I found something that may have been effecting my results ?Robert Flesher wrote:Houdini wrote:Mark,mwyoung wrote:After loading my new i7 laptop with a few chess engines. I am finding Rybka 4, Fritz 12, and Houdini are running faster using 8 logical cores. This is true with NPS and timed test positions. As this is my first i7 cpu, does Hyper-threading with the i7 help instead of hurt chess engines performance. This was not true in the past. In my testing I am clearly faster with Hyper-threading turned on running my chess engines.
Example Fritz 12
Fritz benchmark 4 cores. 5380 Nps 11.21
Fritz benchmark with HT 8 logical cores. 6995 Nps 14.57
I've never experienced any useful improvement from hyper-threading for Houdini. Even if the nps is slightly higher, the overhead of the additional threads could very well reduce the actual playing strength.
Could you try the "autotune" feature of Houdini? This feature was intended for picking the best Split_Depth parameter, but it also functions as an accurate benchmark for multi-thread node speeds.
Double-click on the Houdini executable to open it in a console window.
Then enter the following commands to "autotune" for 4 threads.Houdini will run for about 10 minutes analyzing a number of positions for different values of split_depth. Make sure you're doing nothing else on your computer, just leave it running for 10 minutes. At the end you'll get a summary of the node speeds.Code: Select all
setoption name threads value 4 autotune
Then repeat the procedure for 8 threads.What results do you get?Code: Select all
setoption name threads value 8 autotune
I also find with my I7 920 that HT enabled produces faster NPS, but also solves best move, or tactical problems faster. Until I see proof that 4 threads is faster ( I cannot see it on my machine) versus 4 actual + 4 virtual threads, HT will remain on. Please explain further, as these are not the results I am finding.
When I was turning off HT I did not notice that Houndini was still set to 8 cores when I rebooted. I assumed it would default back to 4 cores. Therefore, I will test again, but it seems that 4 cores while being slower in nodes is a tad faster at solving problems. I will test it many times, however, but MP CPU random search results makes it very hard to tell.