Author: Don Dailey
Date: 21:25:23 02/06/99
Go up one level in this thread
On February 06, 1999 at 23:13:25, Robert Hyatt wrote: >On February 06, 1999 at 14:04:32, Don Dailey wrote: > >>On February 05, 1999 at 17:53:57, Eugene Nalimov wrote: >> >>>On February 05, 1999 at 15:50:20, Dann Corbit wrote: >>> >>>>On February 05, 1999 at 14:52:07, Don Dailey wrote: >>>>[snip] >>>>>There are some companies selling these machines at reasonable >>>>>prices, better than you might think. They can run Windows NT and >>>>>Linux. Crafty would really scream on this machine. I don't know >>>>>if Bob has an Alpha NT port or not, but there is a good chance a >>>>>recompile of Crafty will do the trick. >>>>There is an ALPHA NT version at my website. It is a single CPU compile, but it >>>>is just a #define away from being an SMP version. If anybody wants me to build >>>>an SMP version for the Alpha chip, let me know. We have an Alpha machine being >>>>used in C.A.P. right now. If I did a port for our big ALPHA unix machine, we >>>>might really see some performance. The Alpha NT machine we have is a pipsqueak >>>>(old and not a very high MHz chip). >>>> >>>>>The big problem is how many >>>>>programs will run on an NT Alpha machine? This I don't really know. >>>>>But theoretically, it should be easy to port most NT software. >>>>It they are in C, it should not take much. Especially if they are simply >>>>Winboard engines. All you have to do is recompile it. It is almost always the >>>>GUI stuff that is system specific. And if they already work on Windows 95 or >>>>Windows NT, the port is trivial. >>>> >>>>>Programs like Fritz, which are developed with assemblers won't >>>>>port without a huge effort so don't expect to see them. It's >>>>>the classic tradeoff, if you want the most compatibility and >>>>>comfort, you have to accept more performance constraints. You >>>>>also have to face your fears, superior products usually die >>>>>eventually because the lesser (usually lesser because it is OLDER) >>>>>product is the one that has the most intertia and the most hype. >>>>I would *really* love to try an EV6 machine with 8 cpu's running NT to see what >>>>it could do. While the Alpha machine has native 64 bit integers for the >>>>compiler, the OS is still 32 bit. I think a 64 bit port of NT will also be very >>>>helpful. Tablebase and opening book I/O would be improved, for instance. >>> >>>Here I have Dell dual PII/400 and Digital dual Alpha 21164A/500. >>>Based on my experience, those machines have almost identical >>>performance - even on Crafty, which does a lot of 64-bit operations, >>>Alpha is only marginally faster. My feelings are confirmed by >>>SpecInt95. >>> >>>I know that 21164A is slower than 21264, but Pentium/400 is not >>>the last processor from Intel, too. >>> >>>The real Alpha advantage lays in 64-bit pointers - it's ideal for >>>huge databases. But not for chess, and not with 32-bit NT. >>> >>>Eugene >> >>I don't get this at all. My program runs way faster on an Alpha, >>even adjusted for megahertz. You cannot be doing this test right. >> >>Are you comparing native code applications? >> >>- Don > > >Here's my test results. For Paris, running on NT on a pentium pro 200mhz >machine, we did 80K for our 6 minute benchmark run. Jason fiddled around >for a few days compiling for an alpha, and on the 500mhz 21164 we used in >paris, the same test produces 250K nodes per second, a factor of 3.1 times >faster, for a clock speed of 2.5X faster. Which is actually pretty close, >mhz for mhz, with intel. mhz for mhz may not be a fair comparison to the alpha's. It probably takes a bigger hit (per mhz) than it would if it were running at 200mhz, with bus/memory/cache issues etc. Your program is extremely pentium friendly compared to mine. I did some tests a while back and I noticed that Crafty and Cilkchess have similar nodes per second on the alpha's. But on the pentiums Crafty is way faster. I haven't had any time to figure out why this is so but I know you have worked hard at making it run well on a pentium. The difference isn't some small factor either it was enormous. > But the 21264 is about 2x faster than the 21164 at >the same clock speed, based on numbers I have received... (running Crafty) >which means it blows the PII's away badly, and even the upcoming 500mhz PIII's) >although I wouldn't mind having 4 slot-2 PIII/500's when they are shipped. I >am saving money for them for my quad, for sure. :) I am waiting for the new alpha's! - Don
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.