Rybka Chess Community Forum
How do you think this would compare to buying about 25 quad core computers and connecting them in a cluster?
I think the magical differential is shared memory. When all the cores share memory is totally different than when they don't. I don't really know what the performance advantage is but it must be substantial.
I agree that shared memory results in a substantial performance boost compared to a cluster. The question I was trying to figure out is: How many shared memory cores would you need to equal the performance of a 100 core cluster? I would guess that 48 shared-memory cores would be better (for infinite analysis anyway, not for IDeA).
>I think the magical differential is shared memory. When all the cores share memory is totally different than when they don't. I don't really know what the performance advantage is but it must be substantial.
Not sure. It could be that the 48 cores on shared memory are overloading the memory controller therein. How much of the 64GB of RAM can engines handle?
It's a floating distinction. All multisocket Opteron systems are NUMA, in that every RAM chip “belongs” to a given CPU socket. If CPU0 wants some memory that belongs to CPU1, it has to ask CPU1 for it.
It's still much faster than going over the network, of course, but it's definitely something you have to take into account for a high-performance application when you get enough cores.
/* Steinar */
I buy this but rybka 3 use 4 cores only. :o Maybe rybka for cluster works with all cores?!
? Rybka can use many more cores. Why do you think it's limited to 4?
In this thing rybka 3 use 4 cores only. Maybe this thing is cluster.
In case it's a cluster setup it would 12 cores (=one processor)
This is no cluster - so Rybka can use all 48 cores. But it only runs at 2.2 GHz.
i noticed the other day you can rent 24 cores @ 2.4 ghz for 1K for a month, if anyone is interested in experimenting with something like that. www.softlayer.com.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill