rec.autos.simulators

NASCAR: Smooth SVGA!

Jubal Ragsda

NASCAR: Smooth SVGA!

by Jubal Ragsda » Fri, 24 Feb 1995 17:59:36



>>   Is the memory access speed synchronized with the clock speed of the
board?
>>What would the optimum access time of the RAM be versus the clock speed?

>>Hope this question makes sense.

>>Dave.

>Yes Dave, it does make sense, it goes something like this:

>Memory speed is calibrated in nanoseconds. (Millionths of a second)
>You may have seen things like "1Mb x 9 60ns SIMM". This describes
>the time it takes the memory chip to 'refresh' between data
>requests. So if the processor requests a byte, it will take 60ns
>before the chip is ready to provide the next one.

>Consider a motherboard running at 33MHz, fitted with 60ns SIMMS.
>This means the processor gets one clock tick (ie. executes one
>instruction in the case of a 486DX) every 0.0000000303 seconds. A
>ram chip on the SIMM would take 0.000006 seconds to recover between
>data requests. The numbers may look small, but check out the
>number of zeroes. The processor in theory could wait for around
>1000 ticks between bytes - just picking it's nose ! :-)

>That's where cache memory and something called 'memory
>interleaving' comes in. Cache is a very high-speed block of memory
>placed in between the processor and the main memory. It gets on
>with the job of retrieving data from main memory whilst the
>processor is doing something else. (Some instructions can take
>much longer than one 'tick' to execute.)

>Memory interleaving is a neat little trick employed by
>motherboards. It stores byte 1 in chip 1, byte 2 in chip 2, byte 3
>in chip 3 .... and so on. The advantage of that is, if the
>processor asks for several bytes in succession (as they usually do)
>each chip gives up it's byte in turn - no refresh time to wait
>for. Hopefully, by the time you've come round full circle, and need
>another byte from chip 1, it will have refreshed.

>Hope I'm not boring you guys - just love answering these questions.

>See ya..

I guess that's why the EDRAM based system performs so well with programs
like NASCAR. 16megs of 15ns cache can transfer information at 4 times the
speed of 60ns ram. Right?
David Kadlca

NASCAR: Smooth SVGA!

by David Kadlca » Sat, 25 Feb 1995 08:25:53

blah blah blah blah blah <didn't wanna put the whole quote in here:)>

    Thanks for response. Now, I know there is at least two major types of
memory chips: dynamic and static. I heard that static ram is WAAYYY
faster than dynamic ram. Is that true or not?

David Kadlcak
-------------
Aerospace Engineering Programm

#############################################################################
##                                                                         ##
##                        E m b r y - R i d d l e                          ##
##                                                                         ##
##           A e r o n a u t i c a l   U n i v e r s i t y                 ##
##                                                                         ##
##                                                                         ##
##   To Live Is To Die,       *****************************                ##
##   To Die Happy,            *  Michael Andretti in '95  *                ##
##   Is To Drink Beer.        *****************************                ##
##                                                                         ##
## =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= ##
##                                                                         ##
##       Pilnser Urquell  <== Drink it 24-7 ==>  Gambrinus                 ##
##                                                                         ##
#############################################################################

Jim Di

NASCAR: Smooth SVGA!

by Jim Di » Sun, 26 Feb 1995 04:16:30


    >> Is the memory access speed synchronized with the clock speed of
    >> the board?  What would the optimum access time of the RAM be
    >> versus the clock speed?
    >>
    >> Hope this question makes sense.
    >>
    >> Dave.
    >>
    >>
    Mike> Yes Dave, it does make sense, it goes something like this:

    Mike> Consider a motherboard running at 33MHz, fitted with 60ns
    Mike> SIMMS.  This means the processor gets one clock tick
    Mike> (ie. executes one instruction in the case of a 486DX) every
    Mike> 0.0000000303 seconds. A ram chip on the SIMM would take
    Mike> 0.000006 seconds to recover between data requests. The
        ^^^^^^^^^^^^
Of course this is 6 micro seconds which is 100 times bigger than the
60ns mentioned above.  60 ns is actually just 2 times larger than the
processor clock period.  And although instruction pipelining may play
some role, rarely does an instruction on a CISC achitecture chip
execute in 1 cycle.  So the problem is not as extreme as suggested here.

    Mike> numbers may look small, but check out the number of
    Mike> zeroes. The processor in theory could wait for around 1000
    Mike> ticks between bytes - just picking it's nose ! :-)

[cache and interleaving descriptions deleted]

    Mike> Hope I'm not boring you guys - just love answering these
    Mike> questions.

    Mike> See ya..

    Mike> =================================================================
    Mike> _/_/_/ _/ _/ _/ _/_/_/ : Mike Fisher - Belmin Purchasing
    Mike> _/_/_/ _/ _/_/ _/_/ : Systems Ltd. UK.  _/ _/ _/ _/ _/
    Mike> _/_/_/ :
    Mike> =================================================================

--
+---------------------------------------------------------------+
|Jim Dibb               | My  car doesn't care if your          |
|Sr. Software Engineer  | car climbed MT. WASHINGTON            |
|CLARiiON               |                                       |
|Data General Corp.     |                                       |

+---------------------------------------------------------------+

David Kadlca

NASCAR: Smooth SVGA!

by David Kadlca » Sun, 26 Feb 1995 06:45:55


> I guess that's why the EDRAM based system performs so well with programs
> like NASCAR. 16megs of 15ns cache can transfer information at 4 times the
> speed of 60ns ram. Right?

   How much did you pay for your cache memory? 2 million dollars?

David Kadlcak
-------------
Aerospace Engineering Programm

#############################################################################
##                                                                         ##
##                        E m b r y - R i d d l e                          ##
##                                                                         ##
##           A e r o n a u t i c a l   U n i v e r s i t y                 ##
##                                                                         ##
##                                                                         ##
##   To Live Is To Die,       *****************************                ##
##   To Die Happy,            *  Michael Andretti in '95  *                ##
##   Is To Drink Beer.        *****************************                ##
##                                                                         ##
## =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= ##
##                                                                         ##
##       Pilsner Urquell  <== Drink it 24-7 ==>  Gambrinus                 ##
##                                                                         ##
#############################################################################

Mike Fish

NASCAR: Smooth SVGA!

by Mike Fish » Mon, 27 Feb 1995 21:38:21


[snip]

Thanks Andy - my maths never were very good. Right in principal
though ?

=================================================================
    _/_/_/ _/ _/ _/ _/_/_/ :  Mike Fisher - Belmin Purchasing
  _/_/_/  _/ _/_/  _/_/    :                Systems Ltd. UK.
 _/  _/  _/ _/ _/ _/_/_/   :
=================================================================

Jubal Ragsda

NASCAR: Smooth SVGA!

by Jubal Ragsda » Tue, 28 Feb 1995 00:47:56




>> >> like NASCAR. 16megs of 15ns cache can transfer information at 4 times
the
>> >> speed of 60ns ram. Right?

>> >   How much did you pay for your cache memory? 2 million dollars?

>> No more expensive than traditional ram .8meg is about $350.

>  Can you use more than 512k of Cache or did you by SRAM SIMMs
somewhere?
>IOW, did you put cache mem where regular RAM is supposed be?

This board has special slots for the 15ns EDRAM. I have 16meg of cache and I
believe the board will handle upto 32meg of 15ns cache.
Joe G. Thomps

NASCAR: Smooth SVGA!

by Joe G. Thomps » Wed, 01 Mar 1995 09:31:24

[DRAM timing discussion deleted]

I would disagree.  The problem is much worse than suggested.

In both RISC and CISC machines, the latency of any given instruction may be
5 or more clock cycles.  A pipelined processor (all modern processors)
masks this latency by executing an instruction in stages.  By overlapping
instructions in the pipeline, a total throughput of close to 1 instruction
per clock cycle can be attained (superscalar processors with multiple
pipelines, like the Pentium, can execute > 1 instruction/cc).

Data dependencies, branches, and some specific instructions can stall the
pipeline and will reduce throughput to less than 1 instruction/cc.
However, with well-written code, a figure close to 1 (and certainly better
than 0.5) can be maintained.

Keep in mind that even though the external interface to most new Intel
processors is 30-33Mhz, the processor executes instructions at full speed
out of internal cache.  As an example, if both the internal and external
caches of a DX4/100 miss, a system with standard 70ns DRAM will cause the
processor to wait for 70ns/10ns = 7 clock cycles (14% efficiency!).  While
certain DRAM access modes can help to somewhat alleviate this for contiguous
memory accesses, the problem is very serious.

As another example, while a 98% cache hit rate sounds pretty darn good...
For a 100MHz processor with 70ns DRAM:
   A 98% hit rate yields 89% efficiency
   A 95% hit rate yields 77% efficiency
   A 90% hit rate yields 63% efficiency
   A 50% hit rate yields 25% efficiency!

The efficiency falls off hyperbolically with cache hit rate.  Small
improvements to cache hit rate can dramatically increase system performance.
Please understand that I am *not* implying that it is easy to increase hit
rate beyond simply providing larger caches.

My point is that memory hierarchy performance is *absolutely critical* to
the overall performance of the computer.  It is probably the single biggest
bottleneck to the performance of almost all current computers and I doubt
that the value of a good cache design can be overstated.

Joe Thompson

Roman Tschob

NASCAR: Smooth SVGA!

by Roman Tschob » Thu, 02 Mar 1995 04:27:00

At current cache prices of around $30 for 128K, that makes it about $240 per
Meg.  16 Megs should have cost him close to $4,000!  Wow it can be done, but
with more money than most of us will ever see, or pay for a complete system.
If he had it and spent $4,000 on memory, good for him and enjoy.

Roman

Anssi Saa

NASCAR: Smooth SVGA!

by Anssi Saa » Fri, 03 Mar 1995 01:07:55


:>> >> like NASCAR. 16megs of 15ns cache can transfer information at 4 times the
:>> >> speed of 60ns ram. Right?
:>> >
:>> >
:>> >   How much did you pay for your cache memory? 2 million dollars?
:>>
:>> No more expensive than traditional ram. *meg is about $350.
:>>
:>>

:>  Can you use more than 512k of Cache or did you by SRAM SIMMs somewhere?
:>IOW, did you put cache mem where regular RAM is supposed be?

This is EDRAM they are talking about. More info in appropriate groups,
I'm sure.

Anssi

Scott Bradley Cord

NASCAR: Smooth SVGA!

by Scott Bradley Cord » Fri, 03 Mar 1995 10:05:02


>At current cache prices of around $30 for 128K, that makes it about $240 per
>Meg.  16 Megs should have cost him close to $4,000!  Wow it can be done, but
>with more money than most of us will ever see, or pay for a complete system.
>If he had it and spent $4,000 on memory, good for him and enjoy.
>Roman

Hey!  Smeghead!  Try thinking before you jump down someones throat so
publicly.  Ever considered the possibility of setting up standard memory
(RAM) as cache with a program as common as SmartDrive (Ug, I always hoped
that I'd never mention this program).
        Cheers,
                Scott Cordon.
--
Cheers,
        Scott Cordon


Omar F. La

NASCAR: Smooth SVGA!

by Omar F. La » Fri, 03 Mar 1995 15:57:47



>Subject: Re: NASCAR: Smooth SVGA!
>Date: 2 Mar 1995 01:05:02 GMT

>>At current cache prices of around $30 for 128K, that makes it about $240 per
>>Meg.  16 Megs should have cost him close to $4,000!  Wow it can be done, but
>>with more money than most of us will ever see, or pay for a complete system.
>>If he had it and spent $4,000 on memory, good for him and enjoy.
>>Roman
>Hey!  Smeghead!  Try thinking before you jump down someones throat so
>publicly.  Ever considered the possibility of setting up standard memory
>(RAM) as cache with a program as common as SmartDrive (Ug, I always hoped
>that I'd never mention this program).
>        Cheers,
>                Scott Cordon.
>--
>Cheers,
>        Scott Cordon



------------------------------
Scott, seems like you're jumping into the thread without
knowing the real topic.  We're talking about 15ns SRAM (yes,
16 megs worth!), not about caching a harddrive.  : )
Seems like there are awesome new developments in memory technology
out there of which we aren't all aware...
Craig Parso

NASCAR: Smooth SVGA!

by Craig Parso » Fri, 03 Mar 1995 21:58:55

Talk about the pot calling the kettle black.

Craig

Jubal Ragsda

NASCAR: Smooth SVGA!

by Jubal Ragsda » Sat, 04 Mar 1995 01:56:09







>>Subject: Re: NASCAR: Smooth SVGA!
>>Date: 2 Mar 1995 01:05:02 GMT


>>>At current cache prices of around $30 for 128K, that makes it about $240
per
>>>Meg.  16 Megs should have cost him close to $4,000!  Wow it can be done,
but
>>>with more money than most of us will ever see, or pay for a complete
system.
>>>If he had it and spent $4,000 on memory, good for him and enjoy.

>>>Roman

>>Hey!  Smeghead!  Try thinking before you jump down someones throat so
>>publicly.  Ever considered the possibility of setting up standard memory
>>(RAM) as cache with a program as common as SmartDrive (Ug, I always
hoped
>>that I'd never mention this program).
>>        Cheers,
>>                Scott Cordon.

>Scott, seems like you're jumping into the thread without
>knowing the real topic.  We're talking about 15ns SRAM (yes,
>16 megs worth!), not about caching a harddrive.  : )
>Seems like there are awesome new developments in memory technology
>out there of which we aren't all aware...
>8 ,egs of 15ns cache/ram is $320-350. I paid $700 for 16 meg of 15ns. I belive

the price may have come down even a little more in the last couple of weeks.
David Kadlca

NASCAR: Smooth SVGA!

by David Kadlca » Wed, 08 Mar 1995 06:44:10


  Which is?

David Kadlcak
-------------
Aerospace Engineering Student
Embry-Riddle Aeronautical University
Prescott Campus, Arizona

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
##   To Live Is To Die,       *****************************                ##
##   To Die Happy,            *  Michael Andretti in '95  *                ##
##   Is To Drink Beer.        *****************************                ##
##   -- Vsude dobre, doma nejlip. V Plzni pivo prsi !!!                    ##
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=


rec.autos.simulators is a usenet newsgroup formed in December, 1993. As this group was always unmoderated there may be some spam or off topic articles included. Some links do point back to racesimcentral.net as we could not validate the original address. Please report any pages that you believe warrant deletion from this archive (include the link in your email). RaceSimCentral.net is in no way responsible and does not endorse any of the content herein.