CS501 GDB Solution & Discussion Last Date:19-02-2015
|GDB Announcement||Dated: Feb 12, 15|
The activity of Graded Discussion Board (GDB) will be conducted on February 18, 2015. Students will be able to post their comments for 48 hours only.
“As memory subsystem is one of the most important components of computer and the overall speed of the computer can be improved by improving the performance of memory sub-system. The performance of the memory sub-system depends on two characteristics: Bandwidth and Latency. But, there are often subtle tradeoffs between latency and bandwidth as giving more emphasis on one characteristic can have an adverse effect on other.”
Being a computer architect, you want to design a memory sub-system for Graphics Processing Unit (GPU). While designing, you will focus on which of the above characteristics more and why?
Give proper justification in support of your answer.
Try to provide precise and to the point comments avoiding irrelevant details. For any query, please send your emails at CS501@vu.edu.pk
+ http://bit.ly/vucodes (Link for Assignments, GDBs & Online Quizzes Solution)
+ http://bit.ly/papersvu (Link for Past Papers, Solved MCQs, Short Notes & More)+ Click Here to Search (Looking For something at vustudents.ning.com?) + Click Here To Join (Our facebook study Group)
Please Discuss here about this GDB.Thanks
Our main purpose here discussion not just Solution
We are here with you hands in hands to facilitate your learning and do not appreciate the idea of copying or replicating solutions.
Latency is the time between a given task is initiated and the task is beginning to execute. In terms of data-transfer, the latency is the time between the instructions to transfer is given and the data-transfer actually commence. This delay may occur due to instruction decoding, memory latency, waiting for bus-access and other causes. The latency may have a great impact on performance when transferring small amounts of data, since the latency is a constant start-up cost, which must be paid regardless of how many items are to be transferred
Bandwidth is the amount of data which can be transfer in a given amount of time. This property has great impact on the performance of a graphic processor since all data which shall be used in the computation must be copied to the graphics processor. To measure the bandwidth one has to take into account the latency, since each transfer will have a period of time before the actual transfer commence. The formula for calculating the bandwidth is given as a function of size, time and latency.
bandwidth =size/(time - latency)
From Latency and bandwidth concepts you can make some lines for GDB
If you want to study more about then follow this link Latency and Bandwidth Impact on GPU-systems
Dear check this link http://www.idi.ntnu.no/~elster/master-studs/runejoho/ms-proj-gpgpu-...
Solving bandwidth is easier than solving latency. To solve bandwidth, more pipes are added. For example, in early analog modems it was possible to increase bandwidth by bonding two or more modems. In fact, ISDN achieves 128K of bandwidth by bonding two 64K channels using a datalink protocol called multilink-ppp.
Bandwidth and latency are connected. If the bandwidth is saturated then congestion occurs and latency is increased. However, if the bandwidth of a circuit is not at peak, the latency will not decrease. Bandwidth can always be increased but latency cannot be decreased. Latency is the function of the electrical characteristics of the circuit.
Designing GPU-Accelerated Applications
GPUs hold tremendous performance potential and have already shown impressive speedups in many applications. Yet, realizing that potential might be challenging. In this section we offer a few simple guidelines to help designing GPU-accelerated applications.
Accelerate performance hotspots.
Usually GPUs are used to run only a subset of application code It is thus instructive to estimate the contribution π of the code intended for acceleration in the applications execution time, to guarantee that making it faster will have a tangible effect on the overall performance.
Move more computations to GPUs.
GPUs are not as efficient as CPUs when running control flow-intensive workloads. However, longer kernels with some control flow ported to a GPU are often preferable over shorter kernels controlled from CPU code.
Stay inside GPU memory. Memory capacity of discrete GPUs reaches 12GB in high-end systems. However if the working set of a GPU kernel is too large, as is often the case for stream processing applications, for example, the performance becomes limited by the throughput of CPU-GPU data transfers. Achieving performance improvement on GPUs in such applications is possible for compute-intensive workloads with high compute-to-memory access ratio.
Conform to hardware parallelism hierarchy.
Ample parallelism in the computing algorithm is an obvious requirement for successful acceleration on GPUs. However having thousands of independent tasks alone might not be sufficient to achieve high performance. It is crucial to design a parallel algorithm, which maps well onto the GPU hierarchical hardware parallelism.
Understand memory performance. To feed their voracious appetites for data, high-end GPUs employ a massively parallel memory interface, which offers high bandwidth for local access by GPU code. The GPU memory subsystem often becomes the main bottleneck, however, and it is useful to estimate application performance limits imposed by the memory demands of the parallel algorithm itself.
May be its helpful for u all guys
main es GDB ka matlab sahe tareka se samjha nai koi thora clear to kar do es ko...
LMS open nahi hoa raha :(
Is ka yeh issue purana hai jab bhi load parta hai LMS baith jata ha junab. Aaj to GDB or quiz ka bhi last day hai. I think tomorrow will be allowed as last day. What did u think guys?
dear GDB ka kuch idea he send kar do
plzzz solution upload kar dain time kam rah gaya hai..