I'm having a hard time accepting this. I can't believe that Intel
Engineers
would be so stupid as to even cripple I/O performance in favor of memory
performance. Also, the above statement seems to me as a contradiction.
If
I/O performanace is so poor, why do they recommend a SCSI card? It
itself
is an I/O subsystem with some cards being able to do DMA transfers, ie.
bus
masters.
Is there another way to test DMA transfer speeds besides DMADIAG?
Alex
Quoted text (35 lines)
>On Thu, 19 Dec 1996 09:53:33 -0500, Alex Fernandez >>> >>Subject: RE: P6SNE bad Verite DMA performance? >>> >>Hello Christopher, >>> >> The Pentium Pro processor has extremely poor I/O performance. This is >>> >>one of the reasons SCSI is always recommended. The memory functions >>> >>are optimize at the sacrifice of the I/O sub-systems. There is not >>> >>work about with this issue. >>I'm having a hard time accepting this. I can't believe that Intel >>Engineers >>would be so stupid as to even cripple I/O performance in favor of memory >>performance. Also, the above statement seems to me as a contradiction. >>If >>I/O performanace is so poor, why do they recommend a SCSI card? It >>itself >>is an I/O subsystem with some cards being able to do DMA transfers, ie. >>bus >>masters. >>Is there another way to test DMA transfer speeds besides DMADIAG? >>Alex > SCSI devices require much less internention from the CPU. For >instance some EIDE CD drives may consume as much as 50 % of your cpu's >time. Not good. SCSI on the other hand uses much less. Can't remember >the exact figures. > I think the reference to SCSI has nothing to do with the Reactor >card. He was just refering to what is suggested for general I/O >performance if you buy a PRO.> I'm having a hard time accepting this. I can't believe that Intel
> Engineers
> would be so stupid as to even cripple I/O performance in favor of memory
> performance. Also, the above statement seems to me as a contradiction.I've got the impression that the PPro has a poor PCI-Performance, because
it has a primary bus called GTL+, which is supposed to be good at
multiprocessing. It seems to be difficult to have both good multiprocessing
and PCI performance at once, and intel went for the first.
Besides, the _main_ memory performance isn't that great, compared to current
Pentium chipsets (~40-46 MB/s both). The on-chip 2nd-level-cache is the
thing that's really fast.
--
Ride on...
Boris Woerner
On Thu, 19 Dec 1996 09:53:33 -0500, Alex Fernandez
SCSI devices require much less internention from the CPU. For
instance some EIDE CD drives may consume as much as 50 % of your cpu's
time. Not good. SCSI on the other hand uses much less. Can't remember
the exact figures.
I think the reference to SCSI has nothing to do with the Reactor
card. He was just refering to what is suggested for general I/O
performance if you buy a PRO.
Vincent Aldridge
Been playing games on PC's since 1980
I be the DOOM GOD .... Bow to me....!!
Quoted text (22 lines)
>> >>Subject: RE: P6SNE bad Verite DMA performance? >> >>Hello Christopher, >> >> The Pentium Pro processor has extremely poor I/O performance. This is >> >>one of the reasons SCSI is always recommended. The memory functions >> >>are optimize at the sacrifice of the I/O sub-systems. There is not >> >>work about with this issue. >I'm having a hard time accepting this. I can't believe that Intel >Engineers >would be so stupid as to even cripple I/O performance in favor of memory >performance. Also, the above statement seems to me as a contradiction. >If >I/O performanace is so poor, why do they recommend a SCSI card? It >itself >is an I/O subsystem with some cards being able to do DMA transfers, ie. >bus >masters. >Is there another way to test DMA transfer speeds besides DMADIAG? >AlexSCSI cards will use bus-mastering to transfer data - the CPU does not need
to do I/O for data transfer. IDE requires the CPU to read the data via I/O
from the IDE controller.
/THansen
Quoted text (44 lines)
> On Thu, 19 Dec 1996 09:53:33 -0500, Alex Fernandez > >> >>Subject: RE: P6SNE bad Verite DMA performance? > >> >>Hello Christopher, > >> >> The Pentium Pro processor has extremely poor I/O performance. This is > >> >>one of the reasons SCSI is always recommended. The memory functions > >> >>are optimize at the sacrifice of the I/O sub-systems. There is not > >> >>work about with this issue. > >I'm having a hard time accepting this. I can't believe that Intel > >Engineers > >would be so stupid as to even cripple I/O performance in favor of memory > >performance. Also, the above statement seems to me as a contradiction. > >If > >I/O performanace is so poor, why do they recommend a SCSI card? It > >itself > >is an I/O subsystem with some cards being able to do DMA transfers, ie. > >bus > >masters. > >Is there another way to test DMA transfer speeds besides DMADIAG? > >Alex > SCSI devices require much less internention from the CPU. For > instance some EIDE CD drives may consume as much as 50 % of your cpu's > time. Not good. SCSI on the other hand uses much less. Can't remember > the exact figures. > I think the reference to SCSI has nothing to do with the Reactor > card. He was just refering to what is suggested for general I/O > performance if you buy a PRO. > Vincent Aldridge > Been playing games on PC's since 1980 > I be the DOOM GOD .... Bow to me....!!Quoted text (44 lines)
> On Thu, 19 Dec 1996 09:53:33 -0500, Alex Fernandez > >> >>Subject: RE: P6SNE bad Verite DMA performance? > >> >>Hello Christopher, > >> >> The Pentium Pro processor has extremely poor I/O performance. This is > >> >>one of the reasons SCSI is always recommended. The memory functions > >> >>are optimize at the sacrifice of the I/O sub-systems. There is not > >> >>work about with this issue. > >I'm having a hard time accepting this. I can't believe that Intel > >Engineers > >would be so stupid as to even cripple I/O performance in favor of memory > >performance. Also, the above statement seems to me as a contradiction. > >If > >I/O performanace is so poor, why do they recommend a SCSI card? It > >itself > >is an I/O subsystem with some cards being able to do DMA transfers, ie. > >bus > >masters. > >Is there another way to test DMA transfer speeds besides DMADIAG? > >Alex > SCSI devices require much less internention from the CPU. For > instance some EIDE CD drives may consume as much as 50 % of your cpu's > time. Not good. SCSI on the other hand uses much less. Can't remember > the exact figures. > I think the reference to SCSI has nothing to do with the Reactor > card. He was just refering to what is suggested for general I/O > performance if you buy a PRO. > Vincent Aldridge > Been playing games on PC's since 1980 > I be the DOOM GOD .... Bow to me....!!Quoted text (11 lines)
|I am Vwariff the knight of mirror . . . | |I bow to no one that doth bow to me. If Ye come seeking drink & droll |belly up to the bar. If Ye come seeking war -- there be virtual war if |and only if you ask for it. Ye need only make the dreaded system call |vwariff().On Mon, 23 Dec 1996 23:26:10 -0800, Keith & Cathy Bancroft
>My first forays with a HACK dungeon were in a VAX750 on a REGENT 200 but
>I digress . . .
>AHH for an old fashioned 2 meg VAX with a dungeon full of gamers and an
>Amulet of Yendor for the taking -- the Hell Hound hits!
>Vwariff :-}
YES!!......wasn't that a great game?!?! I played it for nearly 4
years! Still start it up ocassionally >|-)
kev


































