Questions about gpu xpander

[jamestmather]

Hi gang,

I am on the verge of buying a GPU-Xpander Desktop 4 with 4x NVIDIA Tesla C2075 units in it principally to run Octane.

Before I embark on this Frankenstein project I have a couple of questions and was hoping someone could answer them for me.

1. I'm using a Kobalt laptop as a primary workstation (I have a desktop but prefer working with the laptop as I can go anywhere) - apparently most laptops have a BIOS that cannot identify an external graphics controller How to I establish this is the case?

2. Are there any problems using the Cubix (I know Refractive and Cubix don't play well together) but I'm interested in specific problems.

3. Are GFlops Linear? - In other words my system runs at around 500GFlops (480m Nvidia) - apparently the cubix with all guns blazing runs at around 16000 - By my maths this should be approx 32x faster - Is this the case?

4. Any other observations or wisdom to impart? All input is appreciated.

Thanks and Merry Christmas to all,

Best

James

[pixelrush]

Unless you have a specific need for the extra vram I don't think I would buy Tesla.
Although they are slightly better engineered they are very expensive and really no faster than the Geforce models based on the same GPU.Cubix expanders are kind of expensive too for what they are.
I am thinking 4 of these cards would be hard to cool in a small box and with an adequate power supply it would be reasonably heavy although a little bit more portable than a desktop. I am not actually sure you can get a power supply with 8 connectors? or that 2 if required would fit in Cubix? I guess you would have to talk to Cubix about that.
My personal preference if you have that sort of money to spend and need the performance and/or space would be to build a new workstation with a mobo having many pcie3.0 slots and possibly water cooled gpu and with SSD. Perhaps you can use VPN to run it as a private render farm accessed from your laptop on the road - other people would be in a better position to comment on that than me as I don't have experience of that.
Bear in mind too that Nvidia will release a new generation of cards next year that will no doubt be 25% faster for the same money and these will be pcie3.0 I think. Loading scenes ought to be faster with the extra bandwidth.
Octane performance does scale virtually linearly with the number of cards and you can estimate the relative performance between cards via their Gflops so yes 4xGTX580/3gb for instance will run Octane about 10.5x faster than the 480m laptop on its own....I don't think 32x is achievable currently...21x with 4xGTX590 but they only have 1.5gb for each gpu - a bit limited for Octane but maybe OK for some people. Perhaps you are misinterpreting their figures.
In fact a Tesla 2075 seems to have fewer cuda cores than a GTX570 so it is not the fastest Octane card you can buy.
You would be best to contact Cubix regarding any advice of which laptops function best with their products in regard of BIOS.
HTH
Edit: btw to fully utilise a Tesla card with 6gb vram for super heavy Octane scenes you will need 24 or even 32gb system ram and of course 64bit OS. Note too despite the extra space you are still limited to the same number of textures as any other card by cuda and I guess the hardware design. If you have many many polys and A1 renders perhaps.
16 or 24gb would be good for 3gb vram cards. I am not sure a typical laptop would be in that league.
Without sufficient ram your laptop will be paging to disk and take ages to load up Octane with large scenes....laptops typically have slower disks as well.
Honestly I think your laptop would be a bit of a bottleneck to using 4 Tesla with a Cubix expander.
Better to limit yourself to a couple of GTX560ti/2gb with 8gb sys ram for the road.
2gb is a handy space, it will be fast enough to impress clients and get the job done and be quite portable without costing big money. It will be 4x faster than what you have.
Just my opinion...other people may have different ideas.
Merry Xmas.
Edit2: I should mention too that having a laptop and trying to run Telsa may be problematic from a Nvidia driver viewpoint. I think the usual way to handle Tesla is to have a Quadro card in the same chassis sharing a Quadro/Tesla driver. Not sure how it would function in your case. Some people manage to install different drivers and some people need to hack. All Geforce cards would be better to manage I think.

[radiance]

I would recommend purchasing another brand than gpuxpanders,
They have horrible customer support and we have had MANY problems with their products, staff, and customers left with non working goods and no support requests replied....

Take my word for it, they are not an honest company, i would buy from another company, there are quite a few ones out there nowadays with similar products, just google for it.

Imo the best option is to buy a big case and a asus WS workstation motherboard, then you can stick whatever you like into it and also use sli for gaming etc... If need be...

It will also work out much cheaper than an xpander...
And regarding portsbility issues, the xpanders are just as large as a medium size PC case


Radiance

[pixelrush]

Perhaps its best not malign them too much, who knows their operation might have been dogged by repeat influenza outbreaks...and we know how disruptive that can be.

[jamestmather]

I tend to work on a laptop a (I own a skulltrail 8 core built by a mate which Is wicked fast but requires weapons grade cooling as it tends to heat up like Hiroshima given half a chance).

I was hoping to be able to use the laptop (quad core) and then offload the rendering to a small render farm. The cubix option seems expensive but they seem to offer the only solution out there that works without requiring external monitoring etc. As far as I can see it's just plug and play. - also customer stories I have read (including this forum) seem globally pretty positive - I just don't see any other options. Anyone (or indeed radiance) care to weigh in with specific problems relating to cubix products (as regards octane specifically).

Thanks.

J.

[t_3]

I tend to work on a laptop a (I own a skulltrail 8 core built by a mate which Is wicked fast but requires weapons grade cooling as it tends to heat up like Hiroshima given half a chance).

I was hoping to be able to use the laptop (quad core) and then offload the rendering to a small render farm. The cubix option seems expensive but they seem to offer the only solution out there that works without requiring external monitoring etc. As far as I can see it's just plug and play. - also customer stories I have read (including this forum) seem globally pretty positive - I just don't see any other options. Anyone (or indeed radiance) care to weigh in with specific problems relating to cubix products (as regards octane specifically).

Thanks.
J.

at least not the only one; there are a few other makers in this area:
http://www.magma.comhttp:/www.magma.com/default.asp
http://www.jmr.com/en/products/4/pcie-extender/
http://www.netstor.com.tw/_03/03_02.php?OTM
http://www.crypticglobe.com/pcie/item?id=1

[gabrielefx]

Hi gang,

I am on the verge of buying a GPU-Xpander Desktop 4 with 4x NVIDIA Tesla C2075 units in it principally to run Octane.

Before I embark on this Frankenstein project I have a couple of questions and was hoping someone could answer them for me.

1. I'm using a Kobalt laptop as a primary workstation (I have a desktop but prefer working with the laptop as I can go anywhere) - apparently most laptops have a BIOS that cannot identify an external graphics controller How to I establish this is the case?

2. Are there any problems using the Cubix (I know Refractive and Cubix don't play well together) but I'm interested in specific problems.

3. Are GFlops Linear? - In other words my system runs at around 500GFlops (480m Nvidia) - apparently the cubix with all guns blazing runs at around 16000 - By my maths this should be approx 32x faster - Is this the case?

4. Any other observations or wisdom to impart? All input is appreciated.

Thanks and Merry Christmas to all,

Best

James

Hi, read this, I posted it one month ago:

http://www.refractivesoftware.com/forum ... 40&t=10042

I wanted to buy a Cubix expander too but I paid less for more performance, portability, noise and less problems.
With 5000€ (one month ago) I bought one processor 3.3GHZ exacore, one all aluminum elegant case, one professional motherboard, 2 ssd 128GB hds, one Windows 7 Ultimate license, 24GB of ram, 1500W modular psu, one bluray writer, one Noctua heat sink, 4 (four) GTX580 3GB. A lot of stuff!

It's not a Frankenstein configuration. You can transport this workstation adding a lightweight monitor+keyboard+mouse, you need a car to transport your laptop and cubix box filled with Teslas and psu.
With this configuration you can make money and forget about bios or related problems.
For me there was an alternative: a Tyan barebone with 8 GTX, 10.000€ about.
Another alternative was the Supermicro workstation with dual xeons...but really expensive and noisier than mine. Why buy 2 Xeons if Octane doesn't use cpus? One cpu is enough.

Don't waste your money!

[jamestmather]

Thanks for all the suggestions Guys - appreciate the upper classman's knowledge. I'm now toying with getting gtx 580s with some kind of cooling system attached to one of the systems recommended by t_3 (thanks for those) - One or two quick questions:

1. some of the gtx cards make it to 3gb - Gainward, evga and MSI Gtx lightning - I've searched the forums over the past hour and want to be clear on one thing: Do these 3gb cards offer full access to the 3gb (without sli as I understand it doesn't work with gpu renderers). Any recommendations? Or are they all much of a muchness?

2. Cooling - I understand these game cards run very hot - I know my laptop 480 does (if I ever have kids they'll be born with a permanent tan). I was looking at getting the Accelero XTREME Plus cooling - Any other options? Any cards self cooling?

3. Do the systems outlined by t_3 above work just like the xpander? Do I need an external monitor - owing to my circumstances working on a laptop is my only real option (like I said I own a skulltrail but it's not really feasable to use as I travel a lot and was hoping to set scenes up on the laptop (quad core) and then plug in at night and set the beast rendering. Any thoughts on why this wouldn't be feasible? In my perfect world I'd just plug in - tell octane to switch to the other cards, hit go and go to bed. Is this a castles in the air scenario?

Thanks for all the responses. I really appreciate the input.

J.

[pixelrush]

1. GTX580/3gb allows a full 3gb space for Octane but GTX590/3gb is effectively 2x GTX580/1.5gb ie 2x speed and half memory.
2. Here is a review for a water cooled card, but the 1.5gb model - http://hothardware.com/m/Reviews/EVGA-G ... fault.aspx 35deg under load, no fan noise, a bit expensive but still better value than Tesla.
Manufacturers link for 3gb model - http://www.evga.com/products/moreInfo.a ... Family&sw= actually 3 models available at different clock speeds check out the site
HTH

[jamestmather]

Thanks Pixelrush - That's terrific info. Looks like I've found the right weapon - now just got to strap it to a few others and make it a minigun.

Any other suggestions or assistance regarding the best pci/gpu expander welcome.

j.