2006 CPU on 2016 Engine
Or
2016 CPU on 2006 engine.
Which of the two alternatives is best for game quality.
^_^
Old Hardware On New Software
Moderator: Ras
-
Nordlandia
- Posts: 2845
- Joined: Fri Sep 25, 2015 9:38 pm
- Location: Sortland, Norway
-
Jesse Gersenson
- Posts: 593
- Joined: Sat Aug 20, 2011 9:43 am
Re: Old Hardware On New Software
Which is to say:
Rybka 2.2 64-bit on a 18-core xeon
VS.
Komodo 9.3/Stockfish 7 on a 4-core X5355
Rybka 2.2 64-bit on a 18-core xeon
VS.
Komodo 9.3/Stockfish 7 on a 4-core X5355
-
Vinvin
- Posts: 5335
- Joined: Thu Mar 09, 2006 9:40 am
- Full name: Vincent Lejeune
Re: Old Hardware On New Software
For software : with 4 cores, about 470 Elo difference (may be more if Rybka 2.2 scales badly with so many cores).Jesse Gersenson wrote:Which is to say:
Rybka 2.2 64-bit on a 18-core xeon
VS.
Komodo 9.3/Stockfish 7 on a 4-core X5355
For hardware : very rough estimation : 8 times the speed means ~ 400 Elo difference.
-
Jesse Gersenson
- Posts: 593
- Joined: Sat Aug 20, 2011 9:43 am
Re: Old Hardware On New Software
Taking performance number from cpubenchmark.net it looks like the speed difference between the cpus is 2.7 doublings, or maybe 200 elo; not 400 elo.Vinvin wrote:For software : with 4 cores, about 470 Elo difference (may be more if Rybka 2.2 scales badly with so many cores).Jesse Gersenson wrote:Which is to say:
Rybka 2.2 64-bit on a 18-core xeon
VS.
Komodo 9.3/Stockfish 7 on a 4-core X5355
For hardware : very rough estimation : 8 times the speed means ~ 400 Elo difference.
Also the modern programs will perform better at lower depth than the older program which'll further lower the importance of the 2.7 doublings.
-
Nordlandia
- Posts: 2845
- Joined: Fri Sep 25, 2015 9:38 pm
- Location: Sortland, Norway
Re: Old Hardware On New Software
It appears new software somehow compensates for old processors.
However i'm not sure how an engine in 2020 scales on an 2006 CPU.
However i'm not sure how an engine in 2020 scales on an 2006 CPU.
-
Vinvin
- Posts: 5335
- Joined: Thu Mar 09, 2006 9:40 am
- Full name: Vincent Lejeune
Re: Old Hardware On New Software
Oops, bad calculation from meJesse Gersenson wrote:Taking performance number from cpubenchmark.net it looks like the speed difference between the cpus is 2.7 doublings, or maybe 200 elo; not 400 elo.Vinvin wrote:For software : with 4 cores, about 470 Elo difference (may be more if Rybka 2.2 scales badly with so many cores).Jesse Gersenson wrote:Which is to say:
Rybka 2.2 64-bit on a 18-core xeon
VS.
Komodo 9.3/Stockfish 7 on a 4-core X5355
For hardware : very rough estimation : 8 times the speed means ~ 400 Elo difference.
Also the modern programs will perform better at lower depth than the older program which'll further lower the importance of the 2.7 doublings.
*8 means *2*2*2 means 50+50+50 = +150 Elo
-
acase
- Posts: 1004
- Joined: Wed Mar 15, 2006 1:14 am
- Location: Columbus, Ohio USA
- Full name: Andrew R Case
Re: Old Hardware On New Software
Hello Jon,
Nice poll, I think the software advances are more important myself, and certainly enough to overcome the hardware advantage. The programmers have done an excellent job of advancing the engines, and in my opinion the engines have evolved faster and better than the hardware advances.
It should be pointed out also that some of the older programs support only a limited amount of cores (not the case with Rybka) and a limited hash table size.
Nice poll, I think the software advances are more important myself, and certainly enough to overcome the hardware advantage. The programmers have done an excellent job of advancing the engines, and in my opinion the engines have evolved faster and better than the hardware advances.
It should be pointed out also that some of the older programs support only a limited amount of cores (not the case with Rybka) and a limited hash table size.
-
IanO
- Posts: 503
- Joined: Wed Mar 08, 2006 9:45 pm
- Location: Portland, OR
Re: Old Hardware On New Software
This poll is already partially answered in the SSDF rating list data.