Forty-Second File Opens on the Design Floor
Baxter and Loft Architecture has 14 design workstations, each about four years old: a quad-core i5, 16 GB of RAM, a 1 TB 7200 rpm hard disk, and Intel integrated graphics. Designers work with BIM models and 2 to 3 GB assemblies and report 40-second file opens, the pointer freezing for several seconds when switching between a model and a drawing, and near-constant disk activity. Task Manager during a complaint shows CPU at 35 percent, memory at 96 percent with heavy page file writes, and a disk queue that stays above 3. Office is the 32-bit version. Management has approved $900 per machine for upgrades this quarter and wants the largest productivity gain per dollar. The IT contractor has proposed spending all of it on 32 GB RAM kits for every machine; one designer insists the real problem is graphics and wants new GPUs.
1. Interpret the measurements. Which resource is the bottleneck, and what do the symptoms rule out?
Read the worked answer
Memory is the binding constraint, and the disk symptoms are its consequence rather than an independent fault.
At 96 percent memory use the working set no longer fits in RAM, so Windows compresses pages and then writes them to the page file. Every one of those faults is a disk operation, and on a 7200 rpm drive the latency is milliseconds per fault, which is exactly what produces multi-second freezes when switching applications and 40-second file opens. The sustained disk queue above 3 confirms requests are piling up faster than the spindle can serve them.
CPU at 35 percent is the decisive clue: the processor is idle, waiting on I/O and page faults. Adding cores or a faster CPU would change nothing. GPU is not implicated either, because integrated graphics degrade frame rate while orbiting a model or redrawing a viewport, not the time to open a file from storage; the designer is describing a different symptom class. Confirm before spending by watching Resource Monitor for hard faults per second and committed bytes versus the commit limit, and by checking whether the freezes cluster after several applications have been open for a while.
2. Allocate the $900 per machine. Rank the upgrades and defend the order, including what to do about 32-bit Office.
Read the worked answer
Rank by pain removed per dollar.
First, a 1 TB SSD (roughly $80 to $140 depending on interface). This removes the latency penalty on every page fault and every model load, and it is the change users feel immediately. Budget a little labor for cloning or reimaging, because a bad clone wastes the whole purchase.
Second, RAM from 16 GB to 32 GB (roughly $60 to $120 for DDR4). Do it on every machine as well, or at minimum on the five or six heaviest users, because the page file pressure will otherwise return as soon as a large model and a browser are open together. But confirm the machines run 64-bit Windows first; adding RAM to a 32-bit OS is money burned.
Third, spend nothing on GPUs at this stage. The remaining budget is better treated as labor: replace 32-bit Office with the 64-bit build. A 32-bit process has roughly a 2 GB virtual address space (4 GB in large-address-aware builds), so it cannot map a 3 GB model no matter how much physical RAM exists. That is a free download and a reinstall, and skipping it would partially nullify the RAM purchase. Leave a contingency for one failed drive or power supply.
3. Explain to the designer why the GPU is not the first purchase, and describe when it would become the right one.
Read the worked answer
Map each symptom to the resource that produces it. GPU-bound behaviour looks like a low frame rate while orbiting a model, laggy material and shadow previews, slow rendering, or display driver resets. Memory and storage behaviour looks like the app hanging on open and on window switches, with the disk light solid, which is what the designers report. Buying a GPU first would leave the paging untouched and the freezes unchanged, and the designer would conclude the purchase failed.
The GPU becomes the correct next purchase once the SSD and RAM are in and the remaining complaint is viewport responsiveness. Even then, scope it: only the two or three people producing client walkthroughs or renders need a certified workstation card with the vendor's ISV driver, and the machine must have the power supply and slot to support it.
There is also a portfolio argument worth putting to management: these machines are four years old and about to receive their second round of upgrades. Price the downtime per designer-hour, multiply by 14, and compare against a planned lifecycle replacement. If the number is large, a phased refresh beats a third retrofit.