Hyperthreading on or off

[lovetb]

I have a 2 x E5-3696v3 system with 36 cores.

I was playing with Hyperthreading & Ran some benchmarks.

Here are the results.

HT Off 35 threads
command : bench 256 35 24 default depth
===========================
Total time (ms) : 39832
Nodes searched : 2496591345
Nodes/second : 62678031

HT On 70 threads
command :
bench 256 70 24 default depth
===========================
Total time (ms) : 48196
Nodes searched : 3841044383
Nodes/second : 79696331

The depth of 24 was reached faster with HT off, but more nodes were searched with HT on.

We also have syzygy tablebases. So is the depth really that important ?

Your thoughts please.

[Nordlandia]

The consensus say that HT should be off.

«The additional hyper-threads will yield about 25% to 30% extra node speed, but the inefficiency of the parallel alpha-beta search with the higher number of threads will partially offset this speed gain »

See here: http://www.cruxis.com/chess/manual/inde ... gement.htm

[Jouni]

Please run arasan testsuite 4 times with 35 + 70 threads and post results here!

[zullil]

I have a 2 x E5-3696v3 system with 36 cores.

E5-2696 v3?

[lovetb]

I have a 2 x E5-3696v3 system with 36 cores.

E5-2696 v3?

Yes, Typo error. Sorry.

[AndrewGrant]

I have a 2 x E5-3696v3 system with 36 cores.

I was playing with Hyperthreading & Ran some benchmarks.

Here are the results.

HT Off 35 threads
command : bench 256 35 24 default depth
===========================
Total time (ms) : 39832
Nodes searched : 2496591345
Nodes/second : 62678031

HT On 70 threads
command :
bench 256 70 24 default depth
===========================
Total time (ms) : 48196
Nodes searched : 3841044383
Nodes/second : 79696331

The depth of 24 was reached faster with HT off, but more nodes were searched with HT on.

We also have syzygy tablebases. So is the depth really that important ?

Your thoughts please.

Old knowledge is that HT is bad, but that is no longer the case. Recently Noobpwnftw tested 192 core vs 384 thread SF and found an elo gain for using the hyperthreads. I've also tested 16 core vs 32 thread Ethereal (Using Ryzen 1950x) and seen just around +20 elo for using the hyperthreads. Anyone who disagrees these days is either wrong or has an engine with poor SMP.

[Modern Times]

I might be wrong but on the processor side HT has been improving over the years as well which has now swung the argument in its favour ?

[Jesse Gersenson]

As far as I know no tests have been run with HT==on vs HT==off. There are tests, as Andrew points out, which use varying numbers of threads and assume this means HT is off.

If someone has two machines which are the same, we could run a test. I'll see if I can do this.

[lovetb]

Please run arasan testsuite 4 times with 35 + 70 threads and post results here!

Hi, Is there a guide on how to run this test this ?

[Jouni]

Just download test from Arasan chess and run in chess GUI like Arena with "Automatic Analysis" and 30 sec for position.