Author: Matt Taylor
Date: 14:16:15 12/04/02
Go up one level in this thread
On December 04, 2002 at 14:28:23, Vincent Diepeveen wrote: >On December 04, 2002 at 13:55:47, Jeremiah Penery wrote: > >>On December 04, 2002 at 13:32:01, Vincent Diepeveen wrote: >> >>>On December 04, 2002 at 11:42:17, Matt Taylor wrote: >>> >>>>On December 04, 2002 at 10:43:59, Vincent Diepeveen wrote: >>>> >>>>>On December 04, 2002 at 10:21:08, James T. Walker wrote: >>>>> >>>>>>On December 04, 2002 at 08:00:35, martin fierz wrote: >>>>>> >>>>>>>hi, >>>>>>> >>>>>>>i'm on the lookout for a new PC for endgame database computations. i'll probably >>>>>>>be buying a lot of ram, 2-3GB. i see that there is a big price difference >>>>>>>between DDRAM and SDRAM. IIRC the main difference is that you get a larger >>>>>>>bandwidth, but about the same latency with DDR - so i suppose i'm better off >>>>>>>buying SDRAM for my application. any opinions of the experts? >>>>>>> >>>>>>>thanks in advance >>>>>>> martin >>>>>> >>>>>>For what it's worth: I purchased one stick (256M) of DDR ram to compare to my >>>>>>cheap SDRAM. I found no noticable difference in chess performance (just price). >>>>>> I did not do any extensive testing. I simply compared Fritz marks. I suspect >>>>>>that in the future most motherboards will not accept the SDRAM. >>>>>>Jim >>>>> >>>>>I see a big difference. 64 versus 32 bytes cache lines matters >>>>>a lot for DIEP and all software that doesn't fit within L1 cache. >>>>> >>>>>Best regards, >>>>>Vincent >>>> >>>>Cache line size is a part of the CPU, not the ram. There are a number of >>>>transitional products, both P4 and Athlon, that accept both SDRAM and DDR SDRAM. >>>>(However, I have never heard of anyone happy with these products.) >>> >>>the P4 ended up being a lot faster for DIEP when i tested a p4 with ddr ram >>>isntead of RDRAM. >>> >>>P4 with ddr ram (northwood) is like 1.5 : 1 for a K7 >>>used to be 1.7 : 1 to a k7 with rdram. >>> >>>So 1.7 Ghz P4 rdram == 1.0Ghz K7 for DIEP >>> 2.4 Ghz P4 ddr == 1.6Ghz K7 for DIEP (both ddr). >>> >>>DDR is a big step forward!! >>> >>>i don't know where the processor gets 64 bytes instead of 32 bytes in >>>the design. I just know it gets 64 bytes, versus SDRAM 32. >> >>Hash probes in a chess program depend on latency, not bandwidth. RDRAM sucks >>for latency, where it is at least double that of SD/DDRAM. >>For the tablebase creation, I would expect this to depend much more on streaming >>bandwidth - there, RDRAM will win. > >Not at all. Practical bandwidth from DDR ram is better. > >But besides from that in my own generator i can put an entire >6 men bitmap in RAM (one side to move win or loss in N moves to >conversion). > >That means that latency there is the only important thing. >Harddisk speed is simply irrelevant compared to RAM latency. > >You can get next buffer from positions sequential from harddisk >which is about with 160MB/s that's like 1 billion positions. > >getting the like 20 equivalent moves from that from RAM is the >major problem. Of course skipping already the ones that are >already known to be either winning or losing (depending upon >what pass you are at). > >All those lookups are in RAM therefore. harddisk speed is >irrelevant. Is anyway irrelevant knowing how much ram nowadays >systems have. > >A single 6 men with pawns buffer is like 18G entries. > >To give example from my own simple generator (i do not use the >extra optimizations nalimov uses, because it slows down the >indexing function too much as i want to generate all 6 men within a >year instead of many years; cpu speed is the limiting factor): > >159 kbnkbp 00240041 18141551280 > >So that's for kbn kbp in my generator 18.1G entries. > >So looping sequential through the positions in small buffers, >you can easily see that i can do random lookups to memory easily >as this is just a bitmap of: 2267693910 bytes. > >That's very luckily smaller than 3 GB because windows doesn't >allow more memory space than 3 GB if i am correct for my K7 computer >and x86 in general. > >So a U160 SCSI disk and a machine that's just generating can generate >all 6 men within a year. > >Of course i do not save the mate in 'n' positions. So when combining >the bitmaps i just keep left with the 'win/loss bitmaps and create >win/draw/loss from it instead of mate in n or loss in n, which nalimov >saves. > >Then i compress it all. > >Uncompressed i have : > >In total 510 egtbs using 2631231208124 entries > >That's for 1 side to move. For both sides to move, just double it. > >So say 5.22 Tera entry. > >that's uncompressed 5.22 / 5 positions a byte = 1.044 Terabyte to >store. > >That compressed will go to i suppose around 150GB. > >Perhaps i can get it smaller. Still toying with Andrew Kadatch superb >compression there. > >All 5 men fit at 1 CDrom. Theoretical bandwidth on pc2100 DDR is 2.1 GB/sec. Theoretical bandwidth on pc2700 DDR is 2.7 GB/sec. Theoretical bandwidth on pc800 RDRAM is 3.2 GB/sec. I used to have a P4 1.5 GHz with 768 MB pc800 RDRAM. The system had 2.8 GB/sec empirical bandwidth, which is higher than pc2700 DDR SDRAM theoretical bandwidth.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.