slice to each thread. The CPU stops and stores the results of the current thread, loads up the results of the previous thread and restarts that thread. That's called switching time. Run two equal full bore threads on one core and they take time switching during each time slice.

 

With HT enabled there's hardware on the CPU that does more work, it means those two threads don't consume switching time, but they have to be managed onto each LP of the core.

 

This diagram below helps explain how HT improves performance. On the left the HT enabled pair of threads finish earlier, but they have to be on separate LPs of that core. The non-HT pair or threads on the right take extra time to complete due to the gaps between time slices:

 

 

 

 

Poor testing provides poor results. Example:

 

2. Ensure other programs such as those that restrict traffic and inject weather start with their own Affinity Mask by using batch files.

This is one area where you can be going wrong. Do you know those Ai injected are the same each test? As time progresses, so to schedules.

 

I've said it enough already - you need guaranteed repeatability, or your results are void.

 

(By the way Mike. IF10 culls Ai traffic you won't get to see. You probably only want one app to do that. In IF10 you can turn off Full AI Traffic Optimisation in the Ai Selection window).

 

"So, Steve, can you provide some examples of what would qualify, as you see it, as being a 'poor setup'? "

 

I remember a few weeks ago on PM with you Mike, I explained you need to get a baseline. That is, if you can run the sim and PC with no other accoutrements and get a reading. That could be from a GPU tool that saves an fps trace. Drop that into Excel and compute the delta for it there. Instead just run tests with IF10 alone, since it's been designed as a test harness for the purpose. Ai injection, Weather, and everything else is guaranteed identical per run, that's what it does. Introducing any other dll, addon, app, or exe, into the system can throw the results. It's no good me doing traces for you, you got to do that. Get a baseline, IF10 will do that for you, then bring in addons and other stuff see what affect it has on your baseline. I did all that while we were chatting, and finished before you went out.

 

You talk of micro-stutter? But when performance testing you won't be looking at the screen because it's not set up for viewing. You are setting up to produce maximum throughput. With Unlimited and no vsync it's probably aweful, you can set that up with AA after getting you bearings.

IF10 culls Ai traffic you won't get to see. You probably only want one app to do that. In IF10 you can turn off Full AI Traffic Optimisation in the Ai Selection window).

 

Hi Steve,

 

Wow! Perhaps it would be much easier for you to tell us what IF10 Pro can't do! I was not aware of this capability and I'll be sure to check it out.

 

I think perhaps we've reached the point when you should be taking a well-earned break from all this. You've spent a great deal of your valuable time providing a wealth of information me (us?) to digest. Without question your posts have created a very valuable resource for future reference. For that I will always be grateful. The rest is up to me now.

 

Many thanks!

 

Regards,

Mike :-)

Hey Mike, I've enjoyed our time in the PMs and forums and appreciate your enquiring interest in the subject. I'm sure I've done enough, but even so, I've thought of a couple more things. Worth thinking about, while in pursuit of understanding this tricky subject, full of conundrums...

 

How about a dual core? HT on or off? What would be the best AM, and what AM for the addons?

 

Why does the AM=116 configuration work better on three cores than AM=84?

 

With eight cores at your disposal, you could run these dual and triple core setups on your last three cores. So the AM=84 would translate to 21504=01,01,01,00,00,00,00,00 and 116=29696=01,11,01,00,00,00,00,00. You should be able to see the 116 configuration perform better than the 84 configuration. Good luck!

 

I've enjoyed our time in the PMs and forums

 

 

Hi Steve,

 

Thanks, me too. This experience continues to be quite fascinating

 

Here is my first attempt to collect some further data. I've tried to follow your instructions and advice as closely as possible, but there are still a few issues I'm unclear about (explained in my PM)

 

First, the Setup:

 

PILOT'S - FS Global 2010 FTX Compatible

FTX: Global BASE

FTX: Global VECTOR

FTX: Europe (Active)

FTX: Global openLC Europe

REX 4-TD with Soft Clouds

 

Nvidia Inspector - Prepar3D (Default Profile)

 

Prepar3D:

FXAA= OFF, MSAA= 4 Samples, Texture Filtering= 16x, VSync= OFF,Triple Buffering= NO

Target frame rate= Unlimited, Road vehicles= 0

 

Ideal Flight 10 Pro (10.015pb11):

AffinityMask=3

CullObjects=True

CullDistance=64000

CullFromCount=30

CullInterval=60

 

Aircraft: Default Beech Baron 58

 

Departure: Cardiff (EGFF), United Kingdom

Weather: Scattered Clouds. Vis > 20Nm, Winds NE 8 Knots, 18C

Taxiing to takeoff, climbout to altitude 2000ft

Autopilot engaged during climbout

 

Until I get a few things sorted out in my mind about the data collection procedure (as explained in my PM I'm doing this differently from the runs carried out in my original post) I've simply included the first set of results - collected over a period of 2 mins. I think you will be pleasantly surprised - I know I was On screen the sim seemed smooth as butter throughout and I was not aware of any stuttering. Also, the visuals were beautiful.

 

 

Regards,

Mike

Looks good Mike, very good in fact, 01,01,01,01,00,00,00,00 that's AM=21760. AM=1360=00,00,01,01,01,01,00,00. 21760 works best if you've got everything on that PC. If you've got networked PCs coming in drop down to 5440=00,01,01,01,01,00,00,00, or 1360, which hand a little more bandwidth priority to those.

01,01,01,01,00,00,00,00 that's AM=21760. AM=1360=00,00,01,01,01,01,00,00.

 

Yes, I realised my mistake after posting. Managed to edit the text in the graph before being timed out by AVSIM. The test was in fact using AM=1360. Will try 21760 tomorrow.

 

Cheers,

Mike

OK Mike, really good work.

 

Just maybe you can eke a little more out of four cores all stacked up the end with 21760, will be difficult to spot any difference with that very stable looking trace above. The other possibility is that can you get more out of six. Five won't work, but maybe six can. I think that would not produce a better stability, and simply use up cores where other stuff would otherwise go. Try AM=21840 6 cores, keep all addons on AM=3, leaves LPs 2, and 3 free. Should be a good test to see if additional cores given to the sim can improve the sim.