One problem leads to another

1308 0

We all experience bottlenecks in our life and in our work. Often you need to replace the whole system, but sometimes by removing one problem component we can all get from A to B much quicker, writes Rob Jamieson
While driving back from an event the other day I ran into a three-mile tailback. Trundling through the traffic I eventually got to the blockage, which was caused by two old cars that had crashed into each other and a policeman waving people by while the cause of the accident was assessed. It was a very minor accident, and the cost to the insurance companies would be very little, but instead of the policeman pulling off the two slightly damaged cars, thousands of people got stuck in the tailback and had to pay the cost in time and money.

This wasn’t the first time the theme of bottlenecks had cropped up that day. At the event I had done a presentation on workstation tuning and received a lot of questions from different people about how they could improve the machines they had or if they would be better off buying new.

The issues presented were certainly indicative of those faced by the wider DEVELOP3D readership so this is a good opportunity to recount a few of the questions and the suggestions I gave.

One guy used 3D modelling software but had slow performance with large models. Task Manager reported that 3.5GB out of 4GB of memory was in use and he was on the 64-bit version of Windows XP. I explained that Windows hides its use of some memory and he was probably paging i.e. using the hard disk instead of fast memory. As the rest of the system was well specified I suggested simply adding 4GB of memory, which should give an immediate improvement.

The next customer was having a bad time with their IT department, that was specifying a basic desktop unit, then placing an entry-level professional graphics card inside it. The CAD users were suffering all round poor performance. 2GB on CAD workstation today is just not enough – my netbook has more – and the CPU performance was so bad that it slowed down the graphics, so much so that they thought the graphics card was also at fault. As they were planning a complete change in IT I did a little analysis of their requirements and budget and suggested a single socket CPU system with a mid-range professional graphics card and enough RAM.

Loading other software before you load your CAD application can cause problems with memory. How many load Outlook first thing in the morning?

There were several people who were using real time rendering and textures to simulate what their models would look like while creating the parts. As the surface detail was high, the 3D load was bigger and they were pushing their existing graphics cards, which were a few years old, very hard.

Advertisement
Advertisement

From one generation to the next, raw graphics card performance generally increased by 20% to 50%. It varies because the way the CAD software is written can limit the amount of graphics throughput that can be sent to the card. We call this “CPU limited” software. This effectively means the speed of the CPU affects the maximum performance you can get out of the graphics card. However, more modern graphics cards are able to take some of the load off the CPU and we were able to demonstrate this to one CAD user, who produced a memory stick with his own data on. There can be no doubting the performance if you test it with what you do every day and he even switched on the materials and bump effects and had no slow down.

I talked through loading files off a network with two people. This is a good way to centrally store and back up data, but it soon became clear that many people had lots of mapped drives to multiple servers and Windows needed to update these periodically. This of course means using up valuable resources, so it’s a case of less is more.

Loading other applications before you load your CAD applications can cause lots of problems with memory usage. In a quick sample of the 150 attendees over 100 loaded Microsoft Outlook first thing in the morning before any other application. When this happens Outlook sees all of the free memory and uses it to try and make it go faster. It does not know that you are then going to load a CAD application that also wants lots of RAM. Every so often Outlook checks to see if there are any new mails which also takes resources. One solution is to use a web client for mail and reserve most of your RAM for your CAD model. Of course, most applications do have memory management but each does it their own way which often means they use it at the same time.

The last group of people were using laptops as their main work machine. There were many questions about how people could upgrade their graphics card in a laptop. I explained that most people who make laptops prefer you to buy a new one rather than upgrade part of the machine.

The next most raised issue was startup time. One solution is to use a nice new SSD (Solid State Disk). These have fast read times but are expensive to retrofit. Perhaps your next new laptop should have one. The other solution I’ve written about before is to remove some of the startup programs via msconfig (see the July/August edition of DEVELOP3D). One customer then appeared with their laptop and we removed three programs out of twelve with no ill effects.

In most of these cases, sorting out one problem component will remove the bottleneck in the whole system. Unfortunately it’s not always immediately obvious what needs to be done to make things run smoother. I wish I could say the same about the traffic jam on the way home. Of course I’d understand if people had been injured, but why do we have to waste everyone’s time and money trying to apportion blame for minor car accidents? Bah humbug!

{encode=”robert.jamieson@amd.com ” title=”Rob Jamieson”} is a marketing manager at AMD. His thinking behind traffic jams versus cost is entirely rational. Just don’t get him started on speed cameras. This article is his own opinion and may not represent AMD’s positions, strategies or opinions.

Rob Jamieson helps you get from A-to-B quicker by removing compute bottlenecks


Leave a comment