Hyperthreading off = stutters off.

[LoBiSoMeM]

Originally Posted by Whiski

If I turn off HT now, will I have to turn it back on once we get multi-core support?

Whiskey

Multicore processing isn't the same as Intel HyperCrap!

You have an AMD CPU, you don't have HyperCrap!

[yakaddict]

Originally Posted by LoBiSoMeM

Multicore processing isn't the same as Intel HyperCrap!

You have an AMD CPU, you don't have HyperCrap!

Hypercrap? I smell a flame war in the kindling...

[LoBiSoMeM]

Originally Posted by yakaddict

Hypercrap? I smell a flame war in the kindling...

No flame. We today have X4, X6, X8 CPUs, and this proprietary tech from Intel have issues regards performance with some apps. We AMD customers don't need to bother

[ZaltysZ]

I don't see any difference with HT on and HT off in CoD. In fact, none of my sims get stutters related to HT.

There is little utility here: http://technet.microsoft.com/en-us/s.../cc835722.aspx , which will show cpu mapping if it is run with "-c" parameter.

Usual mapping for i7 9xx with HT is: A=[0,1], B=[2,3], C=[4,5], D=[6,7]. This means that virtual CPUs 0 and 1 are in fact a single hyperthreaded core and so on. This is most common mapping, but it is done by BIOS, so mainboard manufacturer still has ability to make a surprise. It is important to know the mapping surely if you are going to use HT with manually set affinity masks.

HT is pretty nice thing, but it requires certain conditions for noticeable performance gains, whose are usually achievable only by specially designed applications, however even non HT aware applications usually benefit from HT to some extent.

Most problems start when two logical cores, belonging to same physical core, are treated as fully separate physical cores. I.e., if two heavy threads are put on cores 6 and 7 (same physical core), then they will run very slow, if compared to situation when they are put on cores 4 and 6 (2 separate physical cores). This must be always be taken into account, when setting affinity masks, and setting manual affinity masks is a must sometimes (i.e. thread scheduler likes to condense everything on to single physical CPU, while thinking it is 2 CPUs).

Those who have stutters with HT on, have you tried running with HT and custom affinity mask? I.e. putting CoD on cores 2, 4, 6 (counting from 0).

[335th_GRAthos]

Wow, thank you Zalty for this information.

Until now, we had Process Affinity =
=1 - core 0
=2 - core 1
=3 - core 0+1
=4 - core 2
=5 - core 0+2
=6 - core 1+2
=7 - core 0+1+2
=8 - core 3
=9 - core 0 + 3
=10 - core 1 + 3
=11 - core 0 + 2 + 3
=12 - core 2 + 3
=13 - core 0 + 2 + 3
=14 - core 1 + 2 + 3
=15 - core 0 + 1 + 2 + 3

I am very bad with (binary maths) so, what number do we need in order to make the game use cores 0+2+4+6 ????

~S~

[Helrza]

Originally Posted by 335th_GRAthos

Wow, thank you Zalty for this information.

Until now, we had Process Affinity =
=1 - core 0
=2 - core 1
=3 - core 0+1
=4 - core 2
=5 - core 0+2
=6 - core 1+2
=7 - core 0+1+2
=8 - core 3
=9 - core 0 + 3
=10 - core 1 + 3
=11 - core 0 + 2 + 3
=12 - core 2 + 3
=13 - core 0 + 2 + 3
=14 - core 1 + 2 + 3
=15 - core 0 + 1 + 2 + 3

I am very bad with (binary maths) so, what number do we need in order to make the game use cores 0+2+4+6 ????

~S~

By manually changing the process affinity mask threw the taskmanager.

[ZaltysZ]

Originally Posted by 335th_GRAthos

I am very bad with (binary maths) so, what number do we need in order to make the game use cores 0+2+4+6 ????

In affinity mask each bit represents one CPU: 1 value - use CPU, 0 value - don't. Bits are counted from right to left, so you need do the same with CPUs. In our case, that would be: 7, 6, 5, 4, 3, 2, 1, 0. Now replace wanted CPUs with 1s and unwanted with 0s. You will get: 0, 1, 0, 1, 0, 1, 0, 1 or simply 1010101. Convert that from binary to decimal (use Windows calculator) and get a mask of 85.

[klem]

Originally Posted by ZaltysZ

In affinity mask each bit represents one CPU: 1 value - use CPU, 0 value - don't. Bits are counted from right to left, so you need do the same with CPUs. In our case, that would be: 7, 6, 5, 4, 3, 2, 1, 0. Now replace wanted CPUs with 1s and unwanted with 0s. You will get: 0, 1, 0, 1, 0, 1, 0, 1 or simply 1010101. Convert that from binary to decimal (use Windows calculator) and get a mask of 85.

Is it better to assign CoD to the first threads of each core in this way or to turn off Hyperthreading in BIOS and assign it to all cores (=15).

And why?

[335th_GRAthos]

Originally Posted by ZaltysZ

In affinity mask each bit represents one CPU: 1 value - use CPU, 0 value - don't. Bits are counted from right to left, so you need do the same with CPUs. In our case, that would be: 7, 6, 5, 4, 3, 2, 1, 0. Now replace wanted CPUs with 1s and unwanted with 0s. You will get: 0, 1, 0, 1, 0, 1, 0, 1 or simply 1010101. Convert that from binary to decimal (use Windows calculator) and get a mask of 85.

[Standing Ovation]

Thanks!
(we started with numbers 1,2,3,4 some years ago; then moved to 15 six months ago; now 85! This is progress my friends! )


I have been flying yesterday with HT = off
I must say I did not have any stutters but, I also realised I never had any stutters before.
So, I wanted to turn HT=on again and test but I also wanted to control better the use of CoD on the cores (and identify how much CPU-time is used by other processes).

Interestingly enough, I had exactly the same concern as SEE (the NV threaded optimisation usage).

~S~

[CaptainDoggles]

Originally Posted by ZaltysZ

In affinity mask each bit represents one CPU: 1 value - use CPU, 0 value - don't. Bits are counted from right to left, so you need do the same with CPUs. In our case, that would be: 7, 6, 5, 4, 3, 2, 1, 0. Now replace wanted CPUs with 1s and unwanted with 0s. You will get: 0, 1, 0, 1, 0, 1, 0, 1 or simply 1010101. Convert that from binary to decimal (use Windows calculator) and get a mask of 85.

For those who enjoy math, the conversion from binary (base 2) to decimal (base 10) works out like this:

1010101 = 1*2^0 + 0*2^1 + 1*2^2 + 0*2^3 + 1*2^4 + 0*2^5 + 1*2^6
        =   1   +   0   +   4   +   0   +   16  +   0   +   64
        = 85