- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I do convolutional neural networks with de1_soc FPGA,I use no num_compute_units and num_simd_work_items in kernel,the FPGA utilization like this:
+--------------------------------------------------------------------+
; Estimated Resource Usage Summary ;
+----------------------------------------+---------------------------+
; Resource + Usage ;
+----------------------------------------+---------------------------+
; Logic utilization ; 34% ;
; ALUTs ; 22% ;
; Dedicated logic registers ; 13% ;
; Memory blocks ; 16% ;
; DSP blocks ; 15% ;
+----------------------------------------+---------------------------;
and the work consume about 2750 ms.
then i use num_compute_units(4) to increase utilization。the FPGA utilization like this:
+--------------------------------------------------------------------+
; Estimated Resource Usage Summary ;
+----------------------------------------+---------------------------+
; Resource + Usage ;
+----------------------------------------+---------------------------+
; Logic utilization ; 75% ;
; ALUTs ; 48% ;
; Dedicated logic registers ; 32% ;
; Memory blocks ; 56% ;
; DSP blocks ; 61% ;
+----------------------------------------+---------------------------;
but the work consume about 3750 ms.
Why did I use num_compute_units to improve the utilization of fpga, but did not get the better results.
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
Thank you for raising this query, we best prefer you to post your query in the below forums for better communication:
https://forums.intel.com/s/
Regards,
Nikky Priya
Intel Developer Zone
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page