#9 GXE @50 MHz? for dream system
Anthony Ta Hui Loh
VRAM) can be operated in a 50 MHz VL-Bus slot??? Also, do you know if
it happily coexists with any other card, like the A.I.R. VL-Bus fast
(but not wide) SCSI-2 host adapter at 50 MHz???
Please send replies directly to al...@bbn.com, in addition any postings
you may make.
I called Number Nine at 617-674-0009 today, but too late for a verbal
answer; their fax-back information doesn't contain that spec :-(.
I'd prefer a report of actual experience anyway.
WARNING:
For those of you who are going on to finish reading this 422-line
posting, be warned that the remainder of it skims over the great many
issues I became aware of while learning to choose the components for
the "dream" system I intend to buy (ordering the week of June 6).
I've worked on computers for years, but avoided learning much about
the 80X86 line until I became convinced the expenditure would yield a
reasonable Unix machine, i.e. run linux, in addition to a DOS, OS/2
and MS Windows platform. I started to learn two months ago.
At first I thought I'd buy a complete system from an integrator. In
fact, I chose a Zeos DX2/66 desktop for use as a Windows machine.
While we are happy with the Zeos and, at ~$3000, it is cost-effective
for Windows, it doesn't have the long-term (three years before
hopelessly obsolete) expandability I desired for a multi-OS platform.
So, I felt forced to build a system for myself out of components.
I tried to learn everything of relevence and made my choices based on
net hearsay, or hearsay from phone calls, friends, or fax backs, and
articles from the many magazines. I don't own any of this stuff, yet,
so I hope people email me any corrections for any serious conceptual
or factual errors.
I hope this posting and more the comments it generates, provides all
the other people going through what I just went through with a broad
overview of the criteria involved in selecting components.
Again, please send comments directly to al...@bbn.com, in addition to
any postings you may make, before I make an expensive mistake :-).
=*=*=*=*=*=*=*=*=*=*=* On the "dream" motherboard *=*=*=*=*=*=*=*=*=*
FLAME ON... I found very little material comparing motherboards. I
realize that it's hard to test them against each other since it would
take a number of different sets of peripherals built up around each
board, in a runoff. Still, I was disappointed at the lack of
available information on motherboards, given the willingness of the
media to compare a dozen N" monitors, a handful of CD ROM drives, or a
score of video accelerators in generally useful reviews. I found
comparisons of complete systems (whose performance was more dependent
on their video accelerators Winmarks) useless for choosing a
motherboard. In fact, few reviewers even mentioned the motherboard
manufacturer while reviewing dozens of systems.
...FLAME OFF.
Without much data, I convinced myself that it was a safe decision to
buy this particular A.I.R. (Advanced Integration Research)
motherboard. There were a couple of positive c.s.i.p.h
recommendations for the company (one mentioned this model) and no
complaints. It's used in Zenon's top of the line offering.
A.I.R. voice (408)428-0800 fax (408)428-0950
2188 Del Franco St. products: at least motherboards and VL-Bus IDE
San Jose, CA 95131 and SCSI controllers, and a VL-BUs Ethernet
card with Netware, DOS, OS/2, Unix support
- Model "486EI" is a 6-layer AT-size PCB with 8 total slots:
3 VL-bus full bus mastering, and
6 capable of full EISA bus mastering.
(Right, 3 VLB + 6 EISA does not add up to 8).
- The board supports either DX/50 or DX2/66 CPU chips (and slower ones
I suppose) in a 238-pin ZIF socket, AMI BIOS (I forgot to ask if
BIOS was in a flash chip and therefore reconfigurable from a floppy,
but I expect so because: frequency reconfiguration is by jumper and
software (the crystal doesn't need to be changed).
- includes 256K 15ns write-back (not write through) cache (direct
mapped I assume, since an A.I.R sales person, with spec in hand,
couldn't tell me if it was the more expensive 2-way or 4-way
associative). SRAM can be added to total 512K cache. (They seem to
have very few and very busy technical support people which is why I
accepted speaking to a sales person on my second call to AIR.)
- Four 72-pin SIMM sockets support 256 Mbytes DRAM using four 64 Mbyte
SIMMS. I am going to buy one 16 Mbyte 72-bin 60ns SIMM. A.I.R.
says that using 60ns DRAM is just a little faster than the 70ns and
80ns DRAM. (I presume I can configure the number of wait states and
the clock frequency on the memory bus to take advantage of the extra
cost of the faster DRAM.) Note that the VL-Bus/CPU clock frequency
is independent of memory bus clock frequency. (Some reviewer
claimed that DRAM could be too fast for a motherboard that isn't
designed to refresh a fast DRAM often enough. That's not a problem
on this board.)
- The board has UMC 82C481BS and 82C482AS chips which I presume
implement the EISA bus. Note that there are no plain ISA slots so I
believe I can avoid 16 Mbyte memory access limitation problems by
avoiding using plain ISA cards (at least for anything fancier than
2S/2P/1G). The board does not provide 2S/2P/1G I/O ports nor any
other on-board peripheral ports; just the (I hope) completely
properly implemented busses. There is no indication its EISA
implementation suffers from chip set bugs, i.e. the one causing
16-bit DMA transfers to fail because the /DMAMEMR MEMORY READ line
on the 82C206 doesn't switch from active low to tri-state fast
enough.
- AIR said revision 1.5 is shipping now, when I called them on June 1,
1993. (Since the afore-mentioned chip set bug is allegedly fixed by
now!)
- Best price I can find on board and CPU (includes 256K cache but zero
DRAM) is DX/50 $865; DX2/66 $950 from Ramtek at 818-718-0894.
- Zenon uses this board in their top-of-the-line "Z-Bus Dream III -
SCSI-2 EISA/VLB" system advertised in the June 1993 "Computer
Shopper", with the AIR fast (but not wide) SCSI-2 VLB and Diamond
Viper VLB accelerator. They also offer the Orchid VLB at 50 MHz,
but say the ATI Graphics Ultra Pro will only run in a 33 MHz VL-Bus.
=*=*=*=*=*=*=*=*= On the DX/50 vs. DX2/66 argument *=*=*=*=*=*=*=*
I'm inclined to believe the quasi-religious arguments that a DX/50
VL-bus system, running a pre-emptive multitasking O/S (i.e. linux) to
take advantage of its higher I/O VESA local bus speed, might overcome
the disadvantage of its slower DX/50 CPU to exceed the performance of
the same system with a DX2/66, running the same O/S. (Anyone have
data? Using what combination of boards?) I'd say this is most likely
true for multiple simultaneous compilations.
On the other hand, I might never put disk I/O on the VLB because...
=*=*=*=*=*=*=*=*= On EISA vs. VL-Bus for disk I/O *=*=*=*=*=*=*=*
In the April 1993 Byte, "Two Ways to Say VL-Bus", p. 179, a reviewer
compares an UltraStor 24F EISA bus SCSI disk controller with an
UltraStor 34F "ISA-bus card that uses VL-Bus for added throughput"
(huh? isn't that true of all VLB cards?) in both an AMI "Enterprise
III" and a Micronics "EISA/VL-Bus System Board" each equipped with two
VL-Bus slots, 8 Mbytes of DRAM and 486 DX33 chips. He finds that
"...surprising. On both motherboards, the EISA adapter ran
significantly faster than the VL-Bus version. On the
Micronics...24F...10% advantage in read... Accoring to UltraStor,
VL-Bus interface boards that need to move large blocks of data are
very sensitive to system architecture (e.g., factors like memory
interleave). ...For either system I would recommend an EISA SCSI card
over a VL-Bus host adapter if maximum performance is required."
(Frankly, this review makes me suspect the UltraStor 34F being used as
a standard is the poor performer, not the two premium motherboards
under test! Anyone have data?)
Zenon says the AIR fast SCSI-2 VL-bus controller has faster disk I/O
than anything they could construct using an EISA controller, but maybe
their tests ran under (the merely) cooperative multitasking MS Windows
which I would guess depends completely on bus speed because only a
single task is accessing the disk so it doesn't utilize DMA
effectively because that task is waiting for I/O completion anyway???
(Also, the AIR disk controller is non-caching, which is better than
configuring both RAM cache and also controller cache, see below)
Many system integrators put both video and disk I/O on the VL-Bus.
However, I'm inclined to use both the VL and EISA busses, since that
allows disk and video DMA to occur on both busses independently, under
a pre-emptive multitasking O/S.
=*=*=*=*=*=*=*=*=*=*=*=* On disk caching *=*=*=*=*=*=*=*=*=*=*=*=*
In at least two reviews (sorry references ar at home) RAM caching
performed better than caching on host adapter cards. RAM caching for
MS Windows is allegedly optimal at 1 MByte, (any larger and checking
for cache misses becomes too much overhead).
Netters claim Unix makes better use of RAM buffers than DOS so that
caching on the host adapter is not important (but no one has said it
contributes to excess overhead).
In either case the conclusion is easy to take: spend $$$ on general
purpose RAM instead of special purpose RAM on the host adapter.
Furthermore, many modern disk controllers (by which I mean the card
attached to the HDA, Head Disk Assembly, which makes the HDA an IDE or
SCSI disk, not the host adapter card) seem to have 64K cache, or more.
I can see how having that cache on the host adapter might be superior
to having it on the disk controller, but I think it's a close call and
strongly dependent on lots of factors like disk fragmentation. I
believe the existence of disk controller cache (which you cannot avoid
paying for) makes the host adapter cache redundant.
=*=*=*=*=*=*=*=*=*=* On the future of the VL-Bus *=*=*=*=*=*=*=*=*=*
I prefer to use the DX/50 chip so I can run the VL-bus at 50 MHz. I
am inclined to believe that VESA card makers will increasingly build
cards that work at 50 MHz. Until VESA, I have never heard of a bus
spec that relates the number of cards on the bus to the maximum clock
speed on the bus: one VESA card at 50 MHz, 2 at 40 MHz and 3 at 33
MHz. (That is the CORRECT standard, isn't it? I've heard so many
other versions on the net.) There are many "2 at 50MHz" combinations
being sold today (i.e. Zenon).
I also noticed that the brand new Pentium boards support VL-bus and
EISA. (I haven't seen a single PCI bus.) Can anyone explain if the
Pentium boards' VL-busses run at the Pentium CPU clock speeds of 60
MHz and 66 MHz??? I'm quite confused about this since those speeds
are dangerously above VESA specs, but clocking slower than the Pentium
CPU clock would (to my mind) ruin the "local" concept.
I think it may, ironically, make more sense to configure a system to
use its VLB for fast SCSI-2 disk I/O and put the video I/O accelerator
on the EISA (or even ISA) bus, because (a) video performance is more
dependent on the accelerator card implementation than its interface
bus, (b) is generally adequate no matter what bus it's on, and (c)
except for BitBlt the video accelerator has largely off-loaded the
need for large memory transfers over whatever bus it is connected to.
The same April 1993 Byte review of the AMI "Enterprise III" and
Micronics "EISA/VL-Bus" referenced above, "Two Ways to Say VL-Bus",
said the identical Diamond "Stealth 24" circuit showed less than
double the improvement for BitBlt when interfaced to the VL-Bus vs.
the archaic ISA bus, and far less than a 25% improvement on "Basic
Graphics" which I take to mean computation offloaded onto the Stealth
accelerator circuit.
=*=*=* On video accelerators for DOS, MS Windows, and X Windows *=*=*=*
I need to run applications under all three: DOS, MS Windows, and X. I
can't afford a coprocessor board and suspect they don't have as
ubiquitous driver support. So, I first carefully reviewed the ATI
Graphics Ultra Pro VLB, but discovered it won't run on a 50 MHz bus.
So restarting my selection process all over again, I found recent
reviews of the #9 GXE encouraging.
Netters say the ATI Graphics Ultro Pro VLB card, whose Mach32
intruction set is a superset of 8514 commands, is great with XFree86
(public domain X) 8514 drivers. And, the GUP will do 1280x1024
non-interlaced at some reasonable vertical refresh rate (netters say
76 Hz?). However, netter's postings convinced me that soon S3X (and
later this year, XFree86 incorporating S3X technology) would provide X
windows drivers to support the S3 86C928 chip on the #9 GXE by the
time I managed to build linux. I was also pursuaded that the generic
S3 chip promised longer-term continuing driver development vs. the
GUP's Mach 32. And it seems an obvious waste to depend merely on the
GUPs 8514 compatibility to do X Windows.
Both the ATI GUP and #9 GXE support:
- virtual screen mode
(GXE: 2048 x 1024 @256 colors w/3Mbytes, @16 colors w/2Mbytes on the GXE).
- on-board storage of fonts
(GXE - have to buy that third MByte of DRAM on top of the the fist
two Mbytes of VRAM)
- reasonable refresh rates at 1280 x 1024
(GXE: 72 Hz)
- good DOS performance instead of the crippled DOS performance
suffered by many other accelerator boards with excellent Windows
performance. (Neither board is as fast as a Diamond Viper or other
P9000 based products. But the P9000 will not recieve public domain
X support anytime soon).
- Number Nine says GXE is available for VESA Local Bus, EISA and ISA.
I think ATI GUP is EISA and VLB, not sure about ISA???
Note ATI GUP is not to be confused with ATI Graphics Ultra Plus (GU+)
whose X11 performance was excellently examined and documented by
Gottfried Rudorfer. He has made his report available via anonymous
ftp from exaia.wu-wien.ac.at, file /linux/ati_perf.taz requires GNU
gunzip and tar to produce his reports. (Basically, his 486 DX2/66
with a ATI GU+ and X8514 server clobbers a Sun 3/50.)
Both ATI GUP and #9 GXE are reputed to have good driver support
(except existing Xfree86 8514 support for the ATI GUP is a bird in the
hand, versus anticipated XS3 support for for the S3 86C928 on the
GXE!!!)
These are features of the Number Nine GXE which led me to prefer it
over the ATI GUP (which I don't believe has such features):
- X Windows-like resolution of 1152x819; xdpyinfo on my Sun 3/50
workstation says: "1152x900 pixels (325x254 millimeters)" (GXE
supports 256 colors w/2MB, 65K colors w/3MB)
- up to 100 Hz refresh rates at resolutions below 1280x1024 (i.e. at
1152x819).
- Flash BIOS (floppy upgradeable)
Also, I believe the GXE is more likely to handle full-motion video (if
applications get developed while I own it) because it seems 50% faster
than the GUP, which itself seemed to be a winner among the 1992
products.
List prices from #9's faxback (June 4, 1993) are:
1Mbyte VRAM ("GXE Level 10") $345
2Mbyte VRAM ("GXE Level 11") $495
3Mbyte (2Mb VRAM, 1MB DRAM) ("GXE Level 12") $595
which are cheaper than the list prices quoted in the few reviews I've
seen for #9 and almost all the real street prices I've seen quoted in
June Computer Shopper! The nasty ploy by Number Nine, to beat the old
2.5 Winmark tests by hard-coding one of the text strings, has probably
hurt their sales by inhibiting the media from even reviewing the GXE
and I believe that handicap and the usual competition has caused them
to drop their prices. (I only found reviews recently.)
=*=*=*=*=*=*=*=*=* What else I'm going to buy *=*=*=*=*=*=*=*=*=*
DTC 2290 EISA IDE Host Adapter
Seems like a great way to attach lots of cheap IDE drives, etc,
cheaply, especially with street prices just below $150. I _presume_
linux drivers will have no problem with it (anybody using it???)
From the fax back:
"8 Mbytes/sec burst, 3.3 to 5.3 Mbytes/sec sustained" "multi-sector
read/writes" (must be supported by disk), "limited only by data
transfer capability of the IDS drive(s)"
"up to 4 IDE disks, two IDE channels support mirroring/ duplexing
under NetWare 386", co-residency is supported for additional fault
tolerance" (I interpret that to mean one DTC 2290 is "co-resident"
with another)
"In addition to hard drive support, also supports up to 4 floppy
disk drives including 2.88 Mbyte"
"supporting 4 IDE drives and 4 floppy drives on a single card"...
"different size, types and performance of IDE can be intermixed on
the same card" (but I saw the board's driver gets configurd for the
slowest IDE drive)
"can support IDE tape drives as well"
Teac 5.25" 1.2 Mb and 3.5" 1.44Mb floppy drives:
For compatibility with all those disks containing drivers, that come
with the boards I'll be buying, for somewhere around $55 each. No
reason to buy 2.88 Mbyte floppy; to small for backups and no one is
handing me 2.88 Mbyte disks. Teac seems to be the most reliable and
you can't save much more than $12 each for cheaper brands.
Colorado Jumbo 250 DC2120 tape drive
Seems to be almost a standard. Need it for compatibility with
other machines using DC2120 tapes. Apparently Unix unfriendly,
but I don't have a choice (anyone using it with linux???)
Quantum 240LPS
Give me that two year warranty vs. all other IDE manufacturers one
year warranties. The last thing I need is a dead disk (256K cache
on controller, supports multi-sector read/writes, reasonably fast.)
Honeywell mouse
Smoother action with those two rollers than any ball, three buttons
for X Windows.
Viewsonic "New Generation" 17 (Model 1782, not a "7") 17" monitor
Great reviews about focus, colors and quality control (often arrives
perfectly adjusted at factory settings), great resolution (1600 x
1200) high refresh rates, has two moire controls, pincushion and
keystone image shape controls in addition to the other more popular
ones. Only adjustment I've ever heard of that it's missing is the
ability to rotate the whole image (i.e put a book under left or
right edge). Should be adjustable for useability until it finally
dies.
case/power supply - I'm getting a 13 bay case and a 400 watt power supply
because I don't want to think about them ever again.
What a pain an upgrade would be! It ain't portable,
but then I won't have to buy any external anythings!
It's easy to find UL approved supplies these days and
FCC B ("home" standard) cases (higher than FCC A
("office" standard).
=*=*=*=*=*=*=*=*=* Thins I'm still fussing over *=*=*=*=*=*=*=*=*=*
keyboard - I'd like to find one for the PC with a Control key just to
the left of the "a" key: any suggestions?
UPS - I found out that "uninterruptible" means a Tripplite will provide
battery backed-up power for 15 minutes at 1/2 of its rated power.
Useful if you never want a data destroying accident to occur while
you're working, but obviously doesn't allow you to walk away for the
weekend.
Power Surge Protection - I found out that one brand (ISOBAR, I'm pretty
sure) is the only brand of power strip to isolates each AC outlet
from the others.
*=*=*=*=*=*=*=*=*=*=* Things I don't need now *=*=*=*=*=*=*=*=*=*=*
Even if I want to buy them next fall, things will likely have changed
drastically. (The rule is "if you bought it, it's obsolete".)
fast SCSI-2 - leaning toward AIR VL-Bus, but it's so confusing and
expensive. Don't need drives anytime soon, I'll be adding
IDE drives for a while.
SCSI drives - almost all have 5 year warranty (in contrast to much cheaper
IDE drives), but Quantum has much fewer parts (i.e. they use
ASICS and mechanical ingenuity) so probably still is most
reliable.
Toshiba 3401 CD ROM drive - pricey but full-featured: two spin rates (spins
faster when it figures out the data is all in one track) and
recognizes Kodak multi-sessions, not just the first Kodak
session where multiple photographs are stored on the CD ROM,
but subsequent visits to the Kodak store for additional
session. Going to hold off on this too. Price wars ahead?
Sound card - Musicians have their own requirements. At this point all
I'm willing to say is that 16-bit boards are cheap enough
(and necessary to reproduce CD quality sound with 44 KHz
sampling) that I wouldn't consider 8-bit boards. Also,
at least the Logitech and Media Vision products support
generic SCSI interfaces (vs. proprietary!) sufficiently
fast for those slow CD-ROMs. Think about buying one of
these before buying a cheap SCSI controller.