|
Hello, all. This is a book, but necessary to cover points.
Recently, I bought a new machine, and wanted to determine, once and for all, if it’s really worth buying quadro video cards over geforce, as well as HD speed testing between
SSDs and traditional drives in RAID. This is relvant to everyone who reads this forum. I did this for my own reasons, wanting to have the best performing machine
possible, because I don’t have time left for waiting unnecessarily. I own my own studio (just me ATM), and have 10 machines in my farm. When I buy new, I want to
spend my money where it will get me the most bang for the buck.
Life happens, and while I was well intentioned on making a very in-depth article about this, I just don’t have the time. However, I do think that it would be of benefit to
all to read this. I will likely post it on my site at some point so it can be found after it stale dates here. My software is Max (2008 in these tests), and I tested on
both x64 & Vista 64, using a GTX 285s in single and SLI, a Quadro 5600 and 4500, together and apart. On the hard drive side, I compared 6 OCZ Core SSD drives, and 16
500GB Seagate 7200RPM drives on a Highpoint RocketRaid card. (Full system specs later.)
All tests were done on an i7 extreme system. If anybody is interested in the full spreadsheet, just ask. I did each test 3 times, on a clean install, with nothing
other than OS, necessary drivers, and software tested. I used the benchmarks noted and two of my own scenes. the 9.1MP scene is free to anyone who asks. These
are not truely exhaustive tests, I admit, but represent a lot of the type of work that I do.
HD performance: (see hdperformance.jpg) Tests were done with IO meter in different configurations. As you can see from the graph, you can get a lot more performace out of
HDs as more drives are added, as you would expect. Probably more performance is available with some other (more expensive) RAID cards. This test determined, for me,
that SSDs just are not raid-friendly enough to compare to a larger traditional raid array. Especially when you consider that a 6 disk raid 5 array in this case was $600
(excluding card) for a 160GB array. The same array in traditional drives would cost $540 for a 2.5TB array… prices from 5 months ago. (I was seeing if it made sense
to have 6 SSDs, and 10 7200.11 drives was faster than 16 7200.11 drives… because I also need at least 2TB of working space.) Ultimately, I chose 16X 7200.11 drives, with 4 RAID
5 arrays, because I didn’t want to chance my system to a drive failure. (Even though I have extensive Ghost backups.) Which is fortunate, because a drive failed on me two months
ago in the middle of a project.
Card performance: I was most interested to see how the Quadro cards stacked up against the GeForce cards; I’ve bought many Quadro cards over the years (always used, on
ebay), but even then, a pair of brand new GeForce cards would always have been 40% less than the used pro hardware. I also did this test with Vista, which, for the most
part, was so slow it wasn’t worth doing more tests. Exception being DX10, which doesn’t work in max, anyway.
Specview: (see specview.jpg) As you can see, in this benchmark, the Quadro cards all significantly outperform their gaming counterparts. The benchmark obviously hit a wall
with Maya, where all the Quadro configs rated at maximum. I wasn’t satisfied with this, however, as Spec doesn’t necessarily represent what I would see… and the ‘real
world’ test below confirmed that.
Max specific: (see bismarck.jpg & 9pt1mil.jpg) At the time, the only relevant 3D software I had was max. (I now have SolidWorks, but no time for testing.) These were
done on basic setups, the first being a large scene of the battleship bismarck that you can see on my site, copied once, with the camera spinning around. The second was a
flythrough of copies (not instances) of a piece of geometry, totalling 9.1 million polys. Vista performance was crap, but we know that. The most surprising thing for me
was that the margin between the 285s and the 5600 was minimal, ultimately only made larger by maxtreme drivers. Around 2% without drivers, 18% with. The 285 SLI
setup was marginally slower, unless I had one monitor plugged into each card (which makes most sense.) You can, therefore, decide if it’s worth an extra $2400 for an 18% gain…
there may be more of a gain from going with SLI 5600s, but I wouldn’t want to finance that. :) I averaged wireframe and shaded viewport runs.
Gaming: (3dmark.jpg, ut.jpg, crysis.jpg) Not that this is relavant to a work machine, but I figured it worth doing the tests when I was at it. No big surprise here - the
gaming cards cleaned the clock of the quadros.
Conclusions: Unless you’re in a ‘money doesn’t matter’ or ‘when every second counts regardless of cost’ situation, don’t waste your time with Quadros. If you do,
save 66% and buy on ebay. I did do some non-scientific testing of scenes that used more extensive features, like real-time shadowing and the like, but I didn’t have time
to do definitive tests. (Damn the need to make a living.) It appeared that the Quadros had a little more headroom here, amplified by drivers, but you could just get
yourself VRay RT, and use your old machines (or even buy a couple of small, cheap systems) to solve that problem for less than the cost of the higher-end card. And then
have a bigger render farm to boot.
Notes: I’ve got the Quadros (both) in my machine right now, due the benefits of maxstreme in 2008. I’ve just installed 2010, and I will be testing the 285s, and see how
they work out there. In talking with a SolidWorks rep recently, he told me how he has had personal experience with gaming cards overheating to the point of fire when put
into heavy SW environments. I don’t know what other mitigating factors may be there, like insufficient airflow, but my guess is if you’re dealing with scenes that heavy,
you likely have the budgets for Quadros.
As far as hard drives, large arrays of 7200RPM Sata2 drives are the most cost effective. SCSI drives are just so expensive, and the RAID controllers are even more.
SSD have issues with write/rewrite which can cause many other issues, and simply are not as RAID ready right now. The newest generation may be better than these. As
it is, with my current system, I see VERY fast load/transfer times, due in large part into the way that the arrays are divided.
System:
Intel i7 extreme /w CoolIt Freezone Elite Peltier Cooler
Biostar X58 motherboard
12GB OCZ Platinum DDR3 (4X3GB)
16 Seagate Barracuda 7200.11 drives
SilverStone Strider 1200W 80Plus PSU
Highpoint RocketRaid 2340 16 Channel Raid
Lian-Li V2000Plus Chassis, custom printed front vented/fan mount front grille
3X120mm 140CFM fans, 2X60MM 50CFM fans.
I hope that some of you will find this useful.
Kris.
Mafic Studios, Inc.
http://www.maficstudios.com
|
Attachment
|
|
|
|
Attachment
|
|
|
|
Attachment
|
|
|
|
Attachment
|
|
|
|
Attachment
|
|
|
|
Attachment
|
|
|
|
Attachment
|
|
|
