Ivanhoe plays singlecore in TCEC 13 !!

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Guenther
Posts: 3111
Joined: Wed Oct 01, 2008 4:33 am
Location: Regensburg, Germany
Full name: Guenther Simon
Contact:

Re: Ivanhoe plays singlecore in TCEC 13 !!

Post by Guenther » Tue Aug 07, 2018 4:49 pm

Uri Blass wrote:
Tue Aug 07, 2018 3:34 pm
lucasart wrote:
Tue Aug 07, 2018 3:30 am
Joost Buijs wrote:
Sat Aug 04, 2018 7:12 am
In the past when Martin was still in charge of TCEC each engine was thoroughly checked before it was entered in the tournament.
Now it seems more like playing Pacman. TCEC will lose a lot of credibility with all these nonsense.
TCEC tournament director is totally incompetent. Was already the case with Martin, is even worse with Anton. They haven't written any code, neither chess engines, nor the website, nor the cutechess backend. They just collect advertising revenue, and sell VIP status to trigger-happy morons for a fee.

What we need instead are programmers that understand what they are doing…

PS: DeusX, really ? It's just LC0 with random changes by someone who has no clue what he is doing. How does that qualify ? Wasn't there supposed to be some clone filtering rule ?
I do not think that it was ok to accept DeusX but I do not think that the changes are random changes by someone who has no clue what he is doing.
True, because there were no changes.
Current foe list count : [101]
http://rwbc-chess.de/chronology.htm

Uri Blass
Posts: 8611
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Ivanhoe plays singlecore in TCEC 13 !!

Post by Uri Blass » Tue Aug 07, 2018 5:53 pm

Guenther wrote:
Tue Aug 07, 2018 4:49 pm
Uri Blass wrote:
Tue Aug 07, 2018 3:34 pm
lucasart wrote:
Tue Aug 07, 2018 3:30 am
Joost Buijs wrote:
Sat Aug 04, 2018 7:12 am
In the past when Martin was still in charge of TCEC each engine was thoroughly checked before it was entered in the tournament.
Now it seems more like playing Pacman. TCEC will lose a lot of credibility with all these nonsense.
TCEC tournament director is totally incompetent. Was already the case with Martin, is even worse with Anton. They haven't written any code, neither chess engines, nor the website, nor the cutechess backend. They just collect advertising revenue, and sell VIP status to trigger-happy morons for a fee.

What we need instead are programmers that understand what they are doing…

PS: DeusX, really ? It's just LC0 with random changes by someone who has no clue what he is doing. How does that qualify ? Wasn't there supposed to be some clone filtering rule ?
I do not think that it was ok to accept DeusX but I do not think that the changes are random changes by someone who has no clue what he is doing.
True, because there were no changes.
No changes in the binary but there were changes in the network and not random changes and I do not agree that Albert Silver had no clue what he is doing.

syzygy
Posts: 4458
Joined: Tue Feb 28, 2012 10:56 pm

Re: Ivanhoe plays singlecore in TCEC 13 !!

Post by syzygy » Tue Aug 07, 2018 6:18 pm

Joost Buijs wrote:
Tue Aug 07, 2018 5:12 am
syzygy wrote:
Mon Aug 06, 2018 10:43 pm
Joost Buijs wrote:
Sat Aug 04, 2018 4:31 am
Yesterday I was watching the game between Deus-X and Tucano and noticed that Tucano shows an incredibly low speed (around 1 mnps, sometimes even lower), so I suspect it was only using one core. Also the configured hash (1024 Mb) seems very low for a lazy-smp engine with these long thinking times. Maybe misconfiguration or is this normal behaviour for Tucano? I don't know.
Don't attribute to misconfiguration what can be explained by false (or inappropriate) sharing of cache lines.

This may also be the problem with Ivanhoe.
Tucano seems to show n/s of the main thread only, so it is impossible to guess how many cores it uses without looking at the task-manager.
Is it a fact that Tucano shows nps only of the main thread or is this mere deduction from what TCEC is showing?
Ivanhoe (same version) runs 15 times faster on my own machine, so it is clear that it is only using 1 core at TCEC.
Do you have a dual-cpu machine?

sedicla
Posts: 146
Joined: Fri Jan 07, 2011 11:51 pm
Location: USA

Re: Ivanhoe plays singlecore in TCEC 13 !!

Post by sedicla » Tue Aug 07, 2018 11:09 pm

syzygy wrote:
Tue Aug 07, 2018 6:18 pm
Joost Buijs wrote:
Tue Aug 07, 2018 5:12 am
syzygy wrote:
Mon Aug 06, 2018 10:43 pm
Joost Buijs wrote:
Sat Aug 04, 2018 4:31 am
Yesterday I was watching the game between Deus-X and Tucano and noticed that Tucano shows an incredibly low speed (around 1 mnps, sometimes even lower), so I suspect it was only using one core. Also the configured hash (1024 Mb) seems very low for a lazy-smp engine with these long thinking times. Maybe misconfiguration or is this normal behaviour for Tucano? I don't know.
Don't attribute to misconfiguration what can be explained by false (or inappropriate) sharing of cache lines.

This may also be the problem with Ivanhoe.
Tucano seems to show n/s of the main thread only, so it is impossible to guess how many cores it uses without looking at the task-manager.
Is it a fact that Tucano shows nps only of the main thread or is this mere deduction from what TCEC is showing?
Ivanhoe (same version) runs 15 times faster on my own machine, so it is clear that it is only using 1 core at TCEC.
Do you have a dual-cpu machine?
Yes, I can confirm that tucano is showing the node count from the main thread. This should be fixed in the next release.
Also the hash is limited to 1024 mb. Also to be addressed.

Regards.

Joost Buijs
Posts: 990
Joined: Thu Jul 16, 2009 8:47 am
Location: Almere, The Netherlands

Re: Ivanhoe plays singlecore in TCEC 13 !!

Post by Joost Buijs » Wed Aug 08, 2018 9:37 am

syzygy wrote:
Tue Aug 07, 2018 6:18 pm
Joost Buijs wrote:
Tue Aug 07, 2018 5:12 am
syzygy wrote:
Mon Aug 06, 2018 10:43 pm
Joost Buijs wrote:
Sat Aug 04, 2018 4:31 am
Yesterday I was watching the game between Deus-X and Tucano and noticed that Tucano shows an incredibly low speed (around 1 mnps, sometimes even lower), so I suspect it was only using one core. Also the configured hash (1024 Mb) seems very low for a lazy-smp engine with these long thinking times. Maybe misconfiguration or is this normal behaviour for Tucano? I don't know.
Don't attribute to misconfiguration what can be explained by false (or inappropriate) sharing of cache lines.

This may also be the problem with Ivanhoe.
Tucano seems to show n/s of the main thread only, so it is impossible to guess how many cores it uses without looking at the task-manager.
Is it a fact that Tucano shows nps only of the main thread or is this mere deduction from what TCEC is showing?
Ivanhoe (same version) runs 15 times faster on my own machine, so it is clear that it is only using 1 core at TCEC.
Do you have a dual-cpu machine?
No, I run it on an Intel compute stick.

Joost Buijs
Posts: 990
Joined: Thu Jul 16, 2009 8:47 am
Location: Almere, The Netherlands

Re: Ivanhoe plays singlecore in TCEC 13 !!

Post by Joost Buijs » Wed Aug 08, 2018 2:00 pm

syzygy wrote:
Tue Aug 07, 2018 6:18 pm
Joost Buijs wrote:
Tue Aug 07, 2018 5:12 am
syzygy wrote:
Mon Aug 06, 2018 10:43 pm
Joost Buijs wrote:
Sat Aug 04, 2018 4:31 am
Yesterday I was watching the game between Deus-X and Tucano and noticed that Tucano shows an incredibly low speed (around 1 mnps, sometimes even lower), so I suspect it was only using one core. Also the configured hash (1024 Mb) seems very low for a lazy-smp engine with these long thinking times. Maybe misconfiguration or is this normal behaviour for Tucano? I don't know.
Don't attribute to misconfiguration what can be explained by false (or inappropriate) sharing of cache lines.

This may also be the problem with Ivanhoe.
Tucano seems to show n/s of the main thread only, so it is impossible to guess how many cores it uses without looking at the task-manager.
Is it a fact that Tucano shows nps only of the main thread or is this mere deduction from what TCEC is showing?
Ivanhoe (same version) runs 15 times faster on my own machine, so it is clear that it is only using 1 core at TCEC.
Do you have a dual-cpu machine?
Seriously now, I have several machines but I stopped building dual CPU machines a long time ago. IMHO dual CPU machines are not worth the price difference, usually you get 1.5 times the performance for 3 times the cost. Pretty soon the AMD 32 core thread-ripper arrives and I would not be surprised if it blows the TCEC machine out of the water for a fraction of the cost.

Not that I will buy one, I never had good experience with AMD, either mainboards or video-cards. Intel will come with an answer on thread-ripper for sure and I'm going to wait for that.
Last edited by Joost Buijs on Wed Aug 08, 2018 2:16 pm, edited 1 time in total.

Milos
Posts: 3387
Joined: Wed Nov 25, 2009 12:47 am

Re: Ivanhoe plays singlecore in TCEC 13 !!

Post by Milos » Wed Aug 08, 2018 2:16 pm

Joost Buijs wrote:
Wed Aug 08, 2018 2:00 pm
syzygy wrote:
Tue Aug 07, 2018 6:18 pm
Joost Buijs wrote:
Tue Aug 07, 2018 5:12 am
syzygy wrote:
Mon Aug 06, 2018 10:43 pm
Joost Buijs wrote:
Sat Aug 04, 2018 4:31 am
Yesterday I was watching the game between Deus-X and Tucano and noticed that Tucano shows an incredibly low speed (around 1 mnps, sometimes even lower), so I suspect it was only using one core. Also the configured hash (1024 Mb) seems very low for a lazy-smp engine with these long thinking times. Maybe misconfiguration or is this normal behaviour for Tucano? I don't know.
Don't attribute to misconfiguration what can be explained by false (or inappropriate) sharing of cache lines.

This may also be the problem with Ivanhoe.
Tucano seems to show n/s of the main thread only, so it is impossible to guess how many cores it uses without looking at the task-manager.
Is it a fact that Tucano shows nps only of the main thread or is this mere deduction from what TCEC is showing?
Ivanhoe (same version) runs 15 times faster on my own machine, so it is clear that it is only using 1 core at TCEC.
Do you have a dual-cpu machine?
Seriously now, I have several machines but I stopped building dual CPU machines a long time ago. IMHO dual CPU machines are not worth the price difference, usually you get 1.5 times the performance for 3 times the cost. Pretty soon the AMD 32 core thread-ripper arrives and I would not be surprised if it blows the TCEC machine out of the water for a fraction of the cost.

Not that I will buy one, I never had good experience with AMD, either being it mainboards or video-cards.
There will not be 32 core thread-ripper that blows TCEC machine out or the watter. TDP is the problem. Moore's law is dead, scaling brings almost nothing to CPUs so no way to squeeze 32 cores with manageable TDP (under 300W) that will run on any reasonable frequency (read over 3GHz).
I have a dual CPU system just upgraded it to 2680v2 from 2670. So 20 cores of 5 years old CPUs (each one only 10% worse than mighty i7 6950X) that perform on par or better than any 20 core v4 Xeon for 1/10 of the cost.

Joost Buijs
Posts: 990
Joined: Thu Jul 16, 2009 8:47 am
Location: Almere, The Netherlands

Re: Ivanhoe plays singlecore in TCEC 13 !!

Post by Joost Buijs » Wed Aug 08, 2018 2:58 pm

Milos wrote:
Wed Aug 08, 2018 2:16 pm
Joost Buijs wrote:
Wed Aug 08, 2018 2:00 pm
syzygy wrote:
Tue Aug 07, 2018 6:18 pm
Joost Buijs wrote:
Tue Aug 07, 2018 5:12 am
syzygy wrote:
Mon Aug 06, 2018 10:43 pm
Joost Buijs wrote:
Sat Aug 04, 2018 4:31 am
Yesterday I was watching the game between Deus-X and Tucano and noticed that Tucano shows an incredibly low speed (around 1 mnps, sometimes even lower), so I suspect it was only using one core. Also the configured hash (1024 Mb) seems very low for a lazy-smp engine with these long thinking times. Maybe misconfiguration or is this normal behaviour for Tucano? I don't know.
Don't attribute to misconfiguration what can be explained by false (or inappropriate) sharing of cache lines.

This may also be the problem with Ivanhoe.
Tucano seems to show n/s of the main thread only, so it is impossible to guess how many cores it uses without looking at the task-manager.
Is it a fact that Tucano shows nps only of the main thread or is this mere deduction from what TCEC is showing?
Ivanhoe (same version) runs 15 times faster on my own machine, so it is clear that it is only using 1 core at TCEC.
Do you have a dual-cpu machine?
Seriously now, I have several machines but I stopped building dual CPU machines a long time ago. IMHO dual CPU machines are not worth the price difference, usually you get 1.5 times the performance for 3 times the cost. Pretty soon the AMD 32 core thread-ripper arrives and I would not be surprised if it blows the TCEC machine out of the water for a fraction of the cost.

Not that I will buy one, I never had good experience with AMD, either being it mainboards or video-cards.
There will not be 32 core thread-ripper that blows TCEC machine out or the watter. TDP is the problem. Moore's law is dead, scaling brings almost nothing to CPUs so no way to squeeze 32 cores with manageable TDP (under 300W) that will run on any reasonable frequency (read over 3GHz).
I have a dual CPU system just upgraded it to 2680v2 from 2670. So 20 cores of 5 years old CPUs that perform on par or better than any 20 core v4 Xeon for 1/10 of the cost.
Moore's law is dead, I agree, but still CPU's are getting less power hungry by each new generation. 300W TDP is not very difficult to manage, I've had AMD video-cards drawing 300W from the PS and they were air cooled with just a single fan. If a Noctua NH-D15 can't cope with 300W, you can always switch to water-cooling.

I know that you can buy occasion Xeon CPU's on eBay for peanuts, even complete servers, but you never know in what condition these CPU's are, the servers they sell make so much noise that you don't want to have that in your living environment.

There is also the scaling issue, for chess a machine with 10 cores at 4 GHz. performs better than a machine with 20 cores at 2.5 GHz.
My 6950X CPU runs without any pain at 4.4 GHz. but then it gets somewhat hot under full load (80 to 90 deg. C.) so I usually run it at 3.8 GHz. to keep the temperature below 50 deg. C.
Last edited by Joost Buijs on Wed Aug 08, 2018 3:00 pm, edited 1 time in total.

whereagles
Posts: 561
Joined: Thu Nov 13, 2014 11:03 am

Re: Ivanhoe plays singlecore in TCEC 13 !!

Post by whereagles » Wed Aug 08, 2018 3:00 pm

Moore's law is dead but survived far longer than expected, really.

Milos
Posts: 3387
Joined: Wed Nov 25, 2009 12:47 am

Re: Ivanhoe plays singlecore in TCEC 13 !!

Post by Milos » Wed Aug 08, 2018 3:10 pm

Joost Buijs wrote:
Wed Aug 08, 2018 2:58 pm
Milos wrote:
Wed Aug 08, 2018 2:16 pm
There will not be 32 core thread-ripper that blows TCEC machine out or the watter. TDP is the problem. Moore's law is dead, scaling brings almost nothing to CPUs so no way to squeeze 32 cores with manageable TDP (under 300W) that will run on any reasonable frequency (read over 3GHz).
I have a dual CPU system just upgraded it to 2680v2 from 2670. So 20 cores of 5 years old CPUs that perform on par or better than any 20 core v4 Xeon for 1/10 of the cost.
Moore's law is dead, I agree, but still CPU's are getting less power hungry by each new generation. 300W TDP is not very difficult to manage, I've had AMD video-cards drawing 300W from the PS and they were air cooled with just a single fan. If a Noctua NH-D15 can't cope with 300W, you can always switch to water-cooling.

I know that you can buy occasion Xeon CPU's on eBay for peanuts, even complete servers, but you never know in what condition these CPU's are, the servers they sell make so much noise that you don't want to have that in your living environment.

There is also the scaling issue, for chess a machine with 10 cores at 4 GHz. performs better than a machine with 20 cores at 2.5 GHz.
My 6950X CPU runs without any pain at 4.4 GHz. but then it gets somewhat hot under full load (80 to 90 deg. C) so I usually run it at 3.8 GHz. to keep the temperature below 50 deg. C.
OC-ing is nice, and that is advantage of only single CPU system but still those 3.8GHz is not all cores frequency but single core boost. When you are running it for chess on all cores frequency goes down. In your case you may at best get 3.4GHz which is just 10% over 3.1GHz all core frequency that you get from 2680v2.
Yes you might have an issue when buying these old Xeons from ebay or aliexpress, but normally if it is a regular sample taken from server machines chance is pretty high you'd get a fully working CPU that can run for many years more. Old servers are bad for not only terrible noise but also power/heat and especially reliability so I would never buy those.

Post Reply