![[tr]](/mwf/v12/flags/tr.png "Turkey")
Hello there,
As far as i know nowdays we dont have any fritz benchmarks on 16/32 core machines over internet,but anyway i've made some calculations about estimated core difference between 8/16/32 core machines
If we use the formula:
-Intel Quad difference between 1 and 4 core is average = 1.8998
-Intel Xeon difference between 8 and 16 core = 1.6043
-AMD Opteron core difference between 4 and 8 core = 1.6342
Then the Fritz benchmark on 16/32 core machines must be average:
CPU / Computer Core/MHz Total MHz Kn/s
Intel QX6700 Quad 1 x 2660 2660 1903
Intel QX6700 Quad 2 x 2660 5320 3755
Intel QX6700 Quad 4 x 2660 10640 7231
Dual Xeon DP 5160 4 x 3000 12000 8199
Intel XEON X5365 8 x 3000 24000 13154
Intel Xeon X7350 16 x 3000 48000 21102
Athlon 64 X2 3800+ 2 x 2000 4000 2357
AMD Opteron 285 4 x 2613 10452 5998
AMD Opteron 270 4 x 2000 8000 4466
AMD Opteron 885 8 x 2600 20800 9802
AMD Opteron 8350 4 x 2000 8000 4591
AMD Opteron 8350 8 x 2000 16000 7502
AMD Opteron 8350 16 x 2000 32000 12260
AMD Opteron 8350 32 x 2000 64000 20036
Notes:
----------
To be sure 100% for 16/32 core machines,we need real public fritz benchmarks
The Fritz Benchmarks for 1/2/4/8 core machine results are have be taken:
http://www.tomshardware.com/2007/07/16/cpu_charts_2007/page30.html
http://www.jens.tauchclub-krems.at/diverses/Schach/fritz9_benchmarks.html
http://www.sedatchess.com/hardwares.html
Best Regards,
Sedat Canbbaz
As far as i know nowdays we dont have any fritz benchmarks on 16/32 core machines over internet,but anyway i've made some calculations about estimated core difference between 8/16/32 core machines
If we use the formula:
-Intel Quad difference between 1 and 4 core is average = 1.8998
-Intel Xeon difference between 8 and 16 core = 1.6043
-AMD Opteron core difference between 4 and 8 core = 1.6342
Then the Fritz benchmark on 16/32 core machines must be average:
CPU / Computer Core/MHz Total MHz Kn/s
Intel QX6700 Quad 1 x 2660 2660 1903
Intel QX6700 Quad 2 x 2660 5320 3755
Intel QX6700 Quad 4 x 2660 10640 7231
Dual Xeon DP 5160 4 x 3000 12000 8199
Intel XEON X5365 8 x 3000 24000 13154
Intel Xeon X7350 16 x 3000 48000 21102
Athlon 64 X2 3800+ 2 x 2000 4000 2357
AMD Opteron 285 4 x 2613 10452 5998
AMD Opteron 270 4 x 2000 8000 4466
AMD Opteron 885 8 x 2600 20800 9802
AMD Opteron 8350 4 x 2000 8000 4591
AMD Opteron 8350 8 x 2000 16000 7502
AMD Opteron 8350 16 x 2000 32000 12260
AMD Opteron 8350 32 x 2000 64000 20036
Notes:
----------
To be sure 100% for 16/32 core machines,we need real public fritz benchmarks
The Fritz Benchmarks for 1/2/4/8 core machine results are have be taken:
http://www.tomshardware.com/2007/07/16/cpu_charts_2007/page30.html
http://www.jens.tauchclub-krems.at/diverses/Schach/fritz9_benchmarks.html
http://www.sedatchess.com/hardwares.html
Best Regards,
Sedat Canbbaz
![[nl]](/mwf/v12/flags/nl.png "Netherlands")
It's all about how well Fritz Chess Benchmark scales, in the future when 32, 64 and 128 cores are common chess software shall scale close to optimal. Deep Junior used Intel 16 cores in the match against Deep Fritz they can provide a 16 cores benchmark when they run it that time. If I remember it well it was round 24 million nodes per second that suggest round 15% scaling losses from 8 to 16 cores if they used round 3GHz CPU's.
2 CPU Kn/s 4 CPU Kn/s 8 CPU Kn/s
AMD Opteron 8350 8 CPU 8200 16400 0.6 9840 19680 0.6 11808 -23.13%
Intel Xeon X7350 4 CPU 12800 25600 0.6 15360
AMD Opteron 8350 8 CPU 8200 16400 0.65 10660 21320 0.65 13858 -16.72%
Intel Xeon X7350 4 CPU 12800 25600 0.65 16640
AMD Opteron 8350 8 CPU 8200 16400 0.7 11480 22960 0.7 16072 -10.31%
Intel Xeon X7350 4 CPU 12800 25600 0.7 17920
AMD Opteron 8350 8 CPU 8200 16400 0.75 12300 24600 0.75 18450 -3.91%
Intel Xeon X7350 4 CPU 12800 25600 0.75 19200
AMD Opteron 8350 8 CPU 8200 16400 0.8 13120 26240 0.8 20992 2.50%
Intel Xeon X7350 4 CPU 12800 25600 0.8 20480
AMD Opteron 8350 8 CPU 8200 16400 0.85 13940 27880 0.85 23698 8.91%
Intel Xeon X7350 4 CPU 12800 25600 0.85 21760
AMD Opteron 8350 8 CPU 8200 16400 0.9 14760 29520 0.9 26568 15.31%
Intel Xeon X7350 4 CPU 12800 25600 0.9 23040
AMD Opteron 8350 8 CPU 8200 16400 0.95 15580 31160 0.95 29602 21.72%
Intel Xeon X7350 4 CPU 12800 25600 0.95 24320
AMD Opteron 8350 8 CPU 8200 16400 1 16400 32800 1 32800 28.13%
Intel Xeon X7350 4 CPU 12800 25600 1 25600
2 CPU Kn/s 4 CPU Kn/s 8 CPU Kn/s
AMD Opteron 8350 8 CPU 8200 16400 0.6 9840 19680 0.6 11808 -23.13%
Intel Xeon X7350 4 CPU 12800 25600 0.6 15360
AMD Opteron 8350 8 CPU 8200 16400 0.65 10660 21320 0.65 13858 -16.72%
Intel Xeon X7350 4 CPU 12800 25600 0.65 16640
AMD Opteron 8350 8 CPU 8200 16400 0.7 11480 22960 0.7 16072 -10.31%
Intel Xeon X7350 4 CPU 12800 25600 0.7 17920
AMD Opteron 8350 8 CPU 8200 16400 0.75 12300 24600 0.75 18450 -3.91%
Intel Xeon X7350 4 CPU 12800 25600 0.75 19200
AMD Opteron 8350 8 CPU 8200 16400 0.8 13120 26240 0.8 20992 2.50%
Intel Xeon X7350 4 CPU 12800 25600 0.8 20480
AMD Opteron 8350 8 CPU 8200 16400 0.85 13940 27880 0.85 23698 8.91%
Intel Xeon X7350 4 CPU 12800 25600 0.85 21760
AMD Opteron 8350 8 CPU 8200 16400 0.9 14760 29520 0.9 26568 15.31%
Intel Xeon X7350 4 CPU 12800 25600 0.9 23040
AMD Opteron 8350 8 CPU 8200 16400 0.95 15580 31160 0.95 29602 21.72%
Intel Xeon X7350 4 CPU 12800 25600 0.95 24320
AMD Opteron 8350 8 CPU 8200 16400 1 16400 32800 1 32800 28.13%
Intel Xeon X7350 4 CPU 12800 25600 1 25600
Hello,
I've analysed both engines(Deep Fritz10 and
Deep Fritz on 8 core achieved 12751 kN/s while the 16 core Deep Junior achieved 25690 kN/s:
http://www.chessbase.com/newsdetail.asp?newsid=3920
I've analysed both engines(Deep Fritz10 and
Deep Fritz on 8 core achieved 12751 kN/s while the 16 core Deep Junior achieved 25690 kN/s:
http://www.chessbase.com/newsdetail.asp?newsid=3920
Hello,
I've analysed both engines(Deep Fritz 10 and Deep Junior 10.1) exactly at the same position of 52th move:
*Used Hardware:QX6700 Quad 2.66GHz
*Hashtable Size = 256 MB
Deep Fritz 10 t4 = 8168 kn/s
Deep Junior 10.1 t4 = 10861 kn/s
Deep Fritz on 8 core achieved 12751 kn/s while the 16 core Deep Junior achieved 25690 kn/s:
http://www.chessbase.com/newsdetail.asp?newsid=3920
I've analysed both engines(Deep Fritz 10 and Deep Junior 10.1) exactly at the same position of 52th move:
*Used Hardware:QX6700 Quad 2.66GHz
*Hashtable Size = 256 MB
Deep Fritz 10 t4 = 8168 kn/s
Deep Junior 10.1 t4 = 10861 kn/s
Deep Fritz on 8 core achieved 12751 kn/s while the 16 core Deep Junior achieved 25690 kn/s:
http://www.chessbase.com/newsdetail.asp?newsid=3920
AMD Opteron 8350 8 CPU 8200 16400 1 16400 32800 1 32800 28.13%
Intel Xeon X7350 4 CPU 12800 25600 1 25600
A few days ago i was thinking in a such way,but now i have a different opinion
and i think your estimated calculations are wrong for 16/32 core machines
If your calculation system was right,then normally XEON X5365 must be:
CPU / Computer Core/MHz Total MHz Kn/s
Intel XEON X5365 8 x 3000 24000 16398
But the right one is here:
CPU / Computer Core/MHz Total MHz Kn/s
Dual Xeon DP 5160 4 x 3000 12000 8199
Intel XEON X5365 8 x 3000 24000 13154
http://www.jens.tauchclub-krems.at/diverses/Schach/fritz9_benchmarks.html
And therefore 16 core Intel Xeon X7350 must be:
CPU / Computer Core/MHz Total MHz Kn/s
Intel Xeon X7350 16 x 3000 48000 21102
The Kn/s of Deep Junior are average 1.329 higher(depends on position) than Deep Fritz,then
if Deep Junior's on 16 core is 25690 kn/s,then Deep Fritz's kn/s must be 19330 (25690:1.329=19330)
I think after all:the time will tell the true !
Please check carefully for 4 and 8 core differences:
http://www.jens.tauchclub-krems.at/diverses/Schach/fritz9_benchmarks.html
Best,
Sedat
Intel Xeon X7350 4 CPU 12800 25600 1 25600
A few days ago i was thinking in a such way,but now i have a different opinion
and i think your estimated calculations are wrong for 16/32 core machines
If your calculation system was right,then normally XEON X5365 must be:
CPU / Computer Core/MHz Total MHz Kn/s
Intel XEON X5365 8 x 3000 24000 16398
But the right one is here:
CPU / Computer Core/MHz Total MHz Kn/s
Dual Xeon DP 5160 4 x 3000 12000 8199
Intel XEON X5365 8 x 3000 24000 13154
http://www.jens.tauchclub-krems.at/diverses/Schach/fritz9_benchmarks.html
And therefore 16 core Intel Xeon X7350 must be:
CPU / Computer Core/MHz Total MHz Kn/s
Intel Xeon X7350 16 x 3000 48000 21102
The Kn/s of Deep Junior are average 1.329 higher(depends on position) than Deep Fritz,then
if Deep Junior's on 16 core is 25690 kn/s,then Deep Fritz's kn/s must be 19330 (25690:1.329=19330)
I think after all:the time will tell the true !
Please check carefully for 4 and 8 core differences:
http://www.jens.tauchclub-krems.at/diverses/Schach/fritz9_benchmarks.html
Best,
Sedat
In my estimation at http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=26865#pid26865 I used 2 X CPU (8 cores) Intel Xeon X7350 2.93 GHz do 12.800 Kn/s in the Fritz Chess Benchmark as reference and I linear scale it up and wrote "shall do max. ... 25.600 Kn/s" for the 16 core Intel Xeon and "shall do max. ... 32.800 Kn/s" for the 32 core AMD Opteron. "shall do max." means in the case chess engines scales perfect (what they don't do at this moment) and I didn't wrote that the 25.600 Kn/s or 32.800 Kn/s is the outcome of Fritz Chess Benchmark. Also I wrote "real live values can be lower because multiple CPU scaling losses" because hardware itself can also have scaling losses by multiple CPU's this means that the AMD 8 CPU's version shall have probably more scaling losses then the Intel 4 CPU's version and make the speed difference between the two systems (Intel 16 cores/AMD 32 cores) smaller then my optimal estimated 28%.
In my other reply http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=27021#pid27021 I show some values by different overall scaling losses. If the AMD 32 core is a better option than the Intel 16 core depends on what chess software you go to run. I guess the scaling losses shall between the 10% and 25% for good written software and compiled for the right CPU. If this is the case then the AMD 32 core shall be between de -3.91% and 15.31% faster than the Intel 16 core depending what chess software it run.
If you want to spend so much money for chess hardware it's good to do deep research. Personal I shall go for the 16 core Intel system because the most chess engines do not scale well but you can also buy both systems and compare them and rent them for $xx a day to the Rybka, Zappa or other teams when they want to check some openings or do other research or maybe want to use it remote in contests :-)
In my other reply http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=27021#pid27021 I show some values by different overall scaling losses. If the AMD 32 core is a better option than the Intel 16 core depends on what chess software you go to run. I guess the scaling losses shall between the 10% and 25% for good written software and compiled for the right CPU. If this is the case then the AMD 32 core shall be between de -3.91% and 15.31% faster than the Intel 16 core depending what chess software it run.
If you want to spend so much money for chess hardware it's good to do deep research. Personal I shall go for the 16 core Intel system because the most chess engines do not scale well but you can also buy both systems and compare them and rent them for $xx a day to the Rybka, Zappa or other teams when they want to check some openings or do other research or maybe want to use it remote in contests :-)
![[kw]](/mwf/v12/flags/kw.png "Kuwait")
Yes it is impossible to tell at the moment which would be better ... the lower Ghz Opteron 32 core system or the faster Intel 16 core system. I believe that the Intel will be stronger for chess ... that is because the chess engines drop dramatically in efficiency as more cores are added. Maybe one day someone will be able to write more efficient software but that is not the case now. Already there have been tests that prove that the fast Intel V8 8 core is faster that the best 16 core Barcelona. I am sure if you will google it you will find that test. So I would guess it would be the same with the new Xeons ... 16 core also might not beat an overclocked 8 core ... just like an heavily overclocked 4 core can beat a vanilla 8 core. The more cores you have ... the less the scaling efficiency ... so you might be in a situation where less cores running faster is better for chess. Until the programmers find more effective scaling implementations for chess engines that is.
One thing to note is that with more cores ... there is one advantage and that is the "luck" factor. More cores will allow the engine to stumble across a strong variation more than less cores ... although I am not sure how that relates to relative ELO gain.
One thing to note is that with more cores ... there is one advantage and that is the "luck" factor. More cores will allow the engine to stumble across a strong variation more than less cores ... although I am not sure how that relates to relative ELO gain.
Instead of paying approximately 20.000 usd for 32 core (8 x AMD Opteron Quadcore 8350 2.0GHz)
or
for 16 core (4x Inte Xeon X7350 2.93 GHz),
the new Intel Penryn Quad will be much better choise and its will be much cheaper too(Intel's 'Penryn' to launch 11 November)
For example,8 core 4x Intel Penryn Quad can be configurated for 5.000 usd
And the fritz benchmark on 8 core of 4x Intel Penryn Quad 4.0 GHz is estimated to be average 20.000 kn/s
Plus single processor engines will run much faster on a such hardware than on 32 core AMD Opteron Quadcore 8350 2.0GHz
Best,
Sedat
or
for 16 core (4x Inte Xeon X7350 2.93 GHz),
the new Intel Penryn Quad will be much better choise and its will be much cheaper too(Intel's 'Penryn' to launch 11 November)
For example,8 core 4x Intel Penryn Quad can be configurated for 5.000 usd
And the fritz benchmark on 8 core of 4x Intel Penryn Quad 4.0 GHz is estimated to be average 20.000 kn/s
Plus single processor engines will run much faster on a such hardware than on 32 core AMD Opteron Quadcore 8350 2.0GHz
Best,
Sedat
It's at least 6 weeks wait and I don't know if you can overclock a Xeon Harpertown but:
2 x Intel Xeon Harpertown quad core 3.00GHz (80W) estimated score round 14.700 Kn/s in Fritz Chess benchmark
2 x Intel Xeon Harpertown quad core 3.16GHz (120W) estimated score round 15.500 Kn/s in Fritz Chess benchmark
If AMD release higher Opteron clock speeds say 2.5Ghz one month later in December 2007 than 32 cores can be interesting again.
2 x Intel Xeon Harpertown quad core 3.00GHz (80W) estimated score round 14.700 Kn/s in Fritz Chess benchmark
2 x Intel Xeon Harpertown quad core 3.16GHz (120W) estimated score round 15.500 Kn/s in Fritz Chess benchmark
If AMD release higher Opteron clock speeds say 2.5Ghz one month later in December 2007 than 32 cores can be interesting again.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill