Conference decwet::windows-nt

    Right... it's an attribute of the SIS chipset.  The Venturis line
    is more aimed at the general business user, with the Celebris line
    more aimed at the power user, which is what this more closely aligns
    to.
    K

Just a stab in the dark, but what was your swapfile size 
for each memory configuration?

r
[Posted by WWW Notes gateway]

    The latest BIOS on WEB/BBS fixes this problem.
    Venturis 5xxx (not 560) BIOS V2.02 (EXE) 
    BG92202A.EXE

    Rod,
    
    The swap file was changed accordingly under different memory
    configuration.
    
    Sam,
    
    The BIOS has been upgraded to V2.02 already and no improvement.
    
    

I don't have a degradation problem: I just haven't seen an improvement
after upgrading a Venturis 4100 from 36MB to 68MB and a Celebris 5100
XL from 48MB to 192 or 196MB. Are there BIOS changes that I should
apply?

Pete

    
    As per my first message, there was no improvement after upgrading the
    Venturis 5100 from 40MB to 56MB, and there was even degradation when
    it was further upgraded to 72MB:
    
                Memory | Time taken by Venturis 5100
               =====================================
                  8MB  |         35-36 sec
                 16MB  |            -
                 24MB  |         12-13 sec
                 40MB  |         11-12 sec
                 56MB  |         11-12 sec
                 72MB  |         16-18 sec
    
    It seems that this might be due to the 64MB memory limit for 256KB L2
    cache.  Any other ideas/comments ?
    
                                    

    re: .6 by HGOV05::JOHNNYSZE
    
    Are you expecting the runtime to approach ZERO as your memory goes to
    INFINITY?  It just doesn't work that way!  How do you know that 11-12
    seconds is not the best that one can do on that machine?
    
    /Bill

    Maybe it's me, but if it feels good at 40Mb and 56Mb, and hurts
    at 70Mb, then I'd stay away from 70Mb.

    
    We have just tested the Compaq PC with more than 64MB memory & the result
    is as follows.  From this it is clear that performance is degrading in
    our Venturis PC. 
    
    System          Compaq Deskpro2000        Digital Venturis
    
    File Size used  11.6 MB                   11.6 MB
    L2 Cache        256 KB                    256 KB
    RAM Size used   40 MB     72 MB           40 MB           72 MB
    Time taken      3 Minute  2 Min.10 Sec.   3 Min. 22 Sec.  3 Min. 40 Sec.
    
    Anyone has further idea ?
    

    reread .1 again.
    reread .1 again.
    reread .1 again.
    re...

    If it's an attribute of the SIS chipset, does it mean that there is no
    workaround ??
    

    What is the SIS chip set?

    SIS  --- "Sorry It's Slow"
    
    
    	8-)

    re: Note 6028.11 by HGOV06::JOHNNYSZE
    
    It means you either don't go over 64MB of RAM, or you replace your 
    motherboard with one that's more 'robust'. 
    
    Ken....

    re.12
     >What is the SIS chipset
    A chipset from a company called SIS.  We sold 800,000 PC based on
    them last year.  BTW, what's Alpha? ;-)
    
    >Sorry It's Slow
    An SIS-based Celebris FX got the highest score (Winstone97 Business
    score of 45) in PC Magazine's last roundup (not the one just out;
    the February MMX roundup... not to be confused with the FX-2 which
    got a 45 too (but only had 256k, all production models are 512k)
    in the 6/27/97 roundup).
    
    The SIS-based PC is designed for business applications, where it does
    very well, not high end CAD. 
    K

    SiS stands for Silicon Integrated Systems (Corp.).
    
    FYI,
    
    /Bill

    Uh could I buy a clue here?
    
    What does the SIS "chipset" _do_?  Why is it "obviously slower" with
    a memory size larger than 64Mb?
    
    FJP

    re: .17 by USPS::FPRUSS
    
    > What does the SIS "chipset" _do_?  Why is it "obviously slower" with a
    > memory size larger than 64Mb?
    
    All of the "chipsets" people talk about when referring to PCs are the
    system support chip sets.  They connect the processor chip to the
    external cache (if it exists), the main memory, and provide some path
    to the I/O.  Current chip sets provide a primary PCI bus and a bridge
    to ISA or EISA.
    
    The 64MB cacheable limit in this particular product gives us the
    undesirable side effect that adding memory beyond 64MB can result in
    slower performance, because once the O.S. touches memory beyond the
    cacheable limit, access times are those of standard dynamic rams (50 to
    70 nanoseconds typically) rather than those of much faster static rams
    (10 to 15 nanoseconds typically) for most memory accesses.
    
    Hope this simplified explanation helps.
    
    /Bill

>    The 64MB cacheable limit in this particular product gives us the
>    undesirable side effect that adding memory beyond 64MB can result in
>    slower performance, because once the O.S. touches memory beyond the
>    cacheable limit, access times are those of standard dynamic rams (50 to
>    70 nanoseconds typically) rather than those of much faster static rams
>    (10 to 15 nanoseconds typically) for most memory accesses.

And to make matters worse, Robert Collins reports that memory is allocated from
the 'top' down (see http://www.x86.org/digest/May97/Feature01.html)

"I  didn't  think this  was  any big  deal,  as I figured  that the  benchmark
 programs would use less than 64 MB of memory, and wouldn't be affected by the
 upper 64 MB of  cacheability.  An  AMD  employee  pointed  out  that  Windows
 allocates memory from top-down.  That means that these benchmarks were run in
 the  non-cacheable  portion of memory (top of memory),  instead of the better
 performing  lower  portion of  memory.  Now,  I've installed  the tag-ram  (a
 whopping  $1.89)  and  re-run  the  tests.  As  you can see,  the results are
 strikingly different than they were before."