[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference hydra::axp-developer

Title:Alpha Developer Support
Notice:[email protected], 800-332-4786
Moderator:HYDRA::SYSTEM
Created:Mon Jun 06 1994
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:3722
Total number of notes:11359

3365.0. "Warp Ltd." by HYDRA::AXPDEVELOPER (Alpha Developer support) Wed Mar 19 1997 17:08

    Company Name :  Warp Ltd.
    Contact Name :  Eric Gullichsen
    Phone        :  
    Fax          :  
    Email        :  [email protected]
    Date/Time in :  19-MAR-1997 17:06:14
    Entered by   :  Marvin Davis
    SPE center   :  MRO

    Category     :  
    OS Version   :  
    System H/W   :  


    Brief Description of Problem:
    -----------------------------
Date: Wed, 19 Mar 97 12:11:12 -0800 (PST)
From: [email protected] (Eric Gullichsen)
To: [email protected]
Subject: bus speed question


Sir/Madam:

I was speaking with some people today at the Palo Alto Alpha
porting center, who referred me to this email address for information.

We have a bus-intensive graphics application we are considering porting
to Alpha. The CPU generates output pixels, and needs to get them into
VRAM as quickly as possible.

So here's the question:
In practice, what is the actual bus throughput from the CPU to VRAM on
the 500 Mhz Digital Personal Workstation?

I know your sales literature quotes theoretical maximum speeds, but
we need to know in reality what throughput can actually be achieved.

For instance, although the PCI bus speed is theoretically 133 Mb/s,
the best throughput we have seen with the latest Intel 430 Triton chipset is
about 28 Mb/s.

Please let me know.

Thanks,


- Eric Gullichsen
  Warp Ltd.
  S.S. Vallejo
  Varda Landing
  Sausalito CA 94965

  FAX:  +1.415.332.8003
  Internet: [email protected]
  Home Page: http://www.warp.com/

T.RTitleUserPersonal
Name
DateLines
3365.1PM contacted, some info sent...AMCUCS::SWIERKOWSKIQuot homines tot sententiaeWed Mar 19 1997 21:0675
Greetings!

  Eric and I met yesterday.  I mailed him the web page with all the "Info-
Centers" listed, (pointing him to both the "Digital Semiconductor InfoCenter"
and "Alpha Workstations" InfoCenter.  Pointed out he'd need detailed feeds 
and speeds from a Product Manager.  I also mailed him (today) the marketing
announcement on the 21164PC and told him that I had identified the Product
Manager for the AlphaStation 600 and sent her mail asking for some pointers
to the technical information Eric asked for.  To date, I have gotten back a
couple of emails from Eric, but no response from the Product Manager.

  As soon as I get a response from the Product Manager, I told Eric I'd pass 
along what I found out.  In case I get hit by a truck, here's the message I
sent off to the Product Manager yesterday.  Let me know if you want the rest
of the email traffic I've had with Eric...



						Tony Swierkowski
						Digital Equipment Corporation
						Software Partner Engineering
						Palo Alto, California
						(415) 617-3601
						"[email protected]"

P.S.	Message I sent to the PM, awaiting a response...

From:	AMCUCS::SWIERKOWSKI  "Tony Swierkowski" 18-MAR-1997 15:52:24.00
To:	SMTP%"[email protected]"
CC:	SWIERKOWSKI
Subj:	Query on behalf of a developer...

Greetings!

  My name is Tony Swierkowksi and I'm an engineer working for the Software 
Partner Engineering group out in Palo Alto, CA.  The Palo Alto SPE center 
and the Marlboro SPE center are the only two SPE groups in the Americas 
that provide members of the "ASAP" (i.e. "Association of Software & Application
Partners") with onsite access to porting centers.

  An ISV in the porting center today is trying to assess whether to undertake
a port of a very compute intensive graphics application and asked me for 
speeds and feeds on an AlphaStation 600 5/500 that I couldn't answer.  Their
application (all done in software) "warps" video images (static or using a
real-time video feed) for display on a workstation providing very fast mani-
pulation (pan, zoom, etc.) of the image(s) being "warped" by their application.
The customer is suspects that an Alpha-based solution may be just the answer,
since top-end SGI workstations just can't deliver (especially for end-users
that want higher-than-normal resolution and/or refresh rates.

  As such they want to be able to push the maximum number of bits from CPU 
registers and/or Memory directly to the VRAM.  Their software does a very small
number of calculations on each pixel to achieve the transformation they perform
and the Alpha is already proven to be very good at these computations.  He was
curious about practical bandwidth via the PCI to the actual graphics controller.
From my understanding, PCI-64 has a theoretical peak of 266MB/second, but in 
reality the customer suspects that this is really limited by the PCI bridge 
between the PCI bus and the CPU/Memory Interconnect.  I told him he'd probably 
need detailed design specification information in order to get real (versus 
theoretical) bandwidth numbers and that I'd try to contact a product manager 
on his behalf, hence this note to you. 

  If you are the right person to at least point me in the right direction for
this sort of thing, I'd be grateful for any assistance you can provide.  Feel 
free to call or send email, cheers...


						Tony Swierkowski
						Digital Equipment Corporation
						Software Partner Engineering
						Palo Alto, California
						DTN:  543-3601
						(415) 617-3601
						"[email protected]"

3365.2Speeds & Feeds sent, customer happy..AMCUCS::SWIERKOWSKIQuot homines tot sententiaeThu Mar 27 1997 19:33170
Greetings!

  The AlphaStation 600 Product Manager (i.e. Connie Doyle) had one of her 
engineers (i.e. Mark Haq) call me last Friday with enough speeds & feeds to
answer Eric's questions.  I crafted up a message (appended below) on Monday
and sent it to him.  Judging by his response, he is happy.  I believe this
call can be closed now...



						Tony Swierkowski
						Digital Equipment Corporation
						Software Partner Engineering
						Palo Alto, California
						(415) 617-3601
						"[email protected]"

P.S.	Message I sent to Eric is below in case anyone wants to know...

From:	AMCUCS::SWIERKOWSKI  "Tony Swierkowski" 24-MAR-1997 16:35:35.45
To:	SMTP%"[email protected]"
CC:	VU,SWIERKOWSKI
Subj:	Speeds & Feeds per Engineering...

Greetings!

  Tony Swierkowski @ Digital here with a follow-up to my query last week I had
sent to the AlphaStation 600 Product Manager.  Side note:  Most PM's are *VERY*
busy - unless they get pre-empted from on high, many days (or even weeks) to 
turn around "What if?" & "How much?" sorts of questions is not unusual.

  Now the good news, the PM must have waved her wand long enough to get one
of the AlphaStation engineers to call me back last Friday with some relevant
technical information re: speeds & feeds.  As you know the AlphaStation 600
originally came in a "AlphaStation 600 5/333" flavor (I think also maybe a 
5/266 flavor?) back when the "AlphaStation 600" family was announced some time 
ago.  Back then, the engineering team wanted to be able to supply a "bleeding 
edge" box and so they provided those systems with two PCI-32 slots and two 
PCI-64 slots.  In theory the industry was going to start screaming for more 
bandwidth than the PCI-32 (i.e. theoretical peak of 133MB/second) could provide
and everyone was going to start designing lots of high-performance PCI-64 
adaptors (i.e. theoretical peak of 266MB/second) for the next generation of 
top-end workstations.  Great idea in theory - never panned out in reality, so...

  The next member of the "AlphaStation 600" family wasn't going to waste time 
designing in PCI-64 slots that nobody built adaptors for.  Instead they went 
for maximum MIPS (using the fastest Alpha Microprocessor available at the time)
and stick to an I/O sub-system using *just* PCI-32 slots.  The result of that 
engineering goal was the "AlphaStation 600 5/500" you had inquired about when 
you and I talked last week.  Like it or not (for now at least) the *fastest* 
compute engine (i.e. quad-issue EV5 chip @ 500Mhz) is wired to PCI-32 slots 
(i.e. theoretical peak of 133MB/second).

  If/when a newer, faster compute engine is designed for the "AlphaStation 600" 
family and *IF* the industry embraces PCI-64 technology, I suppose a follow-on
to the "AlphaStation 600 5/500" could be expected to have not only more raw MIPS
but a faster hose to the I/O world.  Before you ask, I won't discuss possible
future products or speculate further.  But you don't have to wonder too hard to
realize that the "hot" box last month/year is only a stepping stone to next
months/years "hot" box, 'nuff said...

  As to the "AlphaStation 600 5/500" itself, here is what the engineer told me:

 1 PCI-32 slot (133MB/sec peak) using TGA2 graphics in DMA mode:

	- Read  = 48MB/sec
	- Write = 115MB/sec (Longword transfers)
	- Write = 128MB/sec (Quadword transfers)

 1 PCI-32 slot (133MB/sec peak) using TGA2 graphics in Programmed I/O mode:

	- Read  = 24MB/sec (32 byte transfers)
	- Write = 84MB/sec (32 byte transfers)

NOTE: All speeds above are *sustained* (not peak!), unless otherwise noted.

  Another thing the engineer told me about that may be of interest is the 
"CatsEyes" controller (i.e. PowerStorm 4D40T) while more expensive may provide
a performance benefit despite being apparently limited by the theoretical peak
bandwidth of PCI-32 (133MB/second).  The drawback to using this graphics
controller is a) it needs two PCI slots and b) it ain't cheap.  The advantage to
using this graphics controller is it can do Reads & Writes concurrently and thus
effectively you have a combined theoretical peak of something approaching the
PCI-64 slot (in theory).  NOTE: The engineer was *not* a graphics guru, so he
was making his best guess at the details re: the "CatsEyes" controllers.

  I have no idea on the ratio of Read vs Write your application might exhibit
(for that matter I don't know graphics well enough to comment on how much a
given amount of "Write"ing might implicitly or explicitly generate some amount
of "Read"ing (or vice-versa).  From what little I know, it sounds like you
would supply your own driver to push bits out as fast as possible from the
CPU/Memory sub-system (limited by the raw MIPS of a given compute engine) to
the graphics controller's VRAM.  If this is the case, and a "Write" does not
necessarily incur a "Read", I would suspect your application really would only 
be limited by the sustained "Write" rate (using DMA mode) for a given platform.
Hope this helps, cheers...


						Tony Swierkowski
						Digital Equipment Corporation
						Software Partner Engineering
						Palo Alto, California
						(415) 617-3601
						"[email protected]"

P.S.	Here's the URL specific to the "AlphaStation 600" family's I/O
	characteristics...

From:	<zapped by me>
To:	Tony Swierkowski <[email protected]>
CC:	
Subj:	AlphaStation 600 5/333 I/O

This is a multi-part message in MIME format.

--------------1DDC6B537BCD
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

http://www.workstation.digital.com/products/a600/io2.html

--------------1DDC6B537BCD
Content-Type: text/plain; charset=us-ascii; name="io2.html"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline; filename="io2.html"

[Digital]
[Image]
[Image]
AlphaStation 600 System I/O

   AlphaStation                     600 5/333                   600A 5/500
 Max. memory:                              1 GB ECC
 Max. storage,
 internal:                                  36 GB

 Storage:            6 storage bays (two 5-1/4", three 3-1/2", one 3-1/2"
                                           floppy)
 Max. I/O
 throughput:                        PCI: 266 MB/sec. (PCI)
                  3 64-bit PCI slots, 1 32-bit PCI slot, 1
 IO:              32-bit PCI/EISA slot, 3 EISA slots; 2        7 32-bit
                  serial ports, 1 parallel port, 1 floppy      PCI, 2 EISA
                  port; 2 FAST WIDE SCSI-2
 Optional I/O
 support                               FDDI, Token Ring
 Network support    Integrated twisted-pair/ ThinWire/thick wire Ethernet
 Operating systems      Digital UNIX, OpenVMS, Windows NT Workstation
                  Built-in 16-bit, stereo quality audio in/out; FullVideo
 Graphics &       Basic, FullVideoTM Supreme, FullVideo SupremeTM JPEG
 multimedia       video in/out; 3D: PowerStorm, 3D30, 4D40T, 4D50T, 4D60T,
                  ZLXp-L1, and ZLXp-L2

[Image]

News & Events | Promotions | Workstations | Solutions | International

 [HR]
 [WorkStation Home] [Image][Image]

Updated: 27 February, 1997.

) Digital Equipment Corporation, 1995. The DIGITAL logo is a trademark of
Digital Equipment
Corporation. All other trademarks are the property of their respective
owners.

Legal

--------------1DDC6B537BCD--