Dezignstuff SolidWorks Blog

So I bought this Xi computer. The experience of working with the company was pretty good, and I have to say it was a lot better than some of the other machines I have bought recently, like Boxx and Hypersonic. They took the time to explain to me why I needed this or that and why I didn’t need this or that. This helped me make a decent decision, I think. They answered my technical questions and gave reasons for their suggestions that didn’t sound like a sales pitch.

The goal of this computer was to have a desktop (tower) that was reasonably economical, and yet reasonably fast for SolidWorks complex parts. I wasn’t out to break any records, and I wasn’t out to skimp, but I didn’t plan on buying stuff I didn’t need this time. In the past I’ve paid $3-4k for a computer with stuff I never wanted or had the opportunity to use, like SLI, Bluetooth, multiple processor sockets, tons of mother board connections for external power, external scsi drives, and lord knows what else.

Processor

I keep hearing that the processor is really the king with SW data. I also keep hearing that Xeon processors don’t really give any advantage, in fact they give a little disadvantage. Also quads (for straight SW) are probably not the best choice, especially if you have to give up clockspeed compared to a dual core, and especially if price has any thing to do with your requirements.

So, from that, I spec’ed the fastest dual core processor I could find, which was the Core2 Duo E8500 3.16 GHz. This only added a very modest amount to the base cost of the machine as spec’ed by Xi. I could have (and maybe should have) had watercooling and overclocking (up to 3.4 GHz) for an extra $300 or so.

Memory

I don’t do large assemblies. I don’t do huge drawings. I do big complex parts, with lots of features (largest had about 1300 features, average parts have 200-400).  I never ran out of memory even on my 2 gb laptop. Still, this is a box I don’t want to have to make any apologies about for a couple of years, so I tried to be forward looking with it. I spec’ed 4 gb of RAM, with space for 8. Of course I have to go with a 64 bit OS to do this, so I went 64 bit Vista. Not because I’m a big fan of Vista, but because in a couple of years, new drivers will probably not be written for new devices on XP. Vista isn’t really horrible, but it’s not really compelling either.  I had to have an OS, and I couldn’t use OSX or Linux, so there I was.

As to the DDR2 vs DDR3, the fellow from Xi talked me out of DDR3. The difference between 800 MHz and 1333 MHZ sounded important to me, but he explained that due to latency and mobo differences, the difference was not very noticeable. The mobo was much more expensive for the DDR3.

Video Card

This really wasn’t as big a decision as you might think. I knew I was going to go nVidia, I know FireGLs are reputed to work, but I just don’t like seeing that ATI logo on a SW box. I knew I was going to go Quadro, even though I know you can apply a soft quadro hack to a GeForce and get just about the same thing. The choices were the 570, 1700, and 5500/5600. The 5600s just seem ridiculously priced, and the fellow from Xi that I talked to doubted that I would see much of a benefit from the 5600 as compared to the 1700, plus the 5600 adds $2150 – doubling the cost of the machine I wound up with. I didn’t see any reason to skimp on the 570, since the 1700 was only $250 more than the 570.

So, I think I found the sweet spot for what I’m doing, and didn’t have to spend a ridiculous amount of money to get reasonable performance.

Hard Drives

I think hard drives are really a terciary concern in a SW system, unless you are doing a lot of file operations locally. On my little network here at home, I’ve got probably 1.5 TB of storage, about half of which is used. I really didn’t need more space. I wanted fast drives, but the drive speed was not as important as the processor speed. I got 1 10k rpm 130 GB WD Raptor. No RAID. I was going to do a RAID 0, but decided against it mainly because its twice as likely to go belly-up on you. Of all the computer components I’ve had fail, drives have failed more often than anything.

Again the fellow from Xi talked me out of the 15k rpm drives in favor of the 10k Raptors that have been around for a while. If part of his prejudice was because of reliability problems, then he served me well. I want a box I can bang on and forget about the hardware.

DVD

I got a bare-bones DVD reader. With the other computers in the office, I have 3 other DVD writers and an external DVD writer that I don’t use already. There was no reason to load this one up with another. I’ve got an external usb 3.5 floppy drive should I ever need to use that.

And that was that! It came to $1997 before adding Office 2007 Small Business Edition. Not bad. Let’s see how it performs.

Benchmarks

Benchmarks are a touchy thing, especially for SolidWorks. The benchmarks that exist are really meant to benchmark hardware against other hardware, but a lot of people use benchmarks to compare versions of SolidWorks. Some people come up with their own homegrown benchmarks, but they don’t take into account the entire range of capabilities that you can run into. The only real benchmark that has existed for SolidWorks is the Spec.org benchmark for SolidWorks 2007. I’ve heard that SW is working with SPEC on a new benchmark that exercises the processor to the same lavel as the graphics. This benchmark is not available for 2008 or 2009 at this time.

There are two other informal homegrown benchmarks that you can find results for. One is Mike Wilson’s Ship-in-a-Bottle, which comes with a macro that you have to run, and then use Feature Statistics to give your time for 50 iterations of the macro. This one is a simple part that uses some surfacing, and actually makes changes to the model which rebuild with each cycle of the macro.

Of course there is an error with the macro when run in SW09. The files were originally created for 2001 or 2001+.

A second informal benchmark is Anna Woods punch holder. She keeps a spreadsheet of specs and results. The punch holder is a part that could have been modeled differently to rebuild faster, but it’s a single part with a lot of holes and a lot of patterns. My rebuild time for the punch holder on my two older machines was in the 177 second range. At the time when the machines were new, it was a respectable time. The Xi runs this test in 87 seconds. The fastest times I have seen on the SolidWroks forums are in the mid 70s. So 87 is not bad for a box that is not geeked out.

My personal benchmark part is this one, with 263 features, mostly surfaces.

My Boxx box, AMD single core processor, XP32, 3 GB RAM, Quadro 3450 rebuilds the part in 125 seconds, using 1 GB of RAM .

My tablet PC is kind of a strange choice for a CAD box, but I only use it for traveling, and it works fine for occasional simple modeling, user group presentations, and Word docs. It has a dual core 2.4 GHz proc, 3 GB RAM, and mobo based graphics (eek!). Total power mismatch. The processor is far ahead of the video. Anyway, the tablet PC rebuilds the part in 93 seconds. A big improvement over the older machine. The tablet PC also used about 1 gb of RAM for this test.

The Xi rebuilds it in 53 seconds. Now that’s a time savings worth having.

Take a look at the task manager for the Xi rebuilding this part:

This shows that the CPU was between 65% and 100% for the entire rebuild. It never went below 65%, except in the first couple of seconds, as it started. This I think puts to rest the stuff that people say when they claim that SW modeling doesn’t use multiple threads. I think the distinction is that multibody modeling uses mutiple threads more than predominantly single body modeling. The situation was very similar on the tablet PC, although it took almost twice as long to accomplish the same rebuild. Dual processors work for SolidWorks.

Notice the amount of RAM used. This is the difference between XP and Vista – .8 GB RAM. Vista is a veritable RAM hog, but it is also slightly (10%) faster, in other tests that I’ve done side-by-side on the same computer.

The final benchmark I did was to compare the Boxx running Vista to the Xi using the Windows Experience Index. This at least is a consistent way of measuring one set of hardware against another, regardless if it is valid or tuned to SW or not. The results probably require some interpretation. Here are the results for the Boxx and the Xi:

The Xi is clearly ahead in the processor and RAM functions, but the 3450 card in the Boxx makes a big difference. If you discount the Aero interface graphics, the Xi score jumps to 5.3. It is interesting that while the Boxx’s 3450 video card scores higher than the Xi’s 1700 for the Aero interface, the 1700 scores higher for “3D business and gaming”. The fact that they lump these together I guess is Microsoft’s tribute to themselves, or their own DirectX graphics standard which is in opposition to OpenGL, which SW uses.

Access the Windows Experience Index in the Control Panel, System Maintenance, Performance Information and Tools.

So far I’m very pleased with this new machine. It is far faster than what I have been using, and was much less expensive than those boxes were as well. I can recommend both the experience and the hardware. They also have a notebook… It’s not time for me to replace a notebook yet, but when I do, I want to have a look at the Xi offering.