Please Reveiw my potential 3d workstation

[kickercav]

I work for an architectural firm and lately we have been doing alot more 3d modeling. Recently my boss gave me the go ahead to spend $4500 for a 3d workstation.

My question is what CPU should I look for within this price range. The software we will be using is Architectural Desktop 2004 and the latest Version of Autodesk Viz when it comes out (this is not included in the cost of the CPU).Do any of you know what type of CPU's other firms are using to produce high quality images/animations? We will network render with Autodesk Viz but we still want to purchase a super CPU that handles the majority of the rendering/modeling work load.

Also, from what I read a video card helps in rendering when your working on a model (using orbit with shade mode turned on), not while rendering a scene which is done by the cpu's processor, is this true?

Here is spec of what our computer provider offered us, can you please review and make any suggestions/comments:

SuperMicro Workstation 7033A-T
Intel Xeon 3.06GHZ/512 533K CPU
2GB 266MHZ DDR PC2100(4X512MB)
Seagate 80GB SERIAL HD SATA (4 total)
PROMISE SATA RAID 0,5,10 CARD
NVIDIA QUADRO 380XGL/128MB AGP
MS Windows XP Pro
APC 500 UPS BACK UP


The fastest CPU we have is a P4 @2.6GHZ ( w/ hyperthreading) and 1gb of ram. It produces high quality single frame renderings pretty well (using VizRender) but if I was to produce a 30 second walk through it would take about 2-3 days (and since it's my machine I would have to work on some other CPU during that time). Would you know how this CPU would compare againts the above "super" CPU.

Any information on this subject would be greatly appreciated!

 

[Guden Oden]

You may want to look at a dual-processor Opteron system instead of a Xeon setup. Performance for the most part is considerably higher.

 

[jandar]

Look at the ZPro workstations from IBM.

Our CAD guys use em.

Dual 2.8Ghz XEONS with 1MB L2
2GB RAM
2 36GB Ultra320 SCSI
Quadro Video card.

kinda sound like a jet engine taking off when power on.

but you can build a nice dual opteron for pretty cheap as well.

 

[KimB]

Definitely check out a dual Opteron system. Those chips have a very fast memory interface. I don't know how much this will matter, but you can also deal with greater than 4GB memory without issue in an Opteron system.

Edit:
Btw, here are a couple of benchmarks of Opteron vs. Xeon:
http://www.aceshardware.com/read.jsp?id=60000284

As it turns out, this is about the Xeon's best showing vs. the Opteron. In the previous webserver type of benchmarks, the Opteron just slaughters the Xeon. Anyway, I'd suggest looking at some more benchmarks. A simple web search should turn up a few.

 

[Kronus]

http://www.extremetech.com/article2/0,1558,1403007,00.asp

AutoCAD, etc. Xeon Vs. Opteron benchmarks.

I'd go with the Opteron.

 

[epicstruggle]

kickercav, do you know if the software your planning on using can take advantage of multiple cpu's? Others have stated above that it might be worth looking at dual (or more) opterons. But you might want to make sure that the programs your going to use can take advantage of a dual cpu setup.

Hope you find a good solution.

later,
epic

 

[kickercav]

The software we will be using is Autodesk ADT2004 for the modeling and Autodesk Viz and VizRender for the rendering. My current machine can handle the modeling part (though I probably should update my video card) but for rendering it's just too slow for what we need (animations/high quality renderings). Deadlines are deadlines


Thank you all for the great posts and keep them coming!

 

[T2k]

I have dozens of dual Xeons from 2.8-3.06 and two dozen dual Opterons, few w/ 248 and majority w/ 246.

We do polygon-based anims as well as voxel (volume) animations and I can tell you, I'm not gonna buy anymore Xeons. Dual Opterons a WAY FASTER, especially when it comes to rendering.

Negative side: FGL X2 does not work with certain dual Opteron mahcines (Tyan S2885 mobo) under max6. That's it, period.
And when I posted it over R3D, the answer was a big silence...

Another note: I don't know how much you'll pay for that Quadro but I gues pretty same like for FGL X1 - if so, definitely go with X1. Much better card than any Quadro4.

 

[Sazar]

opterons would be the way to go... cheaper too I would think though I am not sure of the pricing of these components..

firegl's will be the cheaper alternative to a quadro but quadro's have proven themselves... IF you have lots of money left over after purchasing your opterons/ddr 400 ECC memory (please do get this instead off pc 2100) figure out how much you have left and invest in the best card you can (subtracting for other components naturally)

t2k did hit on something though with compatability issues... ensure all components are compatible before purchase...

your new rig should rock compared to your old system... even if you went the xeon route...

 

[KimB]

Negative side: FGL X2 does not work with certain dual Opteron mahcines (Tyan S2885 mobo) under max6. That's it, period.
And when I posted it over R3D, the answer was a big silence...

Another note: I don't know how much you'll pay for that Quadro but I gues pretty same like for FGL X1 - if so, definitely go with X1. Much better card than any Quadro4.

Didn't you just say that you had problems with the X2? Why would the X1 be any better?

Anyway, here are some comparisons between the FireGL X1/X2 and the Quadro FX:
http://www6.tomshardware.com/graphic/20030916/index.html

Remember that a Quadro or FireGL will only make it more pleasant to view large meshes, and won't accelerate rendering. I'd say the decision on which video card to purchase should be much less important than what CPU to purchase, and stability/reliability should be the primary concern.

 

[stevem]

The key is to know your software. Be certain your software is multi-CPU/threaded aware if you intend a dual CPU system. Also consider larger cached Xeons & higher bandwidth RAM, as suggested. Opteron CPUs perform extremely well due to EV6 PTP topology. Make certain if choosing a dual Opteron, that the board employs independent banks of RAM for each CPU. Also drop the 500VA UPS. It's next to useless for your purposes. A unit with >2x the rating should be considered...

 

[Saem]

Opteron CPUs perform extremely well due to EV6 PTP topology.

Hrm, IIRC, the K7 used the EV6 PTP. While the Opteron uses HT to enable a PTP topology.

 

[KimB]

The key is to know your software. Be certain your software is multi-CPU/threaded aware if you intend a dual CPU system. Also consider larger cached Xeons & higher bandwidth RAM, as suggested. Opteron CPUs perform extremely well due to EV6 PTP topology. Make certain if choosing a dual Opteron, that the board employs independent banks of RAM for each CPU. Also drop the 500VA UPS. It's next to useless for your purposes. A unit with >2x the rating should be considered...

The memory controller is integrated into the CPU, so memory performance should be essentially independent of the motherboard chosen. And yes, the Opteron uses Hypertransport as its bus technology.

 

[T2k]

Negative side: FGL X2 does not work with certain dual Opteron mahcines (Tyan S2885 mobo) under max6. That's it, period.
And when I posted it over R3D, the answer was a big silence...

Another note: I don't know how much you'll pay for that Quadro but I gues pretty same like for FGL X1 - if so, definitely go with X1. Much better card than any Quadro4.

Didn't you just say that you had problems with the X2? Why would the X1 be any better?

Chalnoth... I'm using X1s instead of X2s in these Opterons.

Why better? Because it's a different card.

Anyway, here are some comparisons between the FireGL X1/X2 and the Quadro FX:
http://www6.tomshardware.com/graphic/20030916/index.html

1. Tom's Garbage not an argue, you knew that very well. Especially don't trust them when VGA comes to the picture - and you're linking something from last September? C'mon...

2. BS, let me tell you. I have many QuadroFX cards as well (few 2k, lot of 3k) and all my polygon guys prefer X1 or X2 - when I replaced some of them with Q30000, they came up within' 30 minutes, crying for their X1/X2...

Believe me: we/they do know better than any newspaper/site/etc you'll link here. Currently GLmax is FAR better than MAXtreme - that's the case.
(What do you think why NV just re-started its MAXtreme drivers?)

Remember that a Quadro or FireGL will only make it more pleasant to view large meshes, and won't accelerate rendering. I'd say the decision on which video card to purchase should be much less important than what CPU to purchase, and stability/reliability should be the primary concern.

ANd also check the prices: X1 costs (I think) some $500, X2 costs ~$700 and QuadroFX 3000, the only 256-bit memory bus Quadro starts over $2000 - and its speed advantage over X2 is ~5-10% only...

PS: QuadroFX cards are ridiculously overpriced, period.

 

[stevem]

Hrm, IIRC, the K7 used the EV6 PTP. While the Opteron uses HT to enable a PTP topology.

Yes HT links. Olde habits die hard...

The memory controller is integrated into the CPU, so memory performance should be essentially independent of the motherboard chosen.

Unless it's multi CPU with shared memory banks.

 

[Sazar]

current HT tops @ 800mhz... afaik the next cpu's out should sport 1000mhz HT though I am unsure as whether or not this is applicable to the opterons as well as the d'tops... the sckt 939's SHOULD have 1ghz HT sooner rather than later...

 

[KimB]

Unless it's multi CPU with shared memory banks.

Um, I have a hard time believing this. Do you know of a specific motherboard that does this? Doing this at the chipset level would incur a very significant performance hit.

 

[stevem]

Um, I have a hard time believing this. Do you know of a specific motherboard that does this? Doing this at the chipset level would incur a very significant performance hit.

Why? It's sensible. All Opteron's can read data from memory at all nodes via their HT links as well as from local memory. So a single memory bank dual Opteron results in something akin to AGLT+ with its shared bus. Performance penalty is a given...

Example 1

vs

Example 2

If only you were as sceptical of a particular IHV...

 

[Guden Oden]

Steve,

I don't think there's a single Opteron mobo planned, much less actually RELEASED, that does not have separate DIMMs for each CPU. You know why? Because it would be plain dumb.

First and foremost, the cost saving would be minimal at best. The Opteron chips themselves already account for a major part of the system cost, the mobo a minor one, and the cash saved on leaving out memory sockets for one CPU even less.

Second, you'd lose a lot of the benefit of having two CPUs to begin with, resulting in a lot of wasted money rather than saved when both CPUs get starved of data. You'd have an on-die memory controller sitting there doing NOTHING, which means a lack of several gigabytes per second of useable bandwidth as well as halved maximum memory capacity!

Third, performance hit would be considerable. It's not smart to save money where it results in the biggest possible performance hit in a system, especially if it means the smallest possible cost saving.

Switching from SCSI harddrives to 10k RPM SATA might be a worthy cost saving for a rendering box or small server, but skimping on a few plastic DIMMs? Jesus!

Got any more bright ideas while you're at it?

 

[Thowllly]

Steve,

I don't think there's a single Opteron mobo planned, much less actually RELEASED, that does not have separate DIMMs for each CPU. You know why? Because it would be plain dumb.

So, do you think his first example is a photoshop or something? Here is another one with DIMMs for only one of the CPUs, and it's been out for a long time:

link

First and foremost, the cost saving would be minimal at best. The Opteron chips themselves already account for a major part of the system cost, the mobo a minor one, and the cash saved on leaving out memory sockets for one CPU even less.

I don't think it's as much about saving money as it's about saving space. For now, all dual Opteron MB with memory for each CPU has been EATX.

Second, you'd lose a lot of the benefit of having two CPUs to begin with, resulting in a lot of wasted money rather than saved when both CPUs get starved of data. You'd have an on-die memory controller sitting there doing NOTHING, which means a lack of several gigabytes per second of useable bandwidth as well as halved maximum memory capacity!

Third, performance hit would be considerable. It's not smart to save money where it results in the biggest possible performance hit in a system, especially if it means the smallest possible cost saving.

Switching from SCSI harddrives to 10k RPM SATA might be a worthy cost saving for a rendering box or small server, but skimping on a few plastic DIMMs? Jesus!

Unless the OS is NUMA aware, the gain from giving memory to both CPUs is pretty small, according to the tests I've seen.

Got any more bright ideas while you're at it?

Nice. Lets insult Steve for a decision made by MB manufacturers. Yes, lets insult him for not being as uninformed as you are.