Conference ricks::dechips

+       INTEL DOUBLES ACCELERATED GRAPHICS PORT BANDWIDTH 

If support from software houses at Intel's Visual Computing 
event last week was somewhat thin, Intel Corp compensated by 
wheeling out eight graphics chip companies to support its 
crucial AGP Accelerated Graphics Port technology. 3Dlabs Inc, 
ATI Technologies Inc, Cirrus Logic Corp, Evans & Sutherland 
Computer Corp, S3 Inc and Trident Microsystems Inc have all 
agreed to support the specification, promising to ship systems 
later this year. AGP is designed to provide graphics 
accelerators with direct access to memory, avoiding the current 
PCI bus bottleneck. It's likely to begin appearing in 
workstations around mid-year, and according to Intel will 
dramatically boost graphics performance while slashing system 
memory costs. Intel also announced a high-speed extension 
called AGP 4x, primarily aimed at high-end workstations and 
twice as fast as the current 2x version, which has a bandwidth 
of 512Mbytes/sec. Intel has plans in place to further boost AGP 
to a 10x mode over the next three years. The enhancements will b
e incorporated into the specification by the fourth quarter of 
this year.

    Hi
    
    Will AGP feature in future versions of Alpha? Or are we going down
    a PCI only graphics path?
    
    From a clone maker's view Alpha will only be attractive if the maker can
    use standard parts in his hot Alpha box. As (and if) Intel starts to
    control more tightly the graphics space, (and who will fight them?)
    then the graphics available will be AGP based. If Alpha does not have
    AGP then that is a minus for Alpha. From the Powerstorm (5205) note in
    humane::digital W/S are definitely not interested in supporting the
    clones, hence we will influence the market only via our own efforts;-)
    
    From an OEM's point of view (and the one that concerns me more) he will 
    only use Alpha for graphics if he can be assured that standard 
    graphics will be available. Same argument as above applies. 
    
    Whilst I agree that this question also applies to our W/S people,
    surely the initial development decision rests with SCO.
                     
    Or what are the arguments against AGP?
     
    Any comments anyone? 

While workstation graphics is (currently) not interested in shipping
Powerstorm graphics into the commodity market, it is certainly true
that we're very interested in AGP support in Alpha chipsets.  Our
"entry-level" graphics are likely to remain commodity-based for the
forseeable future, and the market trend will clearly be toward AGP
based solutions.  

To date I've heard little interest in SCO in directly pursuing AGP
support in their chipsets;  they have told us in workstations to
develop our own solutions if we require near-term support for AGP.
I would love to hear that I'm wrong about this...

    "Our "entry-level" graphics are likely to remain commodity-based for the
     forseeable future."
    
    Bzzzt.  The only option that we ship that I would consider a
    "commodity" part on workstations is the S3 Trio64 (3D10) which
    has gone to EOL with no replacement planned by workstations.
    Although I have heard that there is something being considered
    in the future.
    
    Entry level on DEC workstations is TGA2 (3D30).
    

Sadly, what you say is true for VMS.

In the NT space (which is what we are directed to focus upon),
we do ship things like the Matrox Millenium today, and will
continue to ship such commodity graphics in the future.

But then, you knew that. :)
 

    
    I explicitly didn't mention Matrox because it is a rather high-end
    commodity card.  A #9 S3 Trio card retails on the street for under
    $100, with performance near that of proprietary graphics cards just a
    few years ago.  As does the low-end ATI card.  Both of which can be
    purchased on the street for NT.
    
    But you are correct, because NT graphics options typically have drivers
    written by the vendor, with occasional help with Alpha-specific
    problems from us, NT will get commodity parts even without explicit
    assistance from DEC
    
    UNIX and VMS are another story.  Neither have a low-end strategy.
    

Nevertheless, Matrox *is* a commodity card.  And similar commodity
3D cards will continue to be on our roadmaps.  And such cards are
going to migrate to AGP -- Intel will simply force the issue.

As you have noted in the past, TGA2 under its current pricing
structure hardly fits in the category of "commodity" card.

> UNIX and VMS are another story.  Neither have a low-end strategy.

Well, actually, they do.  The strategy is not to offer low-end graphics. :(
Actually, that's not quite right... I've seen roadmaps that offer
commodity graphics cards as part of the graphics offerings for both
Unix and NT;  they probably don't meet your definition of "low-end",
however.  

Is AGP open or licensed by Intel?  I noticed a few graphics companies missing
from their lineup.  I was wondering if they were going to use AGP like slot 1.

Note that a "fast grahic bus" is how SGI has been pulling out their magic for
quite a while.  SGIs claim was you didn't need that absolute fastest processor.
It was the speed of the system that counnted, not just one component.

ac

Intel controls the AGP spec, but is "giving it" to the world.
They don't need to patent it like slot 1;  by controlling the
specification, they can guarantee that they have a temporal
advantadge in bringing AGP products to market -- witness the
fact that enhanced-AGP spec (AGP4) won't be ready until the
end of the calendar year -- coincidentally about the time that
their internally-developed 3D graphics product is supposed to
hit the streets... (but I'm probably just being paranoid, right? :)

re: .8
> SGIs claim was you didn't need that absolute fastest processor.
> It was the speed of the system that counnted, not just one component.

They're right of course.  But having the fastest processor helps --
witness the benchmark wins with PowerStorm graphics on Alpha systems.
You need the right processor, the right bus, the right memory system,
and the right graphics engine.  

AGP may or may not be the "right" bus -- 64-bit EPCI is a good
alternative -- but Intel is the 800-lb gorilla, and AGP is very
likely to become the standard.

    re .6  "NT will get commodity parts even without explicit
    assistance from DEC"
    
    Are you sure ?
    
    E.g. Where is the multihead-capable driver for Matrox on NT V4. For
    Intel, see Matrox's website. For Alpha, search around, fail to find
    anything, if you work for Digital, ask around and be pointed to a
    product manager, he provides a pointer to code, and you can't let
    customers have it. So off they go to their favourite Intel vendor (or,
    if we're lucky and they really really like us, they stay with UNIX or
    OpenVMS).
    
    Does that sound like "commodity" to anybody ?
    
    regards
    john

    The NT model is the PC model.  The option vendors usually develop the
    device drivers themselves.  From what I've seen, unless the card is
    being sold by Digital, the most we do is assist a vendor in making
    their Intel drivers work on Alpha.  This "should" get easier with the
    byte/word I/O on EV6 platforms.
    
    It also means that not all vendors will bother to make their code
    available on Alpha.  Unfortunately, it isn't something FX!32 can solve
    -- the only solution is making it easy enough, and having enough volume
    to make it worth while to the vendors.
    
    The UNIX/VMS model is to write custom driver/server logic for each card
    that we want to support.  While I would welcome (and even help) a
    vendor that wanted to spin their own... the volume just isn't there to
    make it interesting for a vendor who sells the card wholesale for $50
    to invest in.
    

    re: .9                                         
    
| Note that a "fast grahic bus" is how SGI has been pulling out their magic for
| quite a while.  SGIs claim was you didn't need that absolute fastest
| processor.  It was the speed of the system that counnted, not just one
| component.
    
    Yes, they make much of being a systems company.
    
    Let's take a look at how well they did implementing R10000
    workstations:
    
      * The Indigo2 R10000 was slow.
      * The O2 R10000 was slower still (I'm being fair here).
      * Third time is a charm for a "system company" I suppose.
        They finally got a high performance R10000 workstation out
        the door with the Octane.  A little pricey however.
    
    re: .10
    
| They're right [about a fast graphics bus] of course.
    
    Sure.  The "new" (Impact on octane crossbar) SI, SSI, and MXI graphics
    take huge advantage of the much hyped 1.6GB/second bandwidth.  Too bad
    the command processor on the graphics card becomes the bottleneck.
    They hype the 1.6GB/sec crossbar, but don't talk about the tiny fraction
    of that bandwidth that actually gets through to their graphics cards.
    
|   But having the fastest processor helps -- witness the benchmark wins
|   with PowerStorm graphics on Alpha systems.  You need the right processor,
|   the right bus, the right memory system, and the right graphics engine.
    
    The TWO geometry engines in SGI's SSI (and MXI) has a combined 960MFlops
    peak performance.  One Alpha 21164 at 500MHz has more.
    
    And processed vertices require less "bus bandwidth" than raw vertices.
    
    SGI has claimed that Impact class graphics couldn't be done with PCI.
    Digital is *doing* Impact class graphics with a 32 bit 33MHz PCI.
    
    
    Now back to your regular oh woe is VMS on the desktop rathole.
    
    								-mr. bill

    It's too bad that people think that this is all about "the desktop"
    and that just being the fastest is all that big a deal.  It ignores the
    fact that graphics are not just for PC's or for high-end workstations. 
    They are also a requirement on most servers, and most of our
    workstation sales are not for high-end 3D, or even high-end 2D for
    that matter.  You wanna sell workstations?  You don't need a faster
    one, just drop the prices.  Heck, we had a customer who offered to buy
    8000 Multia's -- if they ran OpenVMS (of course, we have no interest).
    
    I haven't been crying in my beer about "VMS on the desktop" for a long
    time.  I do still care that we will do things like lose a huge server
    account because someone in workstations decided not to continue to
    develop *any* 3D support for OpenVMS -- and tell the customer to dump
    VMS for NT or UNIX (and of course, most of the time they end up as a
    HP or Intel customer when this happens).  They don't give a crap, they
    only care about the WS sales, not the server sales that are lost.
    
    I *do* care when the workstation group decides that they won't do
    low-end commodity graphics for VMS and UNIX - because the low-end
    *server* business depends on it.  And it pisses me off, because until
    the workstation group decided to get their thick fingers involved, 
    the UNIX and VMS groups had this space covered.
    
    Now there is *no* server graphics strategy, and no VMS graphics
    strategy at-all.  And while UNIX may get low-end commodity work
    restarted for servers... it's not clear that it will happen for VMS.
    I can't wait until the crisis happens (DEC only reacts to a crisis).
    
    I'm sorry if I still get a little cranked up about this crap, 'cause
    when the dung does hit the fan, I'm sure I won't be able to get out of
    the way, short of finding a new employer.
    

    Your arguments are compelling, Fred.  There are a few of us left that
    still agree with you.  I suspect, though, that the decisions about [the
    lack of] a graphics strategy for VMS/UNIX are not as complicated as is
    your argument.  It probably boils down to nothing more that there are
    no resources to do it.  No budget.  No engineer.  No deep thought. 
    Just a simple equation -- 
    
      sales down =
      revenue down = 
      R&D budget down =
      downsizing = no one left to think about it, much less actually do it
    
      = sales down = ...
    
      We could still make a lot of money by doing a few simple things. 

I would like to buy VMS -- what's the asking price?