[fen]8/8/8/8/8/1K6/R7/2k5 - - w 1 0[/fen]
In this position these are the solving times for risc2500:
128kb 3min 20 sec
512kb 28 sec
2mb 24 sec
Not much difference between small upgrade and big upgrade.
Try KBN/k positions: mate in 7
[fen]8/8/6K1/8/8/3N4/8/5B1k - - w 1 0[/fen]
128kb 21 min 39 sec
512kb 2 min 26 sec
2mb 1 min 19 sec
mate in 8
[fen]8/8/6K1/8/8/3N4/4B3/7k - - w 1 0[/fen]
128kb One hour no mate
512kb 35 min 31 sec
2mb 6 min 30 sec
This next one I only tried it with the 2mb machine, I knew the 128kb
would not do it and the 512kb would take too long. mate in 9
[fen]8/8/6K1/8/8/3N4/8/3B3k - - w 1 0[/fen]
2mb 32 min 20 secs
Interestingly after I made my reply move it took fully 15 mins to make
it's next move...why?... was the ram exhausted and the processor was
left to it's lonesome?
You notice with the longer mates, the difference between small and
big upgrades magnifies itself. This would be because of the bigger "branching
factor" as they say over on CCC. This refers to the expanding tree of
variations, and the 128kb, then the 512kb rams are soon full and the
processor is left to its own resource, just as in a non hash machine.
It will be interesting to check solving times with ram turned off...I'm
assuming the risc has this facility.
Good to have three risc machines, one with each ram size. I'll do some
more testing later.
nice day all...
Larry
risc2500 solving speed with different ram values
Moderators: Harvey Williamson, Steve B, Watchman
Forum rules
This textbox is used to restore diagrams posted with the fen tag before the upgrade.
This textbox is used to restore diagrams posted with the fen tag before the upgrade.
-
Peter Grayson
- Member
- Posts: 208
- Joined: Sat Aug 04, 2007 3:23 pm
- Location: South Wales, UK
Re: risc2500 solving speed with different ram values
Hi Larry,
Must admit I mothballed my Mephisto Risc 1 Mb and London 030 dedicated machines over 10 years ago, although I have played a few games on the 030 since then, so most of my hash effect testing has involved PC's rather than dedicated machines.
I recall in the heydays of the dedicated machines that a Richard Lang comment in Eric Hallsworth's Selective Search magazine was something along the lines that hash was most useful for endgame positions rather than middle game positions that suggested significant speed gains would be achieved by more hash in the latter part of the game and your results seem to confirm that is the case with the Risc 2500. It would be interesting if similar solving time improvements were seen for middle game positions.
It still comes down to how the author has set out the code for hash use and some engines seemed to clear hash after a move was made and that seems to be the case in the last position when finding the M8 after a M9 announcement may have been expected very quickly. Was your reply the predicted move? If not that may also have some bearing on the solving time.
I think the Mephisto MMII engine continued calculating but dropped ply depth after a predicted move was made that gave some continuity to its analysis.
Your position tests confirm the worth of extra RAM for endgames but back then RAM was very expensive just as were the top dedicated machines and the costs were, certainly for me, a major influencing factor.
If you have opportunity to compare some middle game positions solving it would be interesting to see if the RAM size speed gains were mainly in the endgame.
Best regards,
PeterG
Must admit I mothballed my Mephisto Risc 1 Mb and London 030 dedicated machines over 10 years ago, although I have played a few games on the 030 since then, so most of my hash effect testing has involved PC's rather than dedicated machines.
I recall in the heydays of the dedicated machines that a Richard Lang comment in Eric Hallsworth's Selective Search magazine was something along the lines that hash was most useful for endgame positions rather than middle game positions that suggested significant speed gains would be achieved by more hash in the latter part of the game and your results seem to confirm that is the case with the Risc 2500. It would be interesting if similar solving time improvements were seen for middle game positions.
It still comes down to how the author has set out the code for hash use and some engines seemed to clear hash after a move was made and that seems to be the case in the last position when finding the M8 after a M9 announcement may have been expected very quickly. Was your reply the predicted move? If not that may also have some bearing on the solving time.
I think the Mephisto MMII engine continued calculating but dropped ply depth after a predicted move was made that gave some continuity to its analysis.
Your position tests confirm the worth of extra RAM for endgames but back then RAM was very expensive just as were the top dedicated machines and the costs were, certainly for me, a major influencing factor.
If you have opportunity to compare some middle game positions solving it would be interesting to see if the RAM size speed gains were mainly in the endgame.
Best regards,
PeterG
[fen]1K6/3bNB2/3k4/7R/8/2b1N3/3n1Q2/3Rr3 [/fen]
I looked in the back of "Blackburne's Chess Games" and found this
composed pawnless mate in 4. Suitable problems have to be searched
for, since some problems cannot be solved in any reasonable time, while
others are solved too quickly.
The unupgraded (128kb) machine surprised me with it's speed, being
not much slower than the others:
RISC2500
128kb 51secs
512kb 45secs
2mb 44secs
Once again the small and large upgrades are very similar. I'll be checking out
some more positions, and look for some suitable longer mates.
For the record, here are the moves, in English descriptive:
1. R - K5 BxR
2. Kt(K3) - B5ch BxKt
3. Q - R7 B-Q2
4. Q- R3 mate
By comparison, the Atlanta, normally fast, took 1min33secs.
all the best,
Larry
PS The FEN gives white playing down the board, for some reason.
I looked in the back of "Blackburne's Chess Games" and found this
composed pawnless mate in 4. Suitable problems have to be searched
for, since some problems cannot be solved in any reasonable time, while
others are solved too quickly.
The unupgraded (128kb) machine surprised me with it's speed, being
not much slower than the others:
RISC2500
128kb 51secs
512kb 45secs
2mb 44secs
Once again the small and large upgrades are very similar. I'll be checking out
some more positions, and look for some suitable longer mates.
For the record, here are the moves, in English descriptive:
1. R - K5 BxR
2. Kt(K3) - B5ch BxKt
3. Q - R7 B-Q2
4. Q- R3 mate
By comparison, the Atlanta, normally fast, took 1min33secs.
all the best,
Larry
PS The FEN gives white playing down the board, for some reason.
-
Peter Grayson
- Member
- Posts: 208
- Joined: Sat Aug 04, 2007 3:23 pm
- Location: South Wales, UK
Larry,Larry wrote:[fen]1K6/3bNB2/3k4/7R/8/2b1N3/3n1Q2/3Rr3 [/fen]
I looked in the back of "Blackburne's Chess Games" and found this
composed pawnless mate in 4. Suitable problems have to be searched
for, since some problems cannot be solved in any reasonable time, while
others are solved too quickly.
The unupgraded (128kb) machine surprised me with it's speed, being
not much slower than the others:
RISC2500
128kb 51secs
512kb 45secs
2mb 44secs
Once again the small and large upgrades are very similar. I'll be checking out
some more positions, and look for some suitable longer mates.
For the record, here are the moves, in English descriptive:
1. R - K5 BxR
2. Kt(K3) - B5ch BxKt
3. Q - R7 B-Q2
4. Q- R3 mate
By comparison, the Atlanta, normally fast, took 1min33secs.
all the best,
Larry
PS The FEN gives white playing down the board, for some reason.
Thanks for posting this. It does tend to suggest on these dedicated machines that hash was more useful in the endings.
Out of interest, I blew the cobwebs off the London 030. On power up it says it has 768 kB of memory. It found the mate in 4 in a few seconds, perhaps quicker because I could not scroll to the clock quicker than that so the time here is my assessment.
In the earlier positions, it took
Pos 1
In this position these are the solving times for risc2500:
128kb 3min 20 sec
512kb 28 sec
2mb 24 sec
Not much difference between small upgrade and big upgrade.
L030 Kc4 M7 96s
-------------------------------
Pos2
Try KBN/k positions: mate in 7
128kb 21 min 39 sec
512kb 2 min 26 sec
2mb 1 min 19 sec
L030 Kg5 M7 121s
--------------------------------
mate in 8
Pos 3
128kb One hour no mate
512kb 35 min 31 sec
2mb 6 min 30 sec
L030 Kh5 M8 163s
--------------------------------
Pos 4
L030 I didn't have the patience to see it through!
So the 2Mb R2500 times look good.
Regards,
PeterG[/b]
Out of curiosity I decided to test Hiarcs 13.1 on my Acer Neotouch, it comes up with an alternative 1st move 1.Ng6 which is also mate in 4.Larry wrote:[fen]1K6/3bNB2/3k4/7R/8/2b1N3/3n1Q2/3Rr3 [/fen]
I looked in the back of "Blackburne's Chess Games" and found this
composed pawnless mate in 4. Suitable problems have to be searched
for, since some problems cannot be solved in any reasonable time, while
others are solved too quickly.
The unupgraded (128kb) machine surprised me with it's speed, being
not much slower than the others:
RISC2500
128kb 51secs
512kb 45secs
2mb 44secs
Once again the small and large upgrades are very similar. I'll be checking out
some more positions, and look for some suitable longer mates.
For the record, here are the moves, in English descriptive:
1. R - K5 BxR
2. Kt(K3) - B5ch BxKt
3. Q - R7 B-Q2
4. Q- R3 mate
By comparison, the Atlanta, normally fast, took 1min33secs.
all the best,
Larry
PS The FEN gives white playing down the board, for some reason.
Peter, I'll pull out the London Pro which is basically an
underclocked 030 (I think), and I'll check it's solving times.
Note if you look at : http://www.schachcomputer.info/BT2630-1280.htm
the risc2500 2mb actually beats out a couple of the TASC machines in
position solving.
Chris, you've got my curiosity up now. I'll get out the pocket PC and
do a bit of checking there too. Of course on pocket pc's you can adjust
the ram size...great when you are on a boring train ride
underclocked 030 (I think), and I'll check it's solving times.
Note if you look at : http://www.schachcomputer.info/BT2630-1280.htm
the risc2500 2mb actually beats out a couple of the TASC machines in
position solving.
Chris, you've got my curiosity up now. I'll get out the pocket PC and
do a bit of checking there too. Of course on pocket pc's you can adjust
the ram size...great when you are on a boring train ride
For comparison with the 68030 London, I set up the KBN/k mate in 7
position on the Berlin London Pro 68020
--------------------------------------
Pos2
Try KBN/k positions: mate in 7
128kb 21 min 39 sec
512kb 2 min 26 sec
2mb 1 min 19 sec
L030 Kg5 M7 121s
Berlin London pro Kg5 M7 262 sec
---------------------------------------
This is pretty much to be expected from the lesser 68020 processor.
Other solving times would be proportionally longer also...
jealous regards,
Larry
position on the Berlin London Pro 68020
--------------------------------------
Pos2
Try KBN/k positions: mate in 7
128kb 21 min 39 sec
512kb 2 min 26 sec
2mb 1 min 19 sec
L030 Kg5 M7 121s
Berlin London pro Kg5 M7 262 sec
---------------------------------------
This is pretty much to be expected from the lesser 68020 processor.
Other solving times would be proportionally longer also...
jealous regards,
Larry
-
Peter Grayson
- Member
- Posts: 208
- Joined: Sat Aug 04, 2007 3:23 pm
- Location: South Wales, UK
Interesting comparison. As I recall I only decided to get the 030 rather than the 020 at the last minute. Previously I had the Portorose 68000 and then the Vancouver 68020. I thought the originally purchased Genius2 030 was significantly stronger than the Vancouver and although on the same hardware, the London seemed that much stronger again although machine vs machine results did not show such a big gap for the latter upgrade. Possibly a better engine against humans!Larry wrote:For comparison with the 68030 London, I set up the KBN/k mate in 7
position on the Berlin London Pro 68020
--------------------------------------
Pos2
Try KBN/k positions: mate in 7
128kb 21 min 39 sec
512kb 2 min 26 sec
2mb 1 min 19 sec
L030 Kg5 M7 121s
Berlin London pro Kg5 M7 262 sec
---------------------------------------
This is pretty much to be expected from the lesser 68020 processor.
Other solving times would be proportionally longer also...
jealous regards,
Larry
The earlier 68020 machines such as in the Vancouver I had ran at 12 MHz but there were some later ones running at 20 MHz. These may have been experimental OC'd. Not sure on their history but they are recorded in the Selective Search Magazine dedicated machines rating list for the Lyon and Vancouver at 2239 ELO. The Berlin Pro is shown as 2236 ELO so not much difference. London 030 shown as 2302 ELO
Smiling and not for sale regards,
PeterG