Flame on Linux

[hyrlvlrec]

so when the flame console says “700mb allocated to framebuffer” thats what its doing?

[patdawg]

hyrlvlrec wrote:

so when the flame console says “700mb allocated to framebuffer” thats what its doing?

Exactly…it is allocating 700MB of your SGI’s system memory to be used for the framebuffer. Note that you can change that number up or down by modifying the memory keyword in your init.cfg file. Once we get 64-bit memory addressing in the next year we’ll be able to add gobs of memory, and crank that memory keyword up to ridiculous values.

[Anonymous]

😕
im a newbie here, but smothing is confusing me, lets suppose i have an octane2 filled up with 8gb of ram, then, why should I use a Vpro V12? a V10 can be enough, even a V6, (in case you dont use dual display of v12 for dm5 of course 🙂 )

so lets buy cheap ebay ram? 🙄

offtopic: where should i start to learn flame? books? dvds?

cheers

[-k]

patdawg wrote:

hyrlvlrec wrote:

so when the flame console says “700mb allocated to framebuffer” thats what its doing?

Exactly…it is allocating 700MB of your SGI’s system memory to be used for the framebuffer. Note that you can change that number up or down by modifying the memory keyword in your init.cfg file. Once we get 64-bit memory addressing in the next year we’ll be able to add gobs of memory, and crank that memory keyword up to ridiculous values.

I’m not an expert but this hype sounds a bit too easy to me.

1st) Someone posted a test on flame-news:

—
I made some benchmarks in the last week for texture fill rate only. And I want to post my results to here. This tells both texture fill + accumulation buffer performance of the gfx subsystem. The cpu speed is almost unimportant.

The test is like this, just get a 2K_8bit image into action, add an image, and scale it down to 50, and render 100 frames of PAL. But set the antialiasing to 16 samples, and motion blur to 50 samples.

1. inferno_onyx350_ir4(single RM11-1024): ~4min.
2. inferno_onyx2_ir2(dual RM9-64): ~4min.
3. flint_fx300g(dual 3Ghz xeon): ~18min.
4. tezro_V12(dual 700Mhz): ~32min.

—-

Tezro does not look that good (in this case)….!

2nd) I thought the whole point of the AGP bus is that you could use system memory on PCs. Nobody did it of course but should it not be possible?

3rd) Just because you can use system memory does not tell you how fast this solution is.

4th) Even if so. It might be fun to have 100 layers in Action, DVS but depending on what you do you might only have a few layers but hundreds of blurs , keys, edge detects, color corrects, logic ops, gmasks etc. (that’s at least the case for me far more often than your scenario). Which probably benefits much more from cpu speed

5) Honestly I think it is a JOKE that FFI has not been a 64 Bit application since years. Now they try to sell you this for lots of $€$€. A joke really…

-k

[Anonymous]

test

[loops]

Some clarification if anyone cares:

No SGI system* can use system memory directly for textures. They all have seperate, dedicated memory. Octane MXE only had 4Mbytes (sic), Octane2 and Tezro have about 40 for V6 and V10, about 100 for V8 and V12. RealityEngine and InfiniteReality have shitloads, I’m not so familiar with them. Gigs anyway.

The memory line in the config file sets how much system RAM is used to buffer frames, both from the framestore for playback and as scratch space when scrubbing though clips before they’re processed and various other shit.

One of the reasons SGI systems kick shit out of PCs is their fat busses. An Octane has a 1.6Gbytes/sec connection between the graphics and the front side. Tezro and Onyx have at least double that. This means that you don’t need to be quite so worried about running out of texture RAM – frames can be brought in from main memory hella quickly. This was obviously very much the case on MXE graphics, where if you were picky about your textures staying in the TRAMs you’d only be able to have like 3 PAL layers in Action.

When AGP was launched they did say that cards could just use system RAM for textures, but very few ever did. Someone has recently revived the idea with PCI-Express, I forget who. PCI-Express is starting to get close to SGI-style fatness, but for now the crossbars in Onxyes still rule, particularly since they connect everything in the system at that speed wheras PCI-E is only a single pipe. Plus the crossbars have garunteed low latency, proper CRC error correction, and QoS-like reservable bandwidth.

It would be nice to think Discreet (or “Autodesk Media and Entertainment”, eew) will go to SGI Prism, which has double the crossbar bandwidth of Onyx and potentially some extremely fat graphics…

* Apart from O2 which we’ll discount becuase it was shit. You could load a half-gig image and rotate it around no problems, but the fill rate was only 65 million textured pixels per second absolute tops so it wasn’t much use for Action. Trust me on this, I still run effect* 6 on one :-/

[loops]

BTW Tezro does badly in the that test above because it doesn’t have hardware multi-sampling whereas IR and Quadro both do.

[Anonymous]

I had access to inside info so:

1. Flame will move to linux. Tezro is the last machine from SGI, they abandoned MIPS line and IRIX development.
2. I’ve seen internal benchmarks on both platforms: say goodbye to IRIX.
3. I tested for one week a flint/ibm vs. our flame/octane. Flame’s speed was pathetic, especially on sparks (cpu intensive).

[eltopo]

Kuky wrote:

I had access to inside info so:

1. Flame will move to linux. Tezro is the last machine from SGI, they abandoned MIPS line and IRIX development.
2. I’ve seen internal benchmarks on both platforms: say goodbye to IRIX.
3. I tested for one week a flint/ibm vs. our flame/octane. Flame’s speed was pathetic, especially on sparks (cpu intensive).

… and Lustre is moving to mac or so ThinkSecret.com says…

I guess this upcoming NAB will sort out the future… 😉

[paul_round]

So what happens to Inferno?

[malu05_vb]

Inferno is going to run on Playstation 3!… 😆

[opus13]

paul_round wrote:

So what happens to Inferno?

i heard some rumours from a reasonably reliable source that inferno is going to move to a heavy duty itanium based platform also sourced by sgi.
considering itaniums questionable memory architecture, i really dont think that would be the wisest of ideas. its got FPU out the ass, but it lacks otherwise.

[Anonymous]

malu05 wrote:

Inferno is going to run on Playstation 3!… 😆

And it will have direct client connection on PSP pocket version 🙂

[Author]

Posts