SGI: Hardware

Hi all,

I have questions about the two "Big Iron" graphics options that SGI had available for Onyx systems back in the day. I've read a lot about the two different systems and what they technically are capable of, but I still have some questions. All these questions relate to a hypothetical single system with a single monitor.

1) If I had a single rack with an Onyx 3xx(x) based system (not onyx2), would I see a greater performance benefit from the graphics end by using IR graphics, or IP graphics? I'm not sure how many nodes of IP could be linked together in this setup, or how many IR pipes could be used in this setup either.
2) For IP, how many pipes/nodes could be linked together to create a single system? Is it as expandable as the IR graphics?
3) For IP, does having multiple pipes/nodes linked together allow for a faster graphics system, or does it only allow more monitors, but the same as a single pipe (like a stand alone V12).
4) Is it possible to use any of the Origin 3xxx based systems and add in graphics? (either IR, or IP)
5) Is it possible for a system - like an Octane, Fuel, Tezro - to connect to an IR or IP pipe and use that instead of the built in graphics?
6) I know that the IR is capable of 8x antialiasing, but is that feature only enabled if the software is calling for it? I.E. in Maya, there is no way to force the graphics system to antialias the viewport, but would an IR system draw the graphics with antialiasing by default?

I think those are all the questions, but I'm sure I'm forgetting some. I've always been very curious about the IR and IP graphics and how they would perform in Maya or Alias Studio. I like using my Octane for these apps with the V12, but I'm always dreaming of more....

Thanks!

To answer (1), InfiniteReality. InfiniteReality is InfiniteReality. InfinitePerformance is Octane2 graphics in an onyx box called a V-brick.

IP: "this product provides an economical advanced visualization solution for polygon-intensive applications."

IR: "The ultimate solution for the most challenging demands"

http://www.sgi.com/pdfs/3468.pdf


Depends on the size of the system 2-way O300 vs. 128 way O3000 etc.. You can combine IIRC up to 4 IP pipes (2 vbricks, 2 pipes per brick) with a special compositor box. Recondas has/had one.

for the most part, yes.
No.
IME most applications that support it will do it automatically but I would have to fire up Maya and check it out for myself.

What the guardian said, though I'll mention the V-Brick ( or hacked X-brick ) was intended for connection with an O3000-based system (you should be able to hack a V-or-X-brick to work with the O3x0 too). An O350 can also be configured as an Onyx InfinitePerformance system , in that case the V12 is installed inside the O350 compute module, and the V12 is the later generation XIO2/IP35-Fuel-Tezro type.

Chris, if it might be helpful, there are some test results that compare the V12 to various IR configurations in this thread (a single IR3 graphics pipe with four RM10-256 raster/texture boards loooks pretty strong when compared to a single V12 pipe): viewtopic.php?f=3&t=16722836&start=15#p7319524

The trouble with the Compositor box - for me, anyhow - was that it's a deep, rather than wide thing. The output resolution is limited to something like 1920 x 1200 or maybe even less. What you get is more memory depth and a bunch of ways to combine the incoming displays into a single output. Pretty cool but sadly, you're still stuck at a low-res output

I still have the Compositor, and while it may work the opposite of hamei's hopes to run multiple V12s as one much larger 3840x2400 display, it may come much closer to what Chris is looking for:

An InfinitePerformance-Compsitor combination may also prove helpful in regards to Chris' inquiry about anti-aliasing:
http://techpubs.sgi.com/library/tpl/cgi ... /ch01.html

I don't have any experience as to how (or if) an IP-Compositor system will work with Maya, so you may want to do additional research before deciding on either IR or IP.

Thanks for all the info!
The link Recondas sent comparing the performance of a single pipe v12 to a IR3 system was interesting to see. I wonder how a 4 pipe IP system with the compositor would compare to an IR3 or IR4. I'm wondering about all this in a single rack, not multiple racks. I'm just trying to figure out the max performance for a single monitor output system. I'm wondering if having the 4x IP pipes through a compositor would be 4x the performance of a single v12. If that's the case, than it would be double the performance (or more)of a IR3 system with 2 RM10's. It looks as if it would also inherently enable antialiasing in the software.
Anybody have a system similarly build that could be benchmarked? I see that Recondas has the compositor and 2 pipes. Have you benchmarked that system vs an IR system?

Is there any difference between an Onyx 350 (with v12) and Tezro rackmount besides the faceplate?

Nope. The two are one and the same.

I see so you could technically numalink a Tezro and Onyx350(v12) and end up with a dual V12 with either 2 or 4 DVI out? This would be clearly more pleasant than a g-brick itself.

You may need to change the brick type of the Tezro - I'm not sure if the PROM checks the type to determine if it will link I know you can't use a router with "Tezro" bricks.

You could also NUMAlink two Onyx350s.

If two Onyx 350's are NUMAlinked, are they treated as a single, large system with the combined ram, CPU and graphics? If that's the case, do the graphics get their combined power to run a single display, or is each graphic pipe treated separately? Also, how would this type of a system integrate into the SGI Compositor hardware?

IRIX will present you CPU and RAM as if it was in a single system, but graphics will be treated separately, you have to combine it yourself.

Here are two guides which might help you:
SGI OpenGL Multipipe SDK User's Guide
OpenGL Performer Programmer's Guide


Here's the description of the Compositor:
The older one (VPro only - Onyx 350, Onyx 3000): SGI InfinitePerformance: Scalable Graphics Compositor Owner's Guide ( HTML version - techpubs have some problems with it)
The newer one (works with VPro and SG2 - Onyx 350, Onyx 3000, Onyx4 and Prism): SGI InfinitePerformance: Scalable Graphics Compositor rev. B User's Guide ( HTML version - techpubs have some problems with it)

Yes. You can numalink two O350's with two graphics cards and two dcd's to get four dvi graphics outputs. SGI did this.

There is another option ... four O350's with a router and a V10 or V12 in each.


Different animal. Some people like camels, some people like donkeys, some people like elephants. One or two like horses

In the case of the T221, the monitor will do it for you. Or were you speaking of laying the pipes/channels down side-by-side to create one large desktop ? OpenGL Multipipe (in theory. Have not yet tried it.)

But mia ... I don't think that's what you really want. It's a huge amount of work just to get more terminals on-screen ...

IP is often used as an easy way to get a head on an O3x series (similar to CHALLENGE EXtreme). IR is probably the easy way to get big SGI graphics performance on a server, and it offers pretty much everything that SGI ever offered graphics-wise. IP looks to be a bit more work to get running.

If you really want a challenge you can always try UltimateVision.

Recondas is one of a very few around here who can give a real-world comparison between the two ... but I looked into it quite a bit for a while. It seemed as if Infinite Reality would give you some things that Infinite Performance does not -- if you are writing your own software. If you are just using off-the-shelf commercial software, there's not much point. Or if you are not doing the things IR does well, there's not much point.

This entire question is sort of backwards. The correct way would be, "I want to do X, Y, and Z. Which graphics option is better at doing what I want ?"

And availability of parts... and running costs.....

IR graphics is easy to get up and running and the hardware typical of a lot of SGI older generation stuff is pretty reliable. But it sucks up a lot of power whilst operating and it's not cutting edge - simple things like the standard DG5-2 board's have a 13w3 analogue connection - where as newer Vpro IP stuff is all DVI.

But then unless you go down the route of V12 cards from Fuel's hacked into the O350 chassis, genuine SGI kit for IP graphics is fairly hard to come by where as there is an abundance of Onyx 3000 type machines around.

I have various IR graphics and IP graphics systems here but no compositors I'm afraid. I could very easily run benchmark of V12 versus IR2 / IR3 or IR4 though if it helps you. I have an Onyx 3000 config that I can attach either IR or IP graphics to easily enough, so the CPU / RAM / Disks would be identical.


tjsgifan

I guess the real question is this: If I want the most powerful graphics performance in a single rack system (think of it like the ultimate SGI workstation) to run off the shelf software - Maya, Alias, etc... - would that be a system based around the IP, or IR? I would imagine that having the IP system would also allow for more processors in the single rack as well as having some more advanced OpenGL features like hardware accelerated phong shading in Alias. Not to mention the reduction in power requirements.

I just came across an input device that might be suitable :

Onyx 3900, 32CPU, 64Gb RAM, (2 x 16 CPU Cx Bricks), 1 x I brick for I/O and 1 G-brick with a single 4 RM11 pipe.

That does sound like a sweet system. Also sounds like a very expensive system. The RM 11's alone are hard to come by and expensive, let alone the 2 16 CPU Cx bricks.