ok so im trying to decide if i should get a dual amdxp 1.8 with 4 gigs or ddr ram and and oxygen with a gvx 420 for some where around 12k or get a fully loaded octane 2 for 72k
is the octane2 really worth the price (or is discreet flame now that combustion 2 is out)??? plz help trying to figure this whole thing out
thanks in advance
GHSU3D
Ace's compared Maya with SGI Onyx2 Infinite Reality, Dual AMD, Dual Intel:
This graph is a testimony of the extremely high pace at which x86 performance has improved. The result is that the calculation power of a SGI Onyx2 Infinite Reality system, a workstation that costs $87000 (!!) in 1999, is now eclipsed by a +/- $3000 dual Athlon system.
http://www.aceshardware.com/Spades/read.ph...cle\_id=30000190
A more recent comparison with MP 1800+:
sgi=overpriced relic
unless you need to have a room full of rendering machines
your better off buying 19 amd machines for the money
What?!.......
No way in hell should an x86 system cost 12k in this day and age. That's just way too much. Also I keep hearing that the gvx 420 is real slow...
As for the Octane2, do you really have that much money to throw around? If money was no objective I'd take the Octane2, but damn, $72,000 is a lot of dough for a desktop computer.
How can you get a flame/Octane2 for $72K?
If you can, you'd be a FOOL not to take it.
A new flame usually runs around $275K, and supports uncompressed SDI and HD as a base config. It offers full accelerated 3D space and the most powerful compositing tools in the world. period. There is nothing available on Intel, AMD or MAC today that offers (close to) the same level of realtime creative flexibility for finishing highend projects. ......combustion2 is not flame.
Now, for straight 3D and rendering - these's no doubt that an AMD or Intel/W2K system will beat the pants off SGI in terms of price/performance.
kiwi
Having worked with Inferno/Flame for over 7 years, I can tell you that given $75K I would build a killer C*/Edit*/Max*/T3K/RenderFarm for the same money. As far as I know, the Flame/Octane does not come standard with HD for one thing. For another, the C* system would outrender it, especially with a renderfarm. The integration between edit*/combustion beats anything Flame/Fire can do, and the integration between Max/Combustion beats anything FFI can do. The workflow in C* destroys Inferno. When I first evaluated it, I did a little side-by-side to list the steps I took in Flame vs how I could have done it in C*. In all cases, even if C* suffers a little in terms of interactivity, it's worklfow advantages far far outweigh this. The only tools that Inferno has that C* does not are bicubics, extended bilinear surfaces, import 3D objects and a built-in warper/morpher, and if you care to include it which I don't, the Modular Keyer (which forces you to work with the 3D keyer, a joke IMHO.). Flame has even less featuers, since Discreet decided to differentiate products and trickle-down features from Inferno to Flame to Flint in each release. In terms of 3D import, I believe TILT can emulate this, in terms of "integrated" warper IMHO Flame's sucks (outside of it's interactivity, which is a microcosm of the whole Flame problem, fast interaction, poorly implemented and integrated toolset) and as far as I know RE:vision will be porting it's integrated warper to C* soon and if not, I would just as soon work with Elastic Reality open since it is as about integrated to C* as Flame's warper is to Flame's compositing environment. This leaves mesh/bicubics and I believe it should not be that long before this is possible with plug-ins. For $30K I just put together what I think kills anything I have ever worked on on "high-end" boxes and I will live with the few shortcomings to relieve myself of the many shortcomings of the much more expensive Flame solutions. Which, are not real-time btw.
Just one man's opinion.
Flate,
You make some good valid points, but there are some points that I should clear up......
If the unit is an Octane2 with the DM2 I/O (standard for flame/smoke), then it IS HD-ready, assuming that you have 4 channels of fibre stone storage.
If you are doing HD work, then there is no comparison between discreet/SGI and combustion on MAC/Windows. It blows it away. (always in interactivity and also in rendering speed if a render farm is not being utilized.)
For SD, the line really starts to blur a bit.
Your notion that the workflow in c* "destroys" that of flame/inferno is a bit over-stated. For instance, c* is just now getting it's first taste of node-based or "schematic" workflow. This functionality is just not as robust as what is already a very mature workflow in F/F/I.
Anyhow, in most cases, the workflow for a given "operator" in combustion is pretty close to the same in FFI. In other cases, the workflow is faster and more flexible in F/F/I.
Can you give an example of anything that is faster in combustion?
There is a nice integration between edit* and combustion*, certainly. But I will stand on all of my reputation when I say that discreet smoke already has a better integration of more powerful editing and compositing tools than edit/combustion will ever have in one box.
3D in combustion is a fake. You can only see a 2D representation of the 3D object. You can't deform, texture map, re-light or rotate around the object like you can in F/F/I. -No mesh support whatsoever. While c* may well get this "true" 3D environment in time, it just doesn't compare well against F/F/I.
Morphing/warping, camera matchmove extraction, and a bunch of other neato stuff is available from other 3rd party vendors -and I would guess that some of these technologies may very well make their way into the discreet FX line in short order.
To give credit to your arguements, though, combustion does have all the benifits of the fantastic discreet keyer (screw the modular 3D keyer in flame), color corecter and tracker as F/F/I, -and they freeking rock anything else on Windows or Mac. Add a render farm, and you have a very powerful FX box.
If you have the clients, you can charge $350 - $600 per Hr. with flame. What can you charge your client for combustion on a Mac?
My arguement is mostly that for client-driven creative sessions -or for heavy high-end 3D-heavy comping, combustion has a long way to go before it approaches the interactivity and speed of it's big brothers.
On the other hand, for doing FX compositing on the desktop, I'd put my money on combustion before any of the others (except maybe shake)
***the rumour is that discreet is busy working on their "black ops" project called "toxic". Toxic (codename)will be the replacement for edit, and is rumored to be an AI64/Linux box with integrated 3D just like the discreet advanced systems. There are no details on pricing or any other capabilities. discreet has not publicly disclosed the exisitence of this project, but our sources are "very reliable" and indirectly involved in this effort.
kiwi
I have a whole suite of tests to compare rendering, hardware assisted/software rendered, interactivity etc. I imagine you are right about HD speed but that remains to be seen. And, the rendering farm is a significant point, because I bet dollars to doughnuts that a DualAMD1900+ or DualP4Northwood with a 10-Proc Renderfarm would outrender anything that Onyx puts out. If you look at pure render power of these boxes, the AMDs and P4s kick the MIPS chips. And I do understand that the Flame does hardware assisted render, however my tests show that this is about a 3x speed increase over Texture Off. So in essense when you compare Inferno to Combustion, you are comparing a "renderfarm" to a single-unit and the single-unit still holds it's own. I have not even tested C* with OpenGL render on, that is one of my intended tests. I will let you know, I am pretty sure even the latest Infinite Reality and top MIPS chips would not blow a way a well configured NT with even a small render-farm.
Interactivity is definitley an issue. This is due more to bandwith and memory/cpu/peripheral archtiecture. I think companies like ServerWorks are quickly addressing those problems. I read benchmark tests with the serverworks with PIIIs and the leading Intel chipsets. They performed a suite of Photoshop tests, and the SW came in about 15% faster. But, if you read closely, you will see that the times of all the tests were merely added together for a composite score. Most tests took 4-20 seconds. One test took over a thousand and this is the one that had very little difference between the chipsets. If you isolate Scale/Rotate/Blur type ops, the SW was often twice as fast, same CPUS, RAM , etc. This is what excited me, since most of these benchmark tests are for rendering, and those operations are the precise ones we need for 2D interactivity.
So, imagine what that chipset would look like with dual 2.5Ghz P4s with PCI-X [5x faster than 64/66 PCI]. Well, you don't have to, it is on it's way Q1.
So now we have far faster CPUs coupled with far faster OpenGL coupled with internal bandwith that at the very least rivals SGI.
There is so much more to workflow than the Schematic View. And, just as an aside, these are things that make FFI great in my opinion;
Integrated Keyer/GMask/Tracker
Schematic/Nodes (though terribly poorly implemented, don't get me started)
3D Space
BiCubics and Extended BiCubics
Integarted 3D
Hardly. Let's talk about the "Modular" Keyer. It is fairly useless because the 3D Keyer sucks even if it's algorithm is better which I do not believe. Combustion can easily Modular Key any keyer, including the Discreet Keyer or any third-party keyer. Modular Keyer is simply a mini-batch for a set of tools Discreet has decided make sense together (which they do).
Operators: This is where the whole notion of "Schematics" drives me crazy. Having the option for process trees is great. Having it as a kludge because the compositing environment (Action) is hobbled by being tied to the Graphics Engine is not. In order to apply a series of filters to a series of layers, one must set up this stupidly complicated Batch Set up, with filters feeding nodes which feed Action inputs. To see them in-context, one must set up Context Nodes. What if you have ten layers with ten filters stacked on each? Can you see the result of each filter in any stage of the pipeline or in-context to the whole composite. Sure, if you set in-context nodes, which are limited as far as I remember to two. The interactivity overhead of this is HUGE. I did a job where I was simply feeding a couple of glow/edge filters into a layer and then into Action. How do you set this up? Well, you must enter batch, load your clips, load Filter nodes, manually attatch them, manually create Action input-nodes, manually attatch your layer/filter pipelines to these, enter action, apply layers, exit action, set in-context nodes for the two nodes you find most important, select those nodes, and wait forever for updates because the overhead of "Batch" in this case is ridiculous. What if I want to see the result of the filter in-context to the composite AND at the node where I am making changes? No such luck. II waited as much as 15 seconds for updates on a 4xR12k IR2. Now, how did I do that in C*? Import Layers. Attatch Operators. Modify filters. No node adding, drag-connection, entering/exiting modules/assigning in-context nodes. When I want to make a change to a filter, I just choose the filter and make changes. Updates are extremely fast because there is not some crazy kludged process tree overhead. I can set up one viewport to see the final-composite, and one viewport to see the result of the operation on the layer, and one viewport to see the operation of the filter plus the downstream filter with about three clicks of the mouse. What if I want to view the composite without the filter? In Inferno, I need to exit action, select the node and disconnect it. In C* I simply click on the operator.
So you see, even if Inferno IS faster in terms of absolute update interactivity (which it is not in this case) I would much rather sit with a client who says "hey can you add/change this filter, hey can I see it without the glow, hey can I see them side-by-side" in Combustion any day of the week. That IMHO is the reality of client-driven sessions, not real-time warper updates.
Okay, more? I am tracking a layer in Action. It is hard to track. I need to apply a filter. I 1) exit Action 2) enter Filter 3) select the sourceclip from the desktop 4) load the filter 5) set parameters of filter 6) process 7) exit back to desktop 8) re-enter Action 9) select my original source and load the filtered version instead 10) perform my tracking 11) I need more filter 12) exit action 13) enter Filter 14) select the sourceclip from the desktop 15) load the filter 16) set parameters of filter 17) process 18) exit back to desktop 19) re-enter Action 20) select my filtered source and load the more-filtered version instead 21) perform my tracking 22) select my filtered source and load the origina source instead.
In Combustion, I would 1) apply a filter operator 2) track 3) I need more filter 4) change the filter settings 5) track 6) turn of or remove the filter. Again, I could not care less whether Inferno tracked faster which it does not, because I have a client sitting behind me while I jump through hoops to do the simplest of operations.
I won't even go into Source nodes, suffice it to say I have tested out how to do separate matte, separate fill, replace matte, replace fill and it is a few step operation in C*, whereas in Inferno I feel like I am doing visual programming.
Paint is vector making rotoscoping much easier, since you can modify the stroke parameters and shape anytime you want AND paint is in-context to the composite. Can you imagine the difference is roto-ing or painting a layer in-context to composite???
Text (even in 1.x) is soooo much easier. What if you are doing an animation to each letter in a word? You don't have to worry about creating parent nodes, creating letter nodes, naming them all, offsetting them all, etc. It is done for you. Select. Animate. Done.
Tranfer modes are in-context to the composite. Yes, I know their are "Sparks" for that, but that puts you in stupid batch-land again and then you only have the tools the spark developer gave you. Often transfer modes need blurs to make them work well. You can't do that with Sparks since blur is aplied downstream after the transfer mode calculation.
Tracker had auto-snap which combines the Tolerance and Snap features of the Inferno tracker (don't remember seeing this in FFI, because I would have used it alot!)
You can import Max cameras without jumping through hoops.
You can apply a filter to a portion of ONE Layer, instead of having to create an additional copy of that layer with it's own matte. This saves creation and rendering time. You can modify the selections that define where operations take place easily.
I could go on and on. I took several pages of notes on the advantages of C*, and just a few line items for FFI. I did a side by side for the last 4-5 jobs I did on FFI to see how I would have done them in C* and was amazed.
It reminds me of a Quantel demo where I asked if he could take one layer he was showing us and parent a bunch of other layers around it and animate them all (i.e. solar system). He said, well no, but you could animate the first, then the others, etc. and since we render so fast it would be just as fast. Wrong. When you lose in-context workflow, speed does not buy it back. And I think FFI has such limited in-context capability (even the keyer in Action cannot be keyframed in-context to the composite keyframes) and it's purported speed advantages are far less than people believe.
Well, you can work with rendered 3D scenes FAR more effectively than you can in FFI. And, as I say if you cannot use third-party plug-ins like Tilt now, you will be able to soon. I pesonally would rather "kludge" together Max and C* for now. I don't love using 3D objects in FFI anyway since the lighting is way limited comparative to 3D.
The RealViz in Inferno (not Flame mind you) was at least one release back from the stand-alone RealViz. We had to buy MatchMover separate to get a tough 3D track done, even though the Inferno had internal 3D matchmoving. And then, getting the Max camera data to Inferno was next to impossible. Why? Because Discreet did not make Inferno exporters because they wanted you to have to use Inferno's matchmover instead. That kind of stuff steams me beyond belief.
I can charge $250 per hour. More imporantly, I can take the same exact types of projects and charge a flat-fee and still get it done in the same time frame so I can realize even more. And, let's do the math. A kicked out C*/Edit/Max box with realtime I/O and playback? $30k. $30k/$250 = 120 Hours. Flame does not even have the tools that Inferno has, so you are right about $350 - $600 tops. So $275k/($350-$600) = 450-750 hours break-even. Now you need some more speed from your Flame? Not happening. Need more speed from your C*. $10k and you quadruple the speed. Hell, add in the $245k you saved from not buying the Flame, you have 100fold increase in rendering speed. Need another operator for your Flame job? Spend another $275k. Need another one for your C* job? $10k and you are home. Hell, for the $245k you saved, you can add 25 more operators (staffing not included of course).
I think the only impetus to client-driven sessions is clients who think Inferno is the Holy Grail and a real-time box. If they knew how much time and money of theirs we wasted kluding solutions in Flame, the would leave in droves. I once isolated the steps it should have taken for a composite which included forwards and backwards tracking and source-nodes for the track. I think I had isolated it at 6-8 steps ideally. FFI required 58 steps, those extra 50 steps being direclty attributable to it's workflow limitations.
High-End 3D comping is never going to be done in Inferno except by importing rendered high-end 3D. Without real lighting tools, deformation tools, etc. no one is going to do anything but the simplest 3D in FFI. This puts Combustion WAY ahead of it's "big brothers", since the ability to work with extended .rla gives you far more flexibility in the final composite than you will ever get with Flame, PLUS you can integrate 2D elements with true-shadows and reflections. I think FFI are "big brothers" to Combustion the way Cro Magnon man is a big brother to Homo Sapiens.
When 64-bit boxes with the newer bandwith (memory to cpu, peripherals, memory to graphics, etc.) come out this year with higher and higher speed RAM, it is going to be the player who gets that this is the future of post who takes that future. If Discreet continues to think of Inferno as it's flagship and hobbles Combustion/Edit to "protect" high-end sales, they will be trounced. I keep my finger's crossed thay they realize that a Combustion/Edit/Max integration that completely harnesses the power of 64-bit/hypertransport/opengl/high-speed RAM and the potential for those for amazing interactivity/rendering/SD-HD playback I/O integrated to the apps would rule rule rule.
In short, if you own an Inferno now, you have an amazing piece of technology. If you do not DO NOT waste your money on amazing but outdated technology. Buy into C*, get 90-200% of the functionality, and hope that Discreet uses the experience they have gained with the high-end to make it amazing.
Michael
so, ghsu3d -
We got way off in to left field (with albeit good info).
Here's the bottom line as far as I'm concerned.
If this box is for 3D, then get the AMD.
If this box is for both 3D and compositing, then get the AMD.
If this box is for compositing only, and you have high-end Ad Agency clients who are willing to pay the money for client-driven sessions on a flame (especialy if you need HD), then get the flame/Octane2 if you can get it for the price you stated.
Otherwise, I'll agree with flate -combustion2 is definitely the best bang-for-the-buck.
kiwi
Getting back on track as well GHSU3D 
, I just put together an AMD system for the $12k you mention. Specs;
AMP Dual 1900+ MPX
2 GIGS DDR
Netgear Gigabit Card
Seagate 20GB TapeBackUp
Quadro DCC [okay Geforce ti500 softquadro'd]
40GB IDE System Drive
Atto Ul3D Dual-Channel SCSI/RAID
2xSeagate 18G w Granite Internal Cables and Terminators
Targa 3000
Digital BreakOut Box
You'll need two more seagate 18GB SCSI's if you don't want to drop any frames on the T3K.
-you need 60MB/s for an uncompressed realtime dual-stream editing, and you won't get that past 75% capacity on 2 striped SCSI drives. (all hard drives slow down as they get fuller)
Otherwise, that's one nice setup!
FYI,
kiwi
to FLATE:
Just wanted to thank you on a detailed and eyeopening, post you put up.
Thanks.