Satisfying The Stockholders

In the short term we don't have to satisfy anyone but ourselves.  Some of our very best engineers are working on EF because they believe in it.  I haven't had anyone say "no" for anything I have asked for.  And I have asked for most of the resources in AMD that are appropriate for working on EF.  Considering EF hasn't earned the company one thin dime so far, AMD is being rather liberal with our spending millions of dollars on this.

We can sell every 5870 we make, EF or no EF.  If AMD killed EF tomorrow it wouldn't make a dent in sales at all.  We'd still sell every one we make as fast as we can make them.  So stockholder happiness has nothing to do with what we do or don't do.  At least in the short term.

I am interested in making a reasonable subset of customers happy, for a number of reasons, but where we are right now today is one step along a path that Eyefinity is inexorably moving along.  I am focused on that path.  If there is a feature you want in EF that isn't being delivered, there is a good chance it is because doing it would take us off the path I see in front of us.

Eyefinity for Everyone

I don't recall where I said it, whether here or somewhere else, but one of my Eyefinity goals was making it available for everyone.  AMD has launched the Radeon 5870 and Radeon 5850 by the time of this posting.  Those products don't address everyone.

Read between the lines.

Eyefinity & Display Resolution

In MS' operating systems Eyefinity currently has two exclusive operating modes: SLS and extended desktop.  You can be in one mode or the other, but not both at the same time.

In extended desktop mode Eyefinity works pretty much as dual-monitor systems work today: any combination of resolutions is allowed on your monitors, as long as you have sufficient timing (clock) sources to support them.  In a 5870 card, which has three timing sources, you can support three different monitors in extended desktop modes.  This is not true in SLS mode.

SLS, or Single Large Surface, requires that all the displays used in a display group have identical resolution.  You can have different timings, but you can't have different resolutions.  All monitors currently must have the same resolution and orientation when in the same display group.   Currently Eyefinity exposes a single SLS on the 5870.

Notice I said currently.   These operating conditions are the result of the current software implementation for performance reasons, architectural complexity, and hardware limitations.  A different software implementation could possibly change the preceding.  Is AMD working on that?  I can't say, since I am not acting as their spokesman in this blog.

How big is Eyefinity really?

How big Eyefinity is depends upon what you are asking.

I suppose most people are asking about physically displayable pixels.  Each display is limited to what DisplayPort (DP) supports.  A single DP connection supports a bit more than 8Gbps.  You can consume that data however you wish because DP transmits data in a format that the monitor tells it to.  The most common high end display configuration is likely to be either

• 2560H by 1600V at 30bpp and 60Hz, or
• 1920H by 1080V at 30bpp and 120Hz

Both of these fit within the somewhat more than 8Gbps capability of DP.  For all practical purposes current Eyefinity implementations are limited to a display resolution of 24.6Mpixels (6 * 2560 * 1600).

If you are willing to trade-off refresh rate and pixel depth, you can get larger resolutions, but you still have to live within the SLS (Single Large Surface) display limit of 8192H 8192V. When you add the total horizontal or vertical resolution of your monitor constellation together it has to be less than or equal to 8192 pixels or lines.  Extended desktop under Microsoft OS's doesn't have this limitation, but then again extended desktop doesn't have the particular capabilities that SLS does..

Bad, Bad Bezels

I suppose I shouldn't bother, but I will say it anyway:

You forget about bezels once you start playing a game that you like. 

Yes, if you don't like the game, or you are doing something non-immersive, those bezels in the middle of your field of view are distracting.

Those of you who have never actually used Eyefinity may dispute my claim.  You will forgive me if I dispute your claim, particularly since you haven't actually used Eyefinity with a game you like (like being defined as a game that is normally immersive for you).

I am not arguing the case that bezels are a good thing to have, quite the contrary.  I am making a point that they aren't as distracting as you might think.  That being said that original SunSp*t slide deck, you know the one I frequently refer to, did portray minimal bezels, near-zero bezels if you will, as being desirable.  Try it before you critique it.

More later...

Eyefinity for Gamers

As I mentioned in a previous posting, Eyefinity was originally conceived for gamers.  Sure business and other users will love it, perhaps more than gamers will, but my inspiration came from games.  The first SunSp*t (the original name) slide deck illustrated how it is possible to build large high resolution displays from inexpensive panels.  The specific example I used was six $149 22inch 1440H by 900V panels.  In other words 4320H 1800V and 61" diagonally for $900.  Wow!

At that time I didn't think a configuration other than 3W2H Landscape (hereinafter referred to as 3x2L) might be more desirable.  That thought came much later.

At a later point in time, that stimulating architect (see Why Eyefinity Supports Six Displays) was pressuring me up over mullion management (the technical term for managing bezel interference), particularly since in 3x2L the horizontal line could be very distracting.  My response had been, "well the ISVs will design around it once they see SunSp*t" (because after all it was so cool they had to, right?).  At the same time we were discussing how many business folks would use a 3x1P (3W 1H Portrait) "constellation" of displays, because many documents are historically taller than they are wide.  During that discussion it suddenly hit me that 3x1P would be the most preferred gaming constellation because:

• Three displays are cheaper than six
• HD aspect ratios are more closely preserved than with 3x2L
• and most importantly there is no bezel through the middle of a display

Yes we had discussed this earlier, but it didn't stick.  I am pretty sure I know why.  But late in the development of SunSp*t 3x1P caught on with a vengeance.  It may seem weird that I failed to apply my own maxim about odd dimensions in the vertical direction.  I applied it in the horizontal.  There was a reason for that though.

More...

Why Eyefinity Supports Six Displays

Anand of Anandtech.com got it right.  We had been developing GPUs with DisplayPort (DP) on them, and found that we never had enough DP outputs.  Because of the electrical characteristics of DP, we had to have one physical DP output from the GPU to every connector.  With analog signals such as those used in VGA, you can tie more than one connector to a single GPU output.  You can't do that with DP.  We had at least five DP outputs per chip, and sometimes six DP outputs.  Inevitably the question came up:  "Gee what if we used some of those extra DP outputs to drive more than two monitors."  (Note: one architect is the source of almost all of these stimulating questions.)

We had had two mobile customers asking for three display support for quite awhile. One had been asking back as far as 1998 (I know, because it was me they made the original request to).  At least once per year this request came up, and at least once per year I turned them down.  We didn't support their request because quite frankly the cost was exorbitant (those analog outputs take a lot of die area, which means the cost goes up).

As we refined the definition of the Evergreen products, this latent request was nagging at me, but not so much as the stimulating architect's question.  For various reasons one day I had the sudden impulse, the conviction to do more than two displays.  The reasons are the subject of another post, not this one.  The observation that DP costs less die area (think money) than analog does made it easier for me to make the leap.  When I said we were going to do this, the immediate conclusion was, "Great, we'll do four monitors".  I said, "no we will do six."  The question, "Why six?".  My reasoning was two-fold. 

• First, four was the obvious improvement over two displays (Note: Why?  it's that power of two thing).  If what we were doing leaked, the obvious conclusion would be that we were doing four.  I wanted to ensure that even if there was a leak, there would be little possibility that we would be competitively leapfrogged.

• Second, I thought there was no "constellation" where four made sense from a gaming perspective.  You always had what I called "the crosshair" problem" with 2Wx2L or 4W1H.  The observation is that for gaming the number of panels in any dimension needs to be odd to minimize bezel interference, to not have "the crosshair problem".

Why not more than six displays?  That's an easy answer: die area, package ball-outs, board real-estate.  In other words money.  But six displays does lead to some interesting discussions.  And the "crosshair problem" is less of a problem than you might think.  These are topics for other postings.

More...