[Future Technology Research Index] [SGI Tech/Advice Index] [Nintendo64 Tech Info Index]


[WhatsNew] [P.I.] [Indigo] [Indy] [O2] [Indigo2] [Crimson] [Challenge] [Onyx] [Octane] [Origin] [Onyx2]

Ian's SGI Depot: FOR SALE! SGI Systems, Parts, Spares and Upgrades

(check my current auctions!)

Comparison of Octane MXE to Deskside Onyx2

[Octane] For some problems, the question may arise as to whether a high-end Octane or low-end Onyx2 is the best choice for a project. Due to the nature of real-world tasks, the answer often depends on various technical aspects of how the target problem needs to be solved.

In general, if you're faced with this choice, I would say go with Onyx2, ie. if you're considering the very top end of an Octane's current capabilities, then there is a danger that you will have no room to grow your computational resources should project requirements exceed original expectations - a common occurence I expect. Think of this as rather like buying a hifi speaker system: if one finds oneself contemplating the very top-end of a particular model's capabilities, then one should get the next model up that can handle the task easily, rather than at the peak of its capabilities.

[Octane] This page gives some detailed examples of why Onyx2 is the better choice when faced with such a decision in the world of 3D real-time graphics, especially VR. Obviously, there may be budget limitations which leave one with no choice, but my comments here are targeted more at customers who tend to have higher budgets: aerospace, automotive, government, medical, oil industry, etc. Such industries would prefer to go with cheaper systems whenever possible, but they're much more likely to select a faster system if they believe there is sufficient justification for their purchase or lease. When it comes to what I call 'industrial' quality projects, it is unlikely that managers would be so short-sighted as to limit the scope of a research program by under-equipping the research team. The team's focus should be on the problem, not on getting round the limitations of their equipment.

Many of the comments I give below are based on an email I received from someone working on an automotive VR research project who had exactly the above conumdrum to solve. First some summary tables to convey the basic differences between Octane-MXE and low-end Onyx2 as they are right now (this data is extracted from my SGI Graphics Performance Comparison Tables page; see that page for a 'key', eg. 'R' = Reality, as in Onyx2 Reality):

                           3D        Flat     Lit GZ    Lit        Lit  
                    X      Depth     Tri      Tri/sec   Tex        GZ   
                  Lines    Cued GZ   Strips             GZ         Indep
                           Lines     No Z               Tri/sec    Quads

Octane MXE:       3.88M    2.29M     ?        2.19M     1.42M      880K 
Onyx2 R:          ?        ?         ?        ?         5.5M       ?
Onyx2 IR (9):     ?        ?         ?        ?         11.0M      ?
Onyx2 IR2 (9):    ?        ?         ?        ?         13.1M      ?


                  GF        Tex GFZ   Conv(3x3)   Conv(5x5)   Color
                  Pix/sec   pix/sec                           Lookups/sec

Octane MXE:       240M      138M      33.7M       27.12M      96.4M
Onyx2 Reality:    448M      188M      ?           30M         ?
Onyx2 IR (9):     896M      624M      ?           60M         ?
Onyx2 IR2 (9):    896M      768M      ?           60M         ?


                  Data       CDRS       Design     Advanced      Lightscape
                  Explorer              Review     Visualiser

OCT MXE/250:      18.37      55.46      10.62      20.09         1.55
Onyx2 R/180:      21.72      91.54      15.83      23.70         2.41
Onyx2 R/195:      21.90      92.72      16.27      23.84         2.61
Onyx2 IR/195:     35.65      161.29     20.82      45.43         2.65
Onyx2 IR/250:     41.60      ?          23.96      ?             3.30


                  Characters    Screen     Max Framebuffer    3D Vectors
                  Per Second    Clear      Memory Size        Per Second


Octane MXE:       ?             ?          27MB               ?
Onyx2 R:          ?             ?          80MB               ?
Onyx2 IR:         ?             ?          320MB              ?


NB: Reality series graphics for Onyx2 is being phased out in favour of InfiniteReality. This is probably for several reasons:

I said earlier that the above figures refer to the situation for Octane and Onyx2 right now. When both systems get their new graphics options, I expect Octane's new graphics system to be as fast as InfiniteReality (or better) and, as usual, Onyx2's new graphics to be an order of magnitude faster than that. Over the years, SGI has, in general, always maintained an order of magnitude performance difference between each level of its product line (low-end, mid-range and high-end).

Please don't ask me when new products will be released as I honestly have no idea. One can merely judge from common sense that Octane's current graphics technology, which is based the IMPACT technology from the old Indigo2 days, is showing its age, while InfiniteReality in its basic form is more than 2 years old now. With the release of the Visual Workstation, I find it highly unlikely that the current options for Octane and Onyx2 would be left alone for much more than a month or two.

So, why is Onyx2 better than Octane for VR projects, etc.? Here are some reasons:

There are several issues associated with requiring multiple high-quality display outputs:

  1. One needs lots of frame buffer video memory to support the multiple outputs.

  2. If one require high-quality full-scene multispatial anti-aliasing (definitely needed for good quality VR), then lots of video memory is required (8 times the normal amount).

  3. In order to sustain high frame rates at high-resolution, one needs high pixel fill rates. This is why something like an Intergaph Wildcat is useless for VR applications: 90M textured pixels/second is nowhere near enough to handle complex scenes at 1280x1024 60fps. Besides, Wildcat does not offer quad-buffered 48bit RGBA anyway. InfiniteReality offers 800M pixels/second, as well as features such as dynamic resolution rendering to ensure consistent update rates.

    For VR work, high frame rates are critical; those involved with video conferencing technologies have found a negative effect if the frame rate of a video link is too low - in the same way, a VR display will not be able to offer good insight into solving a problem if the display update rate is low (the level of interaction and cognitive would be insufficient to gain any further understanding of the problem).

  4. If there are to be multiple viewpoints, then that will require multiple preprocessing for the graphics pipe, ie. being able to assign 1 CPU per viewpoint ensures efficient usage of available host compute power.

Many of these factors are related, eg. multiple displays, high quality anti-aliasing and high update rates all imply a need for high pixel fill rates.

In the above example, where at least 3 displays are required (more likely 4 since an instructor display is also usually needed), again Onyx2 is the choice since an option card offers up to 8 display outputs whereas I believe Octane is limited to 2.

I know of a researcher at a medical institution in Alaska who was organising an upgrade to their SGI system. They were considering an Onyx2 Reality as an upgrade over their old IRIS Crimson. After some consideration and demonstrations, they eventually chose InfiniteReality instead (2X faster than Reality). A wise move.


Final notes:

A video clip I have shows Ford Motor Company's Onyx IR system doing real-time VR-based car design, using six IR subsystems. Asking for demonstrations also gives you a chance to talk to SGI people who know more about the subject and can offer detailed advice, and gives an opportunity to enquire about other costs such as software, peripherals, etc. (try out different HMDs before you buy/lease! Maybe you don't need a 1280x1024 HMD in order to obtain the quality level required for the project? Maybe 1024x768 is enough? Or maybe you need better than SXGA, ie. perhaps 1600x1200?)

Your focus should be on solving the problem, whatever that may be: research program, specific task, etc. Don't obtain a system that can't help you solve your problem.


Ian's SGI Depot: FOR SALE! SGI Systems, Parts, Spares and Upgrades

(check my current auctions!)
[WhatsNew] [P.I.] [Indigo] [Indy] [O2] [Indigo2] [Crimson] [Challenge] [Onyx] [Octane] [Origin] [Onyx2]
[Future Technology Research Index] [SGI Tech/Advice Index] [Nintendo64 Tech Info Index]