Hello again guys. Im struggle to understand the results from the profiler of two kernel versions (one with unroll factor of 128 and another with 32) The 32 unroll factor outperforms the 128 factor by 5 seconds for an input matrix of 20000 x 1000. Stats are: 32 | 128 Activity: 96% | 25% Memory(global) BW: 15182 MB/s | 11885 MB/s Kernel Clock Freq: 244 MHz | 185 MHZ Stall %: 14,49% | 15,1 % I don't know what is happening because the stall increases while the best version has better memory bw..

I have explained the math behind saturating memory bandwidth here (http://www.alteraforum.com/forum/showthread.php?t=57099&p=232613#post232613). An unroll factor of 32 is likely more than enough to fully saturate the memory bandwidth. Anything more than that will result in a lot of memory contention and lowered performance. You are losing performance due to two reasons: lowered operating frequency, and lowered memory bandwidth due to higher amount of contention. Trying to decipher the numbers reported by the profiler generally results in more confusion, and I have encountered many cases in which the numbers simply do not make sense and contradict each other or the timing results. My recommendation is not to rely too much on those numbers.

--- Quote Start --- I have explained the math behind saturating memory bandwidth here (http://www.alteraforum.com/forum/showthread.php?t=57099&p=232613#post232613). An unroll factor of 32 is likely more than enough to fully saturate the memory bandwidth. Anything more than that will result in a lot of memory contention and lowered performance. You are losing performance due to two reasons: lowered operating frequency, and lowered memory bandwidth due to higher amount of contention. Trying to decipher the numbers reported by the profiler generally results in more confusion, and I have encountered many cases in which the numbers simply do not make sense and contradict each other or the timing results. My recommendation is not to rely too much on those numbers. --- Quote End --- Thanks mate! HRZ you're my salvation ;) My doubt was because of the stall parameter, it would make sense that were more high because of the memory contention...

--- Quote Start --- My doubt was because of the stall parameter, it would make sense that were more high because of the memory contention... --- Quote End --- Yes, that is also exactly what I would expect to see here. However, I have seen so many cases where the stall percentage simply does not make sense and hence, eventually stopped relying on the profiler. The only numbers that generally seem to make sense are the reported memory bandwidth and the burst size.

--- Quote Start --- Yes, that is also exactly what I would expect to see here. However, I have seen so many cases where the stall percentage simply does not make sense and hence, eventually stopped relying on the profiler. The only numbers that generally seem to make sense are the reported memory bandwidth and the burst size. --- Quote End --- Case solved then, thanks again HRZ B)

--- Quote Start --- Case solved then, thanks again HRZ B) --- Quote End --- Sorry for reviving this topic, but how can i calculate the bandwidth value ? I can't get the right value from the profiler info in the unrolled loop. Also i have a latency of 144 LSU for that read at global memory. http://i64.tinypic.com/o6wr2b.png

Kernel performance with profiling | Altera Community

15 Replies

Altera_Forum
Honored Contributor
8 years ago
--- Quote Start ---
If you are talking about the number reported by the profiler, I think it is in 32-bit words, not bytes.
--- Quote End ---

On profiler reports says that the max burst size is 16.. but 16 bits, bytes, number of requests?

bandwidth
Burst Size
The average burst size of the memory operation.If the memory system does not support burst mode (for example, on-chip RAM), no burst information will be available.

Average Burst Size=7.6(Max Burst=16)

Global memory
Altera_Forum
Honored Contributor
8 years ago
Since the maximum read/write size per kernel clock cycle per DDR3/4 memory bank is 512-bits (standard value hardcoded in the BSP), for a total of 1024 bits for two banks, I have so far assumed that a burst size of 16 equals a read/write size of 1024 bits, hence the unit being 32-bit words. I could be wrong, though.
Altera_Forum
Honored Contributor
8 years ago
--- Quote Start ---
Since the maximum read/write size per kernel clock cycle per DDR3/4 memory bank is 512-bits (standard value hardcoded in the BSP), for a total of 1024 bits for two banks, I have so far assumed that a burst size of 16 equals a read/write size of 1024 bits, hence the unit being 32-bit words. I could be wrong, though.
--- Quote End ---

Thats seems logic :) Last question (for now i guess), im using the board de5_net (http://www.terasic.com.tw/cgi-bin/page/archive.pl?language=english&categoryno=158&no=526&partno=2)

However, when i go to the BSP board spec, it says that i only have 2 GB for each bank (im assuming each dimm socket its a bank).. My question is, my frequency memory clock for DDR3 its 933 MHz?
Altera_Forum
Honored Contributor
8 years ago
The standard DE5-Net board has two 2GB DDR3 banks, each running at 1600 MHz (800 MHz double datarate). There is another variation of the board with two 4GB DDR3 banks, but I am not sure how fast the memory is (probably the same speed as the 2GB one). The first variation must be the board you have.
ADua0
Occasional Contributor
6 years ago
I am asking this question because I think this might be related. In profiler report I get average read burst size of 1 and average write burst size of 1 with optimal possible of 16. I am not sure how to utilize that burst size . Any suggestion to achieve that will really be helpful.
Just to give more information, I am reading 16 data at a time and storing in shift register

Forum Discussion

Kernel performance with profiling

15 Replies

Recent Discussions

recovery timing issue

error in JTAG server (error code 35) and autodetect (unable to scan device chain) Quartus 18.1

Quartus/Signaltap complains about wrong version

License issue

No access to the Self Service Licensing Center (SSLC)