GPU Computing Labs

  Assignment mini-project 2026

• Choose one/two basic operations that you will apply on 1 or 2 arrays
• addition - multiplication - mul-add - division - sqrt - max  - bitwise operation (and) - ...
• Choose one/two datatype: integer, float, double
• Choose and compare either two basic operations (e.g. compare multiplication and division) or two different datatypes (e.g. compare float and double).
  
• 1. Measure the computational and memory performance in function of the array size (automate this!)
  
• performance will converge to its maximum
  
• it will be memory-bound, so constrained by the memory bandwidth
  
• 2. Let the kernel execute on multiple elements
  
• There are multiple ways to select which elements each thread will do: test at least 2 ways (an efficient one and a non-efficient one).
• In the document of lesson 2, you'll find the 2 major ways to do this.
  
• The number of elements is a parameter
• Plot the performance in function of this parameter
  
• 3. Artificially increase the compute intensity by doing more of the same operation on the same data
• Add a loop over the operation and/or use the idea of a pseudo-random number generator (see this code)
• The compute intensity will depend on the loop count which you specify as a parameter
• By doing so, the computational performance will increase as is demonstrated by the roofline model
• Draw the roofline model (which is hardware dependent) for the operation and the GPU
• 4. Each kernel (work item) uses the data of the whole work group.
  
• Compare a version that works with the data in global memory with a version that copies the data to local memory
• See this scheme
  
• Specify which GPU you are using, its parameters (generation, #multiprocessors, #scalar processors on each multiprocessor, frequency, ...) and calculate the theoretical peak performance
• Several characteristics can be queried with cuda.
• With the compute capability you know of which generation your gpu is (see lesson 3 or wikipedia)
• Use the benchmark app of www.gpuperformance.org to empirically measure the peak performance (computational and memory)
  
• By clicking on the data you'll see the measurements (or roofline)
• Compare the performances (theoretical, benchmark, kernels) and try to explain them
• Hint: check 'Add a new project to cmake' on the documentation page.
• Consider the applied rules for the use of AI.
  

---

- Back to top -