What to spec.-up a 3d workstation with?

[cws101]

Need some sound advice on building a 3d workstation box intended to run high-end apps (essentially MayaNT&Soft.). I havent got a silly budget (so no dual xenon, rdram suggestions), but enuf.

I know where i'm at for vid. (FireGl2,Gloria/Syn.3,Gvx1), i've got my 20'(kept that from the O2 i sold to build this), i'll be right with IDE (just go a 75GXP), ...but the platform i'm still pondering (a good 5months there of!).

*I'm purchasing in a tight window unfortunately* -A trip to the states, visiting LA ~23-7 during thanks giving (will shops even be open during this time!?). I need to buy then, or on-line beforehand delivering to my rels there. I tell you this as the time window is going to clash with the DDR thang me thinks. (at least on the 760MP front, but others?)

Im looking for advice on what people have found good for doing _3d work_ (and no, that doesnt (quite) include fraggin!). As i see it, ive got about 4 different options:

1. P3, i815
2. dual P3, VIA 133a (694d, or VP6 soon)
2. Athalon, KT133
4. DDR...pro266 or KT266 (will they be ready+reliable tho by time i buy?)
[BX is out 'cause im hesitant to run a ~$1k prof graphic card out of spec.]

--> What experiences have people hand with these platforms, when using their 3d apps. It'd be great if people who are actually doing work in the industry could comment ...especially people who spec together boxes for production houses, and their experiences there.

I know most of the paper specs on these things, so i know AMD has better FPU which is good for 3d, intel has better industry support for its SSE, etc. What im interested in is real realworld performance fo the platforms.

Thanking you all for your posts,
Campbell.

 

[Goi]

Between the P3/i815E and the Athlon/KT133, definitely pick the latter, as clock for clock its faster than the former, and cheaper too. However, put the dual P3 into the mix and the choice isn't that clea anymore. The dual P3s will probably outperform the single Athlon by a margin, but it will also be significantly more expensive. Depends on where you draw the line I guess.

Nobody now knows for sure when DDR platforms will be out, or which chipset will be the best(speed/reliability). If your timeframe is so limited, then I suggest you stick to a tried and true platform rather than take the risk with a new, probably faster but possibly less stable/buggier DDR platform that you won't be able to return should you wish to. Of course, if they start rolling out soon(like way before you make the trip) and receive immense popularity and great reviews with no apparent problems, then by all means take the plunge.

To sum it all, basically #1 is out, #4 is a maybe, and #2 and #3 are tied, depending on your budget. That's just my opinion.

btw you have the Athlon spelt as Athalon, and have it labelled as #2 when I think it should be #3

 

[cws101]

Thanx Goi.
I think i'm with you on the ddr front, ...im going to miss them, but hopefully miss the drycleaning as well! Besides this will be my first build, so best to go with whats known.

Ok, if i go single, then Athalon (i mean Athlon!, mine sounds more like a race!), dual, then P3 (no-brainer). ...But how does a pumpy Athlon compare to a couple of P3s? I guess dual is always going to do better as far as multi-tasking goes, and rendering (unless dual CPU splits RAM at render, then i'll need more ram to make similar to Athlon w same RAM). Arg.

 

[xtreme2k]

i would go for Dual 700E on a Dual BX mobo since they will give u adequate performance with exceptional stability and speed
they also give u support more than 512MB RAM

 

[Goi]

I don't have hard numbers, and I haven't seen any websites comparing single Athlons to dual P3s, so I can't say for sure. I'm just basing my analysis on common sense. The P3s are behind Athlons in single processor systems, but not by THAT much. They are behind by quite a noticable margin, such that a lower clocked Athlon several speed grades down can beat a higher clocked P3, but I don't think its something that throwing more CPUs at it can't solve. Of course, it requires that the software and OS be able to take advantage of the 2nd CPU. I'm not sure if 3D visualization/rendering software are multi-threaded or have support for multiple processors.

 

[cws101]

On the BX side of things, i think your certainly right about stability ..but speed i think comes in with the O'Cing, where the chipset shows superb memory performance with robust bus scaling...but at default 100 it is just solid. I think opening up the system bus, and running with cas2 PC133 RAM, will be of great benefit for the apps im interested in. Im scared to clock up the BX AGP bus out of spec if i've got $1.2k FireGL2 riding in it, if i go with a nVidia (Gloria/Synergy) card then i might be more prepared as their GeForce chipset seems fairly happy seated @89Mhz interface (thanx to the gamer world's testing!). Dual P3 on a VIA 133a is not limmited to 512, most can do 1-1.5gig, so that aint an issue. And im not going to pay that much more for a dual 133FSB geared cf BX mobo.

I've seen quasi-quantifications on render speeds for different machines equiped with different configurations, and you see things like a gig Athlon pipping a dual 600e on render time. Not too much of a surprise. MayaNT has various aspects which are threaded, of most note is their IPR renderer which is an on-the-fly texturing port, also of note is their batch renderer which will enable you to render out a scene on one processor whilst you continue work on the other (this is the type of thing im interested in, i can wait for a scene render ...time is not money at this stage [if so, i'd buy some time with a mini Athlon farm], ...im interested in the multitaskability of a dual platform .....but can a single chip offer similar!?)

I appreciate your comments guys,
any more suggestions?

 

[xtreme2k]

i personally wont count on VIA as my 'workstation' chipset

too much compatibility and stability problems

 

[cws101]

xtreme2k,
Yeah, stability. Save the time a quicker setup would have lent you by dealing with less maintenance. But you always wonder the what-if tho, people do manage to run via solid.

Can ask what you run on? ...and if you had a chance to purchase a totally new system with whats now, what would you do?

cheers

 

[Viper GTS]

My picks:

MOBO: Dual P3, BX based. Or, the Intel OR840... Granted, you'll have to buy RDRAM, but it's price is dropping rapidly. Currently $1,000 for 512 MB. Prob'ly BX would suit you best, dual slot one. Get one that supports a 133 bus, you'll need it to make the most of your low-cost CPUs.

CPUs: Dual P3 700e's @ 933. There are people around here selling 700e cb0's guaranteed to 933 at default voltage. I have one, &amp; I love it. If you get the OR840 board, you won't be able to overclock the CPUs, so you might want to get &quot;real&quot; 933's. Drop a couple Agilent ArctiCoolers on them, boost the core voltage to 1.7V, &amp; do 2x933 at a very comfortable <40ºC apiece. And the ArctiCoolers are very quiet to boot.

HD(S): Look into the real IDE RAID solutions, &amp; then drop a 128 MB DIMM on for cache. They're expensive, but they'll give you good speed &amp; redundancy. That &amp; 2 or 4 75 GXP's would make a very nice disk subsystem.

On a side note, i815 is not SMP capable. You have to get the i820. I found that out recently when I looked into building a dualie rig. Which, BTW, I'll be doing exactly what I suggested to you except with only 256MB RAM since I don't need 512 MB for web surfing &amp; casual office/corel draw/photoshop work. Hell, I don't need 1932 MHz, either, but I sure want it.



Viper GTS

 

[cws101]

Im starting to be wooed. You talk of dual slot1 mobos that 'support' 133 FSB, ...whose doing 'em? Could you name a few models. What sorta support you talking?

---> There is this remianing problem of AGP running over spec tho, unless these mobos have fixed that!? <----

As for the procs, can you still get Slot1's readily ..is that what ppl around the boards are selling? Or do those 700e's only come in a Slot1 package anyway? Or will i need a couple of 'em slocket (sp?)translators?

Cheers

 

[xtreme2k]

if new system for me

(well i am a gamer so i need less than 512MB ram)

Asus CUSL2 (i815-based mobo) will be for me

it seems to be a bit quicker than the Via 694x and has less problems with the AGP issues


again
i recommend Dually 700E (whether Overclocked or Non-overclocked) this is still gives a good speed and good value

 

[BenSkywalker]

For multitasking performance, a dual rig would definately be better. If you try to render a scene on a single processor unit you can forget doing anything else, it doesn't work.

For raw performance, a 1.1GHZ T-Bird will easily and handily destroy any dual PIII rig you can setup for a like amount, and be very close to if not slightly better then dual PIII 933MHZ rigs(the Athlon crushes the PIII in rendering, it isn't even *close*, not to mention far less then ideal SMP on the GTL+ platform).

Forget ultra high speed hard drives, that is the last thing I think about when working on a scene, you should have enough RAM to avoid swapping, 512MB or more for Maya particularly(nasty RAM pig, but I'm sure you know that). When rendering to file, your HD most certainly won't be the bottleneck.

With what you are looking for, a dual PIII may be the easiest solution. What I do, when needing to render a scene or clip when I still want to work, is send the render off to a second machine which sits on my network(no monitor even hooked up) to render when I need CPU cycles. This is an old K62 machine, slow as all he!!, but it frees my Athlon rig to keep working at full speed while the file is being rendered elsewhere.

If you built two seperate &quot;lower&quot; end T-Bird systems, in the 750-800MHZ range, one barebones with a cheap $5 vid card, RAM(don't need much for a render rig), a small HD(to boot and hold the app plus swap file for scene data), write files back to the main rig and then equip the main machine properly, you may be looking at saving yourself money and getting the best return from the performance aspect.

The raw CPU power isn't going to make a big difference(within reason) when working on models, not nearly as much as the proper video card and system RAM, with the Athlon still holding a decisive edge even in that area.

If you do go with an Athlon, I would advise staying away from Diamond's &quot;Fire&quot; boards. They are excellent offerings when paired with a PIII, but they rely heavily on SIMD for performance and that creates a twofold issue, one is that their drivers are far better optimized for SSE and two SSE is simply better then 3DNow!.

The simplest way to go I would say would be dual PIIIs with 512MB RAM, a ~40GBish 7200RPM hard drive with a solid video card, the ones you are looking at, excluding the GVX1 which is quite slow(the GVX210 is much better, though more costly) in comparitive terms.

 

[Goi]

There you have it...BenSkywalker has spoken
I didn't know The P3 was that much behind the Athlon that a couple of P3s wouldn't even match an Athlon...interesting...

 

[cws101]

I ken your wisdom Ben. And similar posts in the thread. Appreciated.

What are peoples thoughts on the soon-to-be-released Quadro2s? (Gloria&amp;SynergyIIIs).
Ware them like the quadro2? Faith with ELSA team closer to hardware now?

If you go dual bx133, does the overclocked AGP buss pose a concern for prof graphic cards? This is my issue with the dual setup. I want bx action, @133Mhz (@100Mhz not worth/size up?), but i need to know my video will go.

Are the Via 694x duals, with whole system w/in spec, simply not a viable (no pun intended) solution, because of stability/compat.? Could we expect the Abit VP6 to be a better take on the chipset?(than tyan's, MSI's and gigabyte's efforts).

Cheers again.

 

[pm]

Why not go with a prebuilt HP Kayak? With an FX graphics card, these systems are about the fastest that you will get for 3D graphics in the x86 world.

Link to HP's Workstations page

 

[Goi]

pm, don't you mean the HP Visualize? According the your link, the Kayak's only come with Oxygen GVX1, Elsa Synergy II, Matrox G400 or G250(whatever that is). No mention of the FX. The Visualize do use the FX graphics cards, and do have P3 Cumines as an option, although no Athlons/Tbirds.

What's an FX graphics card anyway?

 

[BenSkywalker]

&quot;What's an FX graphics card anyway?&quot;

The kind of gfx card where you decide between it and a new car

They aren't that bad, but for the top of the line card(fx10 now if memory serves) you are looking at a $25K workstation(for a higher end Visualize). Anytime I have used a machine equipped with one, I have been extremely impressed, but unfortunately its' price premium makes it an unwise choice for most people unless you are dealing with higher end demands(automotive engineering, video production for television/movies and such). They(Visualize fx machines) slot nicely between SGI's PC based machines and their O2 workstations, in features performance and unfortunately price.

When the IA-64 is paired with the latest and greatest fx graphics workstation I imagine that it will send many companies, including SGI, back to the drawing board. Too bad they cost so much

 

[BenSkywalker]

Sorry to post twice in a row, but I completely forgot an important aspect-

First, I am not up to date enough on mobos to help out advising on a particular board but I would avoid the BX boards. For the majority of people, a BX board is still the best platform to run a PIII on, simply by OCing it to 133FSB. The problem with what you are looking for is two, one being that AGP will be out of spec and two, AGP4X does do quite a bit of good in the applications you are looking to use this for, much better then the less stable OCed AGP2X.

Unless you drop serious money on a board such as the Wildcat with obscene levels of RAM, the machine will more then likely require AGP texturing fairly regularly. This, coupled with AGP fastwrites, eliminates additional CPU overhead and can also improve performance significantly.

When looking at the killer Wildcat/fx style graphics boards you will notice that they include massive amounts of RAM, but if you look closely you will see that most of the RAM, the texure storing portion, is very slow EDO RAM which can be quite a bit slower then AGP texturing under many circumstances.

Couple this with the ability of AGP4X to handle twice the vertex data(hence, higher geometry throughput) of AGP2X and trying to stick with a full AGP4X compliant board makes the most sense.

The pro market is one area that AGP aids greatly in, both because of the additional possible texture memory and also the higher level of geometry the greater bandwith can handle.

 

[Goi]

Cool, you learn something new everyday. My ex-college actually had some HP Visualizes, I believe they were the C360 model(not even sure if they're Visualizes, but I think so), but they used PA-RISC instead of P3s. I didn't run anything 3D on them so didn't know they were that high performing.

Anyway, any specs on the FX boards? Would like to know more about them

 

[cws101]

Cant afford a pre-built workstation.

Benskywalker, do you have any suggestions on what graphics card to go with? (for running maya primarily). Or a bit early to make a call with upcoming releases? I cant go a new wildcat, but perhaps a secondhand 4000, what do you think they're worth? ($/performance).

Cheers.
c.

 

[redpriest_]

Ben said:

If you do go with an Athlon, I would advise staying away from Diamond's &quot;Fire&quot; boards. They are excellent offerings when paired with a PIII, but they rely heavily on SIMD for performance and that creates a twofold issue, one is that their drivers are far better optimized for SSE and two SSE is simply better then 3DNow!.

I say:

I agree with you mostly, but I have to say one thing about the 3dNow vs SSE issue:

Um, no. SSE is NOT better than 3dNow. Period.

You'd know that if you worked with both.

 

[BenSkywalker]

The biggest problem with selecting a board for Maya is drivers, everyone seems to have some problems running it, with both the Oxygen series of boards and the nVidia based offerings having some more serious issues for the last round of testing I have seen(Oxygen couldn't run in flat shaded mode, nV had issues with certain overlays if I recall correctly, 5.xx series Dets for nV).

In terms of performance, the nV offerings clearly are the best in bang for the buck right now. The older 4000 Wildcat wouldn't stack up well in terms of performance to a Quadro1, let alone the Quadro2 board.

Looking out, or perhaps they already have, I would expect nVidia's driver support to mature extremely well for Maya with SGI now utilizing their chips in their PC workstations. This, in the long run, may end up being a more important factor then performance alone.

With Diamond and Elsa both about to, or recently have launhced a new board you might want to take a look around for some reviews on some of the 3D visualization sites. I haven't seen them or touched a machine equipped with them in anything at all yet, let alone had any sort of time in Maya with them.

 

[pm]

Oops - sorry, Visualize - not Kayaks... as several of you mentioned.

I design chips for a living - the closest that I get to 3D graphics is blowing away my friends in UT.

I can personally testify as to the power of the Visualize workstations. I work on a Visualize C3600 and it rocks (no FX card, though, we don't need 3D in chip design... yet). Vastly faster than the old C360 that I used to use before the upgrade last spring.

A good friend of mine does workstation sales for HP said that it's possible to buy an FX10 without actually buying a Visualize worksation. He said, IIRC, that one is approximately $2.5k. The amazing thing is that they basically have 6 PA-RISC FPU cores on the board to use for geometry acceleration - PA-RISC SpecFP scores are pretty impressive. The FX10 card looks industrial strength too - they have handles on the ends to help push it down.

The Kayaks are still solid 3D graphics workstations (or so my salesman friend assures me) with or without an FX graphics card. Of course it will be more expensive than putting it together yourself - and they have i820's on them (why didn't they use i840's I wonder). But although Direct RDRAM is much maligned, the high bandwidth is often very useful in 3D graphics when coupled with the 4X AGP bus, so this is a situation where the benchmarks favor Direct RDRAM over PC133 SDRAM.