>> If you find ReCache doesn't work for you - even on a Windoze machine -
>> then I'm sorry, but you do have the option not to use it!
>
>This really sounds like it's a result of Intel's policy of making their
>chipsets as cheap as possible.
>It's a well known problem that several of the widely used Intel chipsets
>can't cache memory above 64MB.

True - but the example I cited uses the LX chipset (Intel AL440LX board),
which definitely caches more than 64 MB memory. Or, at least, the system
speed stayed the same when I upgraded from 64 MB to 96 MB.

My other home system - dual PII-350 on a Supermicro P6DBS, BX chipset,
128 MB memory - also shows a big response to ReCache. My work system -
Celeron 266, EX chipset on "unknown" M/B, only 32 MB memory - shows
little response. In fact ReCache's name comes from the fact that
originally I thought of clearing crud out of the L2 cache. I realize
now that this is probably incorrect, but the project name stuck 'coz
I couldn't be bothered to change it ...

>What isn't so well known is that this makes a real difference on Windows,
>since Windows uses memory from the high addresses first, so the first
>programs to be started ends up in uncacheable memory.
>This is why it's sometimes possible to see machine performance drop when
>you add more memory.

With Win 9x, yes. I believe NT tends to allocate from the bottom up.

>I suspect the real reason why ReCache makes prime95 faster is that it uses
>up all the non cacheable memory, then loads prime85 in chacheable.

Are you _sure_ this is the case - most of my development was done under NT
but I did check the effects were similar under 9x - also I thought that
deficient chipsets, like TX, failed to cache _any_ memory once you went
over the 64MB cacheable memory limit?
>
>This will also be the reason for the difference on Linux, since the VM
>model is completely different, and Linux specifically works around some of
>the stupidities of the chipsets..

But I've never seen a case of a system which will run mprime over linux
significantly (>1%) faster than it will run Prime95 over Win NT - except
for one pathological case where memory was insufficient to run _anything_
properly over NT - when the systems were otherwise idle.

This last comment was _not_ intended to start, or provide ammunition for
any existing, OS war - I'm just trying to make the point that Windoze is
not neccessarily a total disaster from the performance point of view -
I actually _prefer_ linux but find myself forced to keep some systems
running Windoze in order to maintain collaboration with fellow workers
who insist on using M*cr*s*ft applications.

Regards
Brian Beesley
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm

Reply via email to