Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: $333.70 per elo point over my pc..

Author: Robert Hyatt

Date: 21:53:51 02/22/03

Go up one level in this thread


On February 22, 2003 at 20:06:44, Jeremiah Penery wrote:

>On February 22, 2003 at 17:31:35, Robert Hyatt wrote:
>
>>On February 22, 2003 at 01:49:45, Jeremiah Penery wrote:
>>
>>>If Xeon suddenly clocked up to 5GHz tomorrow, and all other chips stayed the
>>>same, it would certainly begin eating into that market segment.  Assuming
>>>something near linear scaling, a 5GHz P4 would have SPECint and SPECfp near
>>>1800.  That's _twice_ as fast in SPECint as any currently released "server"
>>>processor (POWER/Alpha/Itanium), and almost 25% faster in SPECfp than _any_
>>>processor.
>>
>>SPEC is not that important for servers.  Ditto for mainframes.  I/O is the
>
>If you don't like SPEC, look at TPC numbers.

Let's back up to reality a minute.  Servers have one type of requirement.
workstations have another.  I'm not about to talk about the PC platform as
a high-performance server. It is _not_.  The raw PCI bus is completely
unable to sustain the kind of bandwidth that the high-end server market is
all about, and that is a problem.  Doesn't matter how fast the CPU goes.

Just compare the I/O bandwidth of "server machines" with the I/O bandwidth
of a PC and you will see what I mean.




>
>In TPC-C, the machine currently with the best price/performance ratio is a
>2.4GHz P4 machine.  The machine with the highest total score is a 272 processor
>P3-900MHz Xeon cluster.
>
>In TPC-H, using the 3TB and 10TB database sizes, the machines with highest
>submitted results are both 128 processor 2.8GHz P4 Xeon clusters.
>
>And, finally, on TCP-W, x86 machines (not clusters) win in both the 10k (8p
>1.6GHz XeonMP) and 100k (16p 900MHz Xeon) item tests.
>
>>name of the game, and multiple channels/busses is the issue.  The PC is still
>>a PC, whether it is 500mhz or 50ghz.
>
>Of course they're not suited to some kinds of big-iron tasks, but there are
>several types of server workloads that they do handle very well.

Only if "very well" has a "price" component...  which was my point.  The
high-end machines are not 100K or 500K boxes...




This page took 0.01 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.