Using a normal processors for HPC

Click For Summary

Discussion Overview

The discussion revolves around the feasibility of using a standard multicore PC as a high-performance computer (HPC) for numerical computations without relying on GPU technologies like CUDA or OpenCL. Participants explore the capabilities of such systems in handling heavy computational tasks and the implications for accuracy and error margins in simulation results.

Discussion Character

  • Exploratory, Technical explanation, Debate/contested

Main Points Raised

  • Some participants suggest that while a standard multicore PC can perform heavy numerical computations, it may not match the speed of GPU-enhanced hardware.
  • There is a question about whether programs can effectively utilize multiple cores or if they will operate as single-threaded processes.
  • One participant explains that the operating system typically assigns processes to cores, but certain specialized programs can run in parallel across multiple cores.
  • Another participant emphasizes that the ability to utilize multiple cores depends on whether the program is designed for multithreading; single-threaded programs will only utilize one core.
  • It is noted that modern cores can handle multiple threads simultaneously, which adds complexity to the understanding of core utilization.
  • A suggestion is made that writing custom programs in languages that support multithreading can allow for better performance on multicore processors, with synchronization mechanisms like semaphores being necessary for thread management.
  • OpenCL is mentioned as a tool that can facilitate the creation of parallel threads without requiring GPU hardware, and it is suggested that using OpenCL can make programs adaptable for future GPU use.

Areas of Agreement / Disagreement

Participants express differing views on the effectiveness of standard multicore PCs for HPC tasks, with some asserting that they can be used effectively while others argue that GPU-enhanced systems are superior. The discussion on multithreading and core utilization also reveals a lack of consensus on the best practices for maximizing performance.

Contextual Notes

The discussion does not resolve the complexities surrounding the definitions of cores and threads, nor does it clarify the specific conditions under which programs can effectively utilize multicore architectures.

hagopbul
Messages
397
Reaction score
45
Hello all:

Can you use current of shelf multicore PC as a high performance computer

Can it handle the job ?
Dose the outcome data tables and simulation results , is correct and with low error margins

With out gpu like cuda or opencl

Best
Hagop
 
Computer science news on Phys.org
Yes and no. Yes you can run heavy numerical computations on it but no GPU enhanced hardware still runs faster.
 
Can you utilities the multiple core or it will look like single core
 
hagopbul said:
Can you utilities the multiple core or it will look like single core
That entirely depends on the program you are running. The correct term is 'thread'. If your program is made to utilize multiple threads ('multithreading') then (by default) it might run on more than one core: if it is a single thread code then it will 'see' only one core.
The term 'core' is a bit misleading since these days a core might run multiple threads in the same time (depending on the core/CPU type).
 
If you write your own program you can write it in a language that allows you to define many independent threads that can be run in parallel. The operating system will run those threads on many cores as cores becomes available. Semaphores between the threads must synchronise the execution of the threads.

OpenCL makes it easier to generate the many parallel threads.
You do not need GPU hardware to benefit from the OpenCL software library. If you use Opencl when writing your program, it will optimise code to run on a multicore processor. If you later get GPU hardware, the same code will use the GPU and run faster again. You are future proofing your program by using OpenCL software.
 

Similar threads

  • · Replies 4 ·
Replies
4
Views
7K
  • · Replies 7 ·
Replies
7
Views
3K
  • · Replies 9 ·
Replies
9
Views
2K
Replies
3
Views
2K
  • · Replies 1 ·
Replies
1
Views
2K
  • · Replies 7 ·
Replies
7
Views
3K
  • · Replies 23 ·
Replies
23
Views
4K
  • · Replies 38 ·
2
Replies
38
Views
10K
  • · Replies 2 ·
Replies
2
Views
3K