Contact submission

[markotoplak]

Submitter: Oleksandr Gakh
It is rather hardware question than technical problem. I have noticed that I can analyze spectral data much faster on Mac computer (Xeon processor, RAM 64GB) then on PC (Xeon processor, RAM 128GB). Thus, is there any specific requirements for hardware set up for faster performance (processor, RAM, video card configuration etc). Thank you

[markotoplak]

Could you share some measurements with us? Are there any particular operations that are slower on Windows? What did you use for installation? Packages from the web site or manual/pip installation?

Whenever I saw such differences before, it was due to different libraries used for matrix operations (different libraries within numpy; they are system dependent). The difference between a slow and a fast matrix library can ~10x for matrix multiplication, but I think, that nowdays all numpy distributions are well optimized.

[AlexGakh]

Hi Marko, I do process the data from FTIR microscope and on PC it is slow from the beginning, even when I download the data, and each step of following analysis (visualization, baseline normalization etc). I have monitored the usage of memory (the PC has 128GB) and it has been used of 30-40% of full capacity. For installation I have used Quasar-0.7.1-Miniconda-x86_64.exe file and I just follow instructions, whale on Mac it was installed an older version of Orange through anaconda navigator. So, if there is a problems with libraries how I can check and change it in order to optimize the analysis. Please, let me know if you need more information, thank you, Alex.

[AlexGakh]

One more question is Quasar better works with CUDA or OpenGL graphic cards? The reason why I am asking that the PC which works slow have the NVIDIA graphic card (Cuda) while the Mac have FirePro (OpenGL).

[stuart-cls]

For the GPU, we don't have any GPU-accelerated calculations that I'm aware of at this time.

For the CPU, I think you should be using a recent Intel MKL (vector math library) since you used the Quasar installer. You can confirm by running "Quasar Command Prompt" and entering conda list mkl. Mine shows mkl with version 2019.4. I think MKL is the fastest library on Windows.

[AlexGakh]

Thank you Stuart, It narrows down the problem what we have with slow calculation, I think it is mayo security stuff which they put on our PC computers (MAC somehow bypass it). Thanks, Alex From: Stuart Read [mailto:notifications@github.com] Sent: Tuesday, May 5, 2020 10:11 AM To: Quasars/quasar.codes Cc: Gakh, Oleksandr, Ph.D.; Comment Subject: [EXTERNAL] Re: [Quasars/quasar.codes] Contact submission (#19) For the GPU, we don't have any GPU-accelerated calculations that I'm aware of at this time. For the CPU, I think you should be using a recent Intel MKL (vector math library) since you used the Quasar installer. You can confirm by running "Quasar Command Prompt" and entering conda list mkl. Mine shows mkl with version 2019.4. I think MKL is the fastest library on Windows. — You are receiving this because you commented. Reply to this email directly, view it on GitHub<#19 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/APMEYGGFALKIKC3RJTAXDS3RQAT77ANCNFSM4MQDWTSA>.