> > 1) All systems using SDRAM & the Intel 820 chipset seem to be very
> > disappointing. The 820 chipset was designed to use RDRAM and the
> > required "translation" in the memory bus slows things down somewhat.
The new Intel i815 chipset uses PC133 SDRAM natively, for coppermine support
up to 1GHz or whatever. They benchmark similiarly to a 440BX based
motherboard when both run at the same clock speed (i.e. both at 100Mhz, or
both at 133MHz with the BX overclocked).
> This cheaper version of this Asus CUV4X MB uses the VIA 82C chips : I am
> now very happy not having paid the extra 100 $ to buy the Asus Intel's
> version, which as you say may have given the same result !
Try the new Asus CUSL2. From all reports, this is a winner. Said i815
chipset, uses PC133 SDRAM. 6 PCI slots, no ISA slots. has AGP Pro 4X (the
kind that actually works correctly in 4X mode with advanced graphics cards
like the Geforce family), unlike the VIA stuff). The chipset has a lame
built in Intel graphics controller, this is probably adequate for 17"
monitor business systems, but is easily disabled by simply installing a AGP
card for higher resolutions or gaming.
> > 2) A PIII-650 has 100 MHz memory bus speed, just like a PIII-500.
> > Since Prime95 drives the memory bus quite hard, perhaps one shouldn't
> > expect too much speed increase. Especially when you have the 820
> > chipset MTH bottleneck in the circuit.
>
> I understand. I didn't think at Prime95 so much using the bus (in fact, in
> Paris the Metro is far more convenient). I was also not aware that a new
MB
> could make a bottleneck to a not-so-recent processor... I do appreciate
> Asus for years because they are very reliable (imho), but now i'm kind of
> deceived.
Prime95 has a working set of about 5MB now when doing lucas lehmer tests on
current range numbers. These CPUs only have 256k cache. about 5 or 10 times
a second, the CPU is walking through the 4MB FFT arrays a NUMBER of times,
including both reads *and* writes in rather complex patterns. Its my
understanding from what George has told us repeatedly that the LL tests
positively HAMMER the memory bus to keep the FPU pipeline busy.
>
> > 3) The smaller cache in the PIII-650E relative to the "old" PIII-650
> > (which ran at 2.0V with 512K L2 cache at 0.5x core speed) will result
> > in increased loading on the memory bus, which may more than offset
> > the clocks saved by the improved cache design.
>
> Ho ho, another tricky point here. It seems to me this 256K PIII wind up
> beeing another kind of "light" version of a serious cpu. Decreasing cache
> size may be just like gluing a tape on the border of a large 21" screen to
> make a 15" of it ?
>
> George also talked about sdram speed, it may be the point too.
>
> > Commercial benchmarks exercise the graphics & disk subsystems as well
> > as the memory subsystem (in fact a lot will run in the L2 cache); the
> > extra raw speed will definitely help boost these benchmark figures,
> > even if Prime95 speed remains depressed.
>
> OK i'll check if I can find some serious testing around the
> memory/cache/bus/sdram ... in fact, we got it, it's called Prime95, thanks
> to George. By the way, why not making a special version of Prime (with no
> net connection), that does only the testing of a number of iterations on
> some different numbers (defined in the ini file), and either give this
> program as a "comparative mesurement" of machines, and/or have a site that
> takes the users measures and publish "average" result depending of
> materials ?
>
> So we could know if a solution is good before we buy...
>
> > I've switched to Athlon since the Athlon memory bus design is
> > superior even if the cache is slower. My Athlon 650 is turning in
> > Prime95 performance equivalent to George's PII-400 scaled up to 750
> > MHz.
>
> Congratulations Brian. But in fact, even if I run my 5 machines 24h, they
> don't serve me for Prime95 only (i'm poor). This PIII 650 seems to be
quite
> good for 3D calculations, PAO and graphics (along with a good agp card).
> So, i'm not so complaining. Even if understand in a general meaning it may
> be closer say to a 555 then a 650...
>
> Sylvain Perez
>
> PS: my fault, I forgot to put an objet to the initial email, this gives
the
> present awfull title of this thread.
>
>
> _________________________________________________________________________
> Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
> Mersenne Prime FAQ -- http://www.exu.ilstu.edu/mersenne/faq-mers.txt
_________________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.exu.ilstu.edu/mersenne/faq-mers.txt
