HT off, no AM.

We are trying to get away from this mantra, since it is doomed as we take delivery of CPUs with more and more cores, and waste's the user's investment in expensive HT technology.

 

Eventually software will take care of multi cores and LPs, but for the time being we need to worry about the AMs.

 

Sure with 4 cores and no HT we may as well have it at AM=0=(1111)=NoAM.

 

But since we can have an aggressive sawtooth action with an app mixing with the sim main thread, we might need to restrict it to avoid that core the app is using with AM=14=(1110). Actually that's not bad and not far off using all four cores. However it's better to allow the sim to spread into four parts, and so again the benefit with HT enabled is we can use AM=116=(01,11,01,00) to split into four parts, uses three cores more efficiently than HT disabled, as seen quite obviously in the graphs.

 

Even two cores is not far off the performance of three or four, so with HT enabled and Overclocking we can use only two cores with an AM=60=(00,11,11,00) or go slightly higher with 52=(00,11,01,00) at the slight expense of background loading with three jobs. 116=(01,11,01,00) is the best you can do on three cores and 4 cores is not much better so don't worry.

Steve - with all do respect for your brilliant advice and knowledge of the ESP engine (I´m a big fan of your posts!), the HT=off and no AM works absolutely best on my 5ghz 3770k. It is able to run with HT on with the same clock speed etc, but I can´t get rid of these problems:

 

1: Stutters. 

2: Texture blurriness after longer flights (like the sim/terrain engine can´t keep up with the position of the airplane). It looks like P3D "needs" four threads minimum to cope with ORBX and other heavy sceneries. I can´t completely cure this problem without leaving the AM out of the .cfg. 

 

I´ve tried any number of Affinity Masks and combinations, but my machine works best with no AM and HT=off.  (This calculator is brilliant: http://www.gatwick-fsg.org.uk/affinitymask.aspx?SubMenuItem=hardware )

 

I recently did a full reinstall of Windows 10, and secure erased all my SSDs to completely wipe the NANDs, and added a brand new one dedicated to P3D for good measure. I did some similar tests as the OP on a saved flight over EGLL, and took note of FPS and FrameTimes with Fraps. The main cause of stutters and erratic frame times on my machine (given v. 3.1) are: 

 

- Unlocked framerate (way to variable, and not smooth). 

- SLI (using one card is smoother, although not as powerful when entering heavy clouds)

- ORBX FTX Vector (roads seem to create a lot of random looooong frame times in an otherwise smooth flight).

 

The biggest hit on frame rates are by no shadow of doubt detailed scenery - and advanced add ons like PMDG NGX/777. (With a default aircraft, the Baron i.e, and default scenery, the unlocked frame rate makes the sim pump out >70FPS peak, but with ORBX, PMDG and a detailed airport, the FPS is struggling to hit 30).

 

I´ve managed to get the sim run smooth now, with HT off and no AM, no SLI, locked FPS (internally) at 32, and NVI locked at 30 (cures some of the "lag" while panning). Interestingly the same "combo" applied to 3.0 did not work at all - a complete sluggish stutterfest. Worth mentioning - I´m running ActiveSky, PFPX etc on a virtual machine, running on another computer, to save workload on the sim machine.

 

Although AM off works best on my ageing 3770k, I´m not doubting that your statement of using HT=on and an appropriate AM to match the given CPU is right, on the more modern CPUs.

 

OK, so you are saying you get a different result with HT Off and AM=0, and HT On and AM = 85?

 

You'll need to check that first...

I use a I7 4770K running a 4ghz with HT on can anyone tell me if i would be better of with an AF and if so what should it be

 

many thanks

While you're checking that Kris, remember that I'm just stating facts here. For example, there's no, No AM, there's AM=0=(1111) which equates to AM=15=(1111), that's enabling all 4 cores. Then there's the HT equivalents 170 and 85. You should get the same performance. If not you could have a problem of some kind.

 

 

I use a I7 4770K running a 4ghz with HT on can anyone tell me if i would be better of with an AF and if so what should it be

 

many thanks

Discussed a few posts ago - with a 4 core HT On you would do fine with 116. You can try 85 but may interfere with an addon, so try them both. Otherwise not much to think about with the four core.

Oke i have also tested my stutters. This is with framelock 30 in nvidia inspector, vsync and triple on in prepar3d

http://puu.sh/mEczu/dfaeb2c583.png

 

i have this

 

TESSELLATION=1

TEXTURE_MAX_LOAD=1024

TERRAIN_SHADOW_CAST_DISTANCE=2

CLOUD_SHADOW_CAST_DISTANCE=5

OBJECT_SHADOW_CAST_DISTANCE=1

TEXTUREMAXLOAD=27

TEXTURE_BANDWIDTH_MULT=160

LOD_RADIUS=3.500000

TESSELLATION_FACTOR=30

AffinityMask=14

 

As you can see i have a few big spikes. Is this fine or can i improve this? I have an i5-4690k @4.4ghz and gtx 670

 

Steve - with all do respect

Remember that HT on will produce more heat, at 5 GHz you could be pushing that ceiling. Unless I get a chance to troubleshoot your rig, who knows what the issue is. I'm just dropping knowledge on you.

I´ll try to clarify my observations:) With No AM i mean that the [JOBSCHEDULER] part is not in the config, and the main thread is placed on the fist core. Like I said, I´ve tried all combinations of AMs possible for my CPU, being careful not to use an invalid AM for the given state of HT set in the bios.  

 

HT ON: 

 

AF: Not in config: Erratic frame times, and higher VAS usage. 

AF: 85: Erratic frame times, but lower VAS usage. 

AF: 84: Better frame times, same VAS as 85, but slower texture loading, possibly due to one less texture thread. 

AF: 117: Same as 85, maybe a bit better. 

AF: Other variants of 117, where one or two HT-threads are used to give a total of 4 threads to P3D, while leaving Core 0 + HT 0 alone: Same results, pretty much. 

 

HT OFF:

 

AF: Not in config: Smooth. 

AF: 14: Better FPS numbers, but slow texture loading, similar to AF84. I´ve seen the same on my machine since P3D 2.x. 

 

I´ve tried altering the CPU clock etc, to see if the "lower performance with HT on" phenomenon could be caused by an unstable (albeit not noticeable in terms of "normal" instability symptoms), but I´t dosen´t change the impression that HT=off in bios and no [JOBSCHEDULER] is the best solution.

 

 

 

While you're checking that Kris, remember that I'm just stating facts here

 

I absolutely understand - I´m not arguing that. I´m just sharing my observations from my four year old computer:) 

Remember that HT on will produce more heat, at 5 GHz you could be pushing that ceiling. Unless I get a chance to troubleshoot your rig, who knows what the issue is. I'm just dropping knowledge on you.

 

Absolutely. I´ve spent a lot of hours building and clocking over the years. Right now I have a serious overkill custom water loop, with 3 huge rads and copper EKWB blocks on both GPUs and CPU. The temps running P3D in HT on is somewhat 10c higher than HT off, although the temps are peaking about 70 with HT ON, and 60 with HT off. Same VCORE on both. (The VDROOP levelling is a bit less with HT on - due to higher workload). 

funkyj4ever, on 21 Jan 2016 - 4:36 PM, said:

Is this fine or can i improve this? I have an i5-4690k @4.4ghz and gtx 670

With 4 cores no HT you can have 2, 3, or 4, running the sim. so an AM=14=(1110) is 3 cores, and you're avoiding core zero. So if you got some app running there it won't be any trouble. But apps have got to run somewhere. Those settings don't present a problem. Your rig should work great with 2, 3, or 4 cores. Try AM=0=(1111), and AM=12=(1100), obviously, I've lost count of saying the sim runs best with four jobs.

 

kristolz, on 21 Jan 2016 - 4:52 PM, said:

AF: 85: Erratic frame times, but lower VAS usage.

Assuming you are nowhere near the GHz ceiling, what I'm saying is that 4 cores no HT is an answer that cleans up many problems in a HT enabled system without a doubt. That's your experience right now, and the answer to cleaning up the HT act may not be too easy if it was clear anyway. Extremes of systems quite often don't follow the regular course of action anyway. But I'm thinking 4.8GHz and HT enabled with the offending process locked down would outperform that HT off overall.

 

OK Kris,

 

How about this as an example: You have an app running that spawns 2 threads. In HT=Off mode, the second thread starts on the next core. But with HT enabled that second thread may start on the other LP of the one core. So although it does get the benefit of being on two LPs it shares throughput with that core. So when I see posts like yours, I'm fairly sure it's a problem of setup that can be solved, but perhaps not easily. In the example the app could start with an AM=5=(00,00,01,01) solving the problem of starting on the same core. Just like we do with the sim.

Food for thought; if we had a 6 core do we use "NoAM" then? After all, no AM on a 6 core equates to an AM=63=(111111) on an 8 core that's AM=255=(11111111), on a 10 core that's AM=1023. The "No AM" mantra has been debunked, it's history. The HT On/Off situation is manageable if understood, but it's complicated.

 

Overall, on a highly threaded system, HT Off cores are slow in comparison to HT On cores (what else are we paying for?) taking more or less twice as long to swap processes on the core. With 50 threads spawned by P3D, don't worry, there will be plenty of time consuming switching. But HT shares the same cache as two threads on the same core so it's more efficient there too. The extra work done by the HT mode produces more heat, more work=more heat. Unfortunately, it's got to be managed.

am 0

 

http://puu.sh/mEfaH/68d8c33b5c.png

 

am10

http://puu.sh/mEi9a/6468b29adc.png

 

cant see much difference?

One trace looks obviously worse than the other. But like I said many times earlier, it's surprising how much you can get out of even two cores in terms of fps, that's why I measure fps-Delta.

 

 

Still puzzled by HT problems? Here's one way to get better results with HT Off on a four core, than with HT On and AM=85=(01,01,01,01) is likely if you have a program, starts on core zero, and with HT disabled its second thread hits core 1, but with HT enabled it will hit the second LP of core zero. Core zero is hit twice as hard, putting two threads on it, and that is contending with the main sim thread twice as much too. HT on or off we can get the same problems. Our mission, should we choose to accept it, is to get the best out of the chip, that's going to be with HT enabled. Try out the theory by moving apps around with Task Manager, before running the sim.

am 0

 

http://puu.sh/mEfaH/68d8c33b5c.png

 

am10

http://puu.sh/mEi9a/6468b29adc.png

 

cant see much difference?

Another interesting thing you managed to point out is that with no AM and HT enabled, the first four threads of the sim spread onto 2 cores, giving just two cores to the sim, same as AM=10. That's another reason why the results are "similar".

 

...that'll be a bit of a bombshell, no wonder HT has a bad name.

 

 

:excl:  :p0305:    No AM is doomed      :smile:

I use a i5 so i dont have HT

Thanks for your input Steve - it's greatly appreciated:)

 

I might have to make a new 4.8 OC profile and try some lower clock speeds to see if that helps, and give HT=on another go. To be fair - I don't notice any difference between 4.9 and 5 with regards to sim performance (only in benchmarking, but I'm not flying a benchmarking tool.. ), so it might not be that bad after all!