Hello all,
Something I thought I'd share with you. I've been mucking around with my
system trying to squeeze some more performance out of ICR2 lately. The
system is a P133 Intel Endeavor PCI, 256k async cache, 32M 60ns EDO RAM,
Diamond Stealth 64.
I'd been using PMINFO32.EXE (from the NASCAR standard install) to
"benchmark" performance, and scratching my head as to why it was
indicating 7 wait states & 51Mb/sec for memory access. I'd already tried
overclocking (thanks to SRN for the info) to no avail.
So one of the things I did was pull out the memory cache SIMM. I nearly
fell off my chair at the results! PMINFO32 now shows 0 wait states and 92
MB/second memory access. Back in ICR2, I can now run a full field with no
loss of frame rate - previously it would bog down badly showing more than
10 other cars.
I guess it shows that ICR2 accesses memory in such a way as to completely
defeat the memory cache, and that I've been somewhat naive to believe that
memory cache would add major value to anything other than programs with
modest memory requirements. Other benchmarks (eg NXBench in NeXTSTEP)
indicate that I've taken a small performance hit in some areas, and
increased performance in others. I think I'll be leaving it out from now
on - the only reason I got the P133 was to get better performance in ICR2!
So, any thoughts, similar experiences, etc? I certainly hope this
information will be of use to others.
James
--
James Bevan
NeXTMail & MIME welcome
Removing your L2 cache should give you a _significant_ drop in
performance, in fact reports elsewhere show that increasing from 256k to
512k of cache give a noticable performance boost in ICR2.
The only thing I can think of is that perhaps you are the unwitting
victim of a "fake cache" chips scam. Some motherboards were sold with
"cache chips" which actually didn't work - this was, and is, a common
scam. If you have any way to try replacement cache chips, or upgrade to
a motherboard with pipe-cache where you can plug modules in and out, I
would be interested in your results.
How did you think to REMOVE your cache chips?! I would never have done
that, and I'm a hardware nut!
Cheers!
John
_________________________________
__ _____| |_____ __
________| |__| :| John Wallace | |__| |________
\ :| |::| :| Team WW Racing TSW | |::| | /
> :| |::| :|_________________________________| |::| | < / :|__|::|____/ * Sim Racing News * \____|::|__| \
/______:/ \::/ http://sneezy.dcn.ed.ac.uk/simnews/index.htm \::/ \._____\
http://www.math.ohio-state.edu/~harmon/simnews
>Removing your L2 cache should give you a _significant_ drop in
>performance, in fact reports elsewhere show that increasing from 256k to
>512k of cache give a noticable performance boost in ICR2.
I agree, it should! Having said that, the increased frame rate is
definitely there. I would guess that ICR2 accesses hugely disparate parts
of memory, causing the L2 cache to fault often enough to actually decrease
performance. If this is the case, then your comment about using 512k of
cache makes sense - perhaps 512k cache is enough to service ICR2's huge
memory accesses.
Like I said in my original post, I have taken a small performance hit in
apps with modest memory accesses, which I guess I expected.
Thanks for the suggestion. I did consider that, and tried out the
fake-cache detector from one of PCW's cover CDs. It indicated that the
cache was there and working OK. Maybe this shows up a difference between
pipeline burst & async cache (not that I'm familiar with the internal
differences between them)?
I found I was assuming that L2 cache cures all ills, so in the spirit of
experimentation I tried challenging my assumption. If you're going to
experiment, you might as well try everything. Didn't see a performance
increase when I removed the processor, though ;-)
/
> \ :| |::| :| Team WW Racing TSW | |::| | /
> > :| |::| :|_________________________________| |::| | <
> / :|__|::|____/ * Sim Racing News * \____|::|__| \
>/______:/ \::/ http://sneezy.dcn.ed.ac.uk/simnews/index.htm \::/ \._____\
> http://www.math.ohio-state.edu/~harmon/simnews
All the best
James
--
James Bevan
NeXTMail & MIME welcome
Quoted text (7 lines)
>Cheers! >John > _________________________________ > __ _____| |_____ __ >________| |__| :| John Wallace | |__| |________j
James,
I vote the last two sentences as the funniest comment I have read all
year, GOOD WORK.!!!!
Anthony
I bought a system in back in december with what was called 'Write-back
cache' now I've heard stories about it that it could be a faked chip
that does nothing (and surely no caching!) So I wonder what is the
best way to determine if it's doing its job or not... If not then I'll
send my system back and ask to get REAL cache!!
thanks in advance!
btw: When will sim racing news be released? I cant wait! :)
Benchmark your system, then go into the BIOS and disable the cache RAM.
Benchmark the system again. If there is no difference, you've been
duped. If the speed falls, your cache (and cash!) is safe. Remember only
to turn off the EXTERNAL cache. If you turn off internal cache (inside
the CPU) your benchmark will surely fall.
If the internal cache is missing, Intel have a LOT of explaining to do!
Em....soon!? It's basically done and sitting here on my hard disk but
I'm waiting on one or two bits and pieces. Should be tomorrow.
Cheers!
John
_________________________________
__ _____| |_____ __
________| |__| :| John Wallace | |__| |________
\ :| |::| :| Team WW Racing TSW | |::| | /
> :| |::| :|_________________________________| |::| | < / :|__|::|____/ * Sim Racing News * \____|::|__| \
/______:/ \::/ http://sneezy.dcn.ed.ac.uk/simnews/index.htm \::/ \._____\
http://www.math.ohio-state.edu/~harmon/simnews
