Wednesday, April 24, 2013

Specifying a Computer to run Modern Design Applications

Much discussion surrounds system specs for running modern design applications. I've made previous posts about system specs (wow that's old!), Jay Zallan has done so recently, but I have a slightly different take on it now after my recent benchmark tests. To make it simple there are 5 main things to consider when spec'ing out a system, something I've touched on previously. I've listed them here in order of importance.
  1. OS
  2. CPU
  3. RAM
  4. Video/Graphics Card (GPU)
  5. Hard Drive
The tough part comes when trying to spec a system for a user. Do you spec one configuration only that covers the majority of needs or have multiple ones depending on a users focus? A designer might be more concerned with running Adobe products, Rhino, Sketchup and Revit early in the project. A sustainability expert would be running still different software while someone wrapping up construction documents would be spending a majority of their time in Revit. Until recently, we have been specifying one spec for all positions, one that can handle large Revit projects with ease.

The thing I noticed with my recent benchmarks testing Revit is that it did not benefit as much from an expensive GPU as I previously thought. Also, I found that steps taken that should benefit from a SSD did not produce as much benefit as I'd hoped. When it comes to OS, you better be on a 64bit platform or most of the rest won't help you.

If you are on a tight budget like I am and trying to get the most benefit for your money, I would go with the fastest CPU you can get as that seems to drive the most benefit for productivity. Next, add as much RAM as you can get, starting with 16GB minimum as Autodesk recommends.

As for a GPU, Revit did great on my benchmark tests with a mid-range card, the Quadro 600. The jury is still out on what happens with it in other applications like 3ds Max, Rhino or Sketchup. Initial experiences from our in house 3ds Max expert Chris Grant said he felt that viewport performance was very snappy, more so than his desktop running a Quadro 4000 with dual Xeon CPU's. We need more testing in other applications, but I predict the Quadro 600 will suffice for them as well. My thought is that because on our BOXX 4050 & the CPU was running overclocked, the CPU was able to hand the data off to the GPU faster for it to process things.

When it comes to hard drives, there are a few things to consider. If you are running a regular physical hard drive (HDD), RPM of the disk speed as well as transfer rates from disk to CPU/RAM affect how fast data can be read/written. You can get some pretty high RPM drives these days for HDD. I noticed on the spec sheets from BOXX that their system with a HDD running @ 7200RPM had the same transfer rate as their SSD. Getting to the data is one thing, sending it around the computer is the other part. I need to run more testing on a similarly configured BOXX with a Solid State Drive (SSD) to confirm my suspicion, but I would have to guess that running with a HDD vs SSD does not offer the performance boost when using Revit that one would hope for.

When does Revit access the hard drive? Per the Revit Model Performance Technical Note, it is only during model load, model save and any hard disk swapping managed by the Windows OS.  What is hard disk swapping? According to this page on Wikipedia, it is part of virtual memory (VM) implementation in the operating system that allows disk storage for data that does not fit into physical RAM. So to me, I interpret that to mean Revit only uses VM if it runs out of RAM. If you have enough RAM, you shouldn't need VM with Revit. This in turn means Revit only really accesses the hard drive when you first open a model and when you save / sync with central. Overall, this is a far less occurring activity during the day and for only short bursts compared with model creation/editing and viewport regeneration. Also, with many large models, they are central files and a good amount of that save time is sending data to/from the server over your network. So I guess that's a long winded way of saying I don't see the benefit from the added cost of a primary SSD. I'm not saying it won't speed things up here and there, I'm just looking at the overall picture of trying to get the most value for my money.

In summary, I can't stress enough that you need to spend as much of your budget as you can on the fastest CPU you can get your hands on. It will just make life so much more productive and pay for itself much faster than other bits you might splurge on. If I had a rich uncle give me a bag full money to go buy whatever desktop I wanted, it would have the fastest CPU (multi-socket/multi-core), best dual GPU setup with a couple SSD's and as much RAM as I could stuff into it. Until then, I'll squeak by with a fast CPU and get by with lesser components elsewhere.

Monday, April 8, 2013

Benchmarking the 3DBOXX 4050 XTREME vs Dell T3600

Well, the numbers are in and I have to say that I'm very happy where they landed! In short, I would be looking at a 28% productivity boost to run the BOXX 4050 instead of the Dell T3600. In my previous post, I mentioned the specs between the Dell T3600 & the BOXX 4050 Xtreme. I've since updated that post slightly to more accurately reflect the T3600 specs that we use. Now for some numbers breakdown.

The Benchmark Tests
I have to give credit to for their posting an updated version of the original AUbench Revit journals and updating the scripts to handle the new versions of Revit. You can download a copy of the same benchmark tool I used (If not a current member you'll need to signup, its free). This benchmark allowed great consistency and automated the majority of testing between systems and versions of Revit. I also ran this benchmark on a couple laptops we have and some of our older systems to get some historical context as time marches on. I also kind of wanted to see how the gap increased the father back we went on T3500 & T3400 models we have floating on the floor. Each machine and each version of Revit ran the benchmark tool 3 times to ensure I could get a fair average.

I also performed some manual tasks on both the T3600 & the 4050. It consisted of working with a Revit 2011 model that was 495MB in size. I took the model and timed the tasks of upgrading it from 2011 to 2012 and then to 2013. Other things were done and timed and noted. One thing to note was that the upgrade from 2012 to 2013 took a long time, much longer than I expected on both systems. Something to watch for as projects are upgraded.

Items of Note

  • As each version of Revit progressed, things took slightly longer to complete, as shown in the total averages below.
  • Many tasks I previously thought of as something being driven by the GPU are still handled by the CPU. For example, the benchmark measures time it takes the viewport to regenerate with Hardware Acceleration (HA) both on and off. However, between the 2 systems, the T3600 has a better video card than the 4050. So I expected the tests with HA on to run faster on the T3600 than on the 4050. However even with a lesser video card, the 4050 still outperformed the T3600!!! This proves that even though view regeneration should be handed off completely to the GPU, the CPU still processes much of the data. Therefore money spent on a faster CPU will go farther than a fancy GPU.
  • My laptop I use that was purchased 2 years ago with a SSD primary HD, nice video card and 1.73GHz i7 CPU got spanked by a new, midrange laptop handed out to Project Managers in Model Creation and Rendering. However, it's on-board graphics card was no match for my Quadro 2800M, my laptop owned it in those tests. So, there is some credence to having some kind of a graphics card and not relying solely with an on-board GPU integrated into the motherboard. This confirms what I have felt over the time of using my laptop, the SSD doesn't offer as much of a boost in performance as I'd hoped. Put the money in the CPU!

In Conclusion
I know what you are thinking - "How do I convince management/purchasers/IT to switch to BOXX from their cherished status quo?" I've found that you can throw graphs and percentages and "everyone else is doing it" until you are blue in the face. What really gets them is cold hard cash. How much $$$$ will the company save if they use this new desktop that costs more than the status quo?

The steps taken in the benchmark tests represent things that would be done many times throughout the day, but for arguments sake, let’s say a person doesn't spend all their time using Revit. In fact, I’ll go out on a limb and say we can only apply the increase in speed over the T3600 once per day for an hour. Let’s run that number out over the course of a year and say the persons billable rate is $75/hour (just a guess for a position that involves heavy use of Revit).
$75/hour * 250 working days per year  = $18750

This means that if the BOXX runs 28% faster than our T3600, we would save about $5250 per year on salary as we could get 28% more work done. We would save even more if the person was on an older system.
Even if the MSRP on the BOXX was roughly $3200, ROI is paid in less than a year!

So, the next time you are due for replacing a system that has become outdated or you are simply looking for the fastest system $$$$ can buy, I'd recommend you check out BOXX and their 4050 Xtreme!

Numbers & Graphs!
For simplicity, I'll list times & percentages first, followed with some graphs.
Hard numbers

Pretty graphs