Porting SW from SGI O2 to Linux

We have a medical software running on SGI, typically O2 workstation with OpenGL. Due to the eeeeextremely high cost, we want to see the possiblity of porting it to PC based hardware.

Could someone enlighten me on 1). speed and 2). graphics performeance on an AMD/Intel based system. If so is there any advantages for using ATI or NVIDIA cores? We do not want to switch at a later time, since all changes has to go through FDA clearance again.

Thanks for your kind help.

Could someone enlighten me on 1). speed

speed of what?

2). graphics performeance on an AMD/Intel based system.

Kind of all depends on a lot more than just the CPU you’re using.

If so is there any advantages for using ATI or NVIDIA cores?

I think nvidia is the clear winner between those two in terms of Linux support and GPU abilities.

We do not want to switch at a later time, since all changes has to go through FDA clearance again.

You know, if it is really really that important that you need FDA clearance, you should probably get slightly more professional advice than just what a bunch of yahoos on a public bulliten board say.

Originally posted by rts:
You know, if it is really really that important that you need FDA clearance, you should probably get slightly more professional advice than just what a bunch of yahoos on a public bulliten board say.

Thanks for your help.

The fact that we do not change the parts for FDA application again has nothing to do with professional services, IMHO. It was professional that adviced us on using SGI because of it is “the best” on graphics. But we just can not ignore the fact that PC is getting cheaper so fast and nobody knows how long SGI can last. I would stay simply the company that has longivity such as Microsoft for our other products.

The speed is referred to a graphic rerendering on Linux vs. O2. I am learning graphics computing slowing. One issue I think is how many overlays we are using now.

Thanks again.

The speed is referred to a graphic rerendering on Linux vs. O2.

Well, you’re comparing an operating system to a machine there, which doesn’t make a whole lot of sense.

With some high-end graphics board and a fast enough CPU, a GNU/Linux-based PC would probably give your O2 a run for its money.

Originally posted by rts:
[b] Well, you’re comparing an operating system to a machine there, which doesn’t make a whole lot of sense.

With some high-end graphics board and a fast enough CPU, a GNU/Linux-based PC would probably give your O2 a run for its money.

[/b]

Thanks, RTS.

Let’s suppose that there is a generic graphic-extensive program that can be run on both high-end-hardware-based linux and O2. How the performence compared in general? I know it is a very loose term.

May be the only way is to find out some benchmark test result which I do not know where to find.

My experiences:

I have both an O2 (one of the first R10000 150 Mhz, Irix 6.3) and an Intel/Nvidia combo (P3-933Mhz, GeForce2-GTS, Linux 2.4) Using OpenInventor on both. My apps run much faster on the Intel/Nvidia combo. I have also noticed that Performer demos are faster on the Intel/Nvidia combo. Depending on what your software requires and (how they implemented it)you may get different results, the two chip sets put different functions into hardware. My software pushes polygon count, not feature count. (no multi textures, 3d textures, volume rendering (I convert volumes to poly skins), etc).

Stability is another feature you should consider. The latest 2 nvidia drivers and latest XFree86 code have not crashed on me but it was not that long ago that I would often get into a state where I would have to restart the X11 processes to be able to use any 3d acceleration. (several glx calls would seg-fault). enjoy

I develop for both Linux and SGI, and have tested the same app (around a million polys, lit, smooth-shaded, no texture) on all of the machines in the office. The Linux boxes are mostly around 1gHz PIIIs with ELSA Gloria II video cards. The video performance is about the same on these machines as the O2s. Where the PIII really shines here is processor speed - disk accesses are about ten times faster on the Intel boxes, for example.

If all you care about is graphics, it’s a crap-shoot - you could go either way. If any other form of computing power is important to you, go with the Intel/AMD/Etc. line for sure. The cost doesn’t go down much since you have to spend a bit on a good graphics card, however. I picked up 3 02s on Ebay for about $600 a piece.

Hope this give you some insight into the decision.

Chris

Thanks to all the Linux gurus!

I am really happay to hear the PC based hardware can be faster than SGI from you guys. I have been puzzled that how can an O2 running at 150 MHz / 66 MHz RAM and be faster than a P4 1.7GH with 800Mhz RAM? My SGI sales person said SGI is simply faster no matter at what CPU speed.

Does it mean that the 64 bit intel CPU is going to be major blowup for the ultra-high priced unix computer?

Thanks again for your knowledge.

Michael

I do not think new processors will make a big difference. The SGI machines is for graphics. It still exists some reasons for SGI:

  • Some stuff is not hardware accelerated on consumer cards but on SGI.
  • programs
  • It seems like a lot of people seems to prefer SGI

mhu8,
Speedwise, there is only one area that the O2 should outperform a high-end PC: work with real-time video (particularly with MJPEG format) and real-time editing that includes loads of textures (enough to swamp the the texture ram on the PC, but not the O2’s unified memory) and perhaps a few other tricks (rotating cube with video streams to each face). The O2 is rather old now, and the graphics never were spectacular for pure OpenGL acceleration. The SGI Octane2 could easily outperform PC solutions for some tasks, but all… and it’s way expensive. If your “medical software” involves combinations of heavy texturing, video, and real-time, then you may still want to stick with an O2 (I’m thinking rotating MRI images). Your SGI salesperson may just want to gouge you, but it’s odd that (s)he didn’t suggest one of SGI’s x86/Linux computers. For strict OpenGL performance, their PCs would be more performant.

We are working with acquired CT/MRI data, but not real time. CT image is typical 8 bit deep and 512x512 each slice. We usually deal with 20 to 30 slices at a time for treat cancer patient. Thus we need to reconstruct all these CT images into a 3D representation. The second reason, besides high cost, is we do not know how long SGI will be around. Therefore we just want to get out alltogether.

Do you know some OpenGL graphic cards that do the following:

(1) has alpha blending for transparent surfaces.
(2) has 3d texture mapping.
(3) has overlay planes for user interface.
(4) has 12 plane visual.

Thanks again for your enlightenment.

Michael

This topic was automatically closed 183 days after the last reply. New replies are no longer allowed.