You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
<pclass="lead"><b>Accel-Sim v1.3.0 and AccelWattch v1.0 have officially been released!</b><br>
96
+
<pclass="lead"><b>Accel-Sim v1.2.0 and AccelWattch v1.0 have officially been released!</b><br>
97
97
Accel-Sim is a simulation framework for simulating and
98
98
validating programmable accelerators like GPUs. For full details,
99
99
please see our recent <ahref="https://people.ece.ubc.ca/~aamodt/publications/papers/accelsim.isca2020.pdf">ISCA 2020 paper</a> and download slides from <ahref="ISCA2020-presentation-v3.0.pptx"> here</a>.<br><br>
100
100
101
-
AccelWattch is a power modeling framework that is extensively validated for modern GPUs. AccelWattch is highly accurate for NVIDIA Volta GPUs and enables reliable design space exploration. Please see our recent <ahref="http://paragon.cs.northwestern.edu/papers/2021-MICRO-AccelWattch-Kandiah.pdf">MICRO 2021 paper</a> and download slides from <ahref="https://drive.google.com/drive/folders/1FX3G4B7Whcy4gCEcGS6Am9adMpks9CY3?usp=sharing"> here</a>.<br><br>
101
+
AccelWattch is a power modeling framework that is extensively validated for modern GPUsand enables reliable design space exploration. Please see our recent <ahref="http://paragon.cs.northwestern.edu/papers/2021-MICRO-AccelWattch-Kandiah.pdf">MICRO 2021 paper</a> and download slides from <ahref="http://paragon.cs.northwestern.edu/talks/2021-MICRO-AccelWattch-Kandiah-slides.pptx"> here</a>.<br><br>
102
102
103
103
To keep you up-to-date with the recent news on Accel-Sim and AccelWattch, please join our Google group <a
<li> Introduction: The Accel-Sim ISCA 2020 paper [<ahref="https://www.iscaconf.org/isca2020/papers/466100a473.pdf">paper</a>, <ahref="ISCA2020-presentation-v3.0.pptx">slides</a>, <ahref="https://drive.google.com/drive/folders/1Q4-y6QTzS_1JoRmTUV31QKpOES9ZY8oG?usp=sharing">video</a>]</li>
@@ -218,17 +218,48 @@ <h3>Manual</h3>
218
218
<li> Original GPGPU-Sim 3.x manual [<ahref="http://gpgpu-sim.org/manual/index.php/Main_Page">manual</a>, <ahref="http://www.gpgpu-sim.org/micro2012-tutorial/">slides</a>, <ahref="https://www.youtube.com/channel/UCMZLxSL7Ibn6uCvwdZcGqFQ/videos">tutorial videos</a>] </li>
<li> The AccelWattch MICRO 2021 paper [<ahref="http://paragon.cs.northwestern.edu/papers/2021-MICRO-AccelWattch-Kandiah.pdf">paper</a>, <ahref="https://drive.google.com/drive/folders/1FX3G4B7Whcy4gCEcGS6Am9adMpks9CY3?usp=sharing">slides and video</a>]</li>
If you have any questions about Accel-Sim and AccelWattch, please feel free to join our <ahref="https://groups.google.com/forum/#!forum/accel-sim">Google group</a>.
222
+
If you have any questions about Accel-Sim, please feel free to join our <ahref="https://groups.google.com/forum/#!forum/accel-sim">Google group</a>.
223
+
<hr>
224
+
<br>
225
+
226
+
<h3>AccelWattch Overview</h3>
227
+
<aname="overview"></a>
228
+
AccelWattch estimates the constant and static power consumption of a GPU architecture using analytic modeling and hardware power measurements.
229
+
<ol>
230
+
<li><b>Constant Power Model</b>: Accounts for the power consumed by components such as the GPU fans and peripheral circuitry in the presence of DVFS in modern GPUs.</li>
231
+
<li><b>Static Power Model</b>: Accounts for the chip static power in the presence of power gating, thread divergence, intra-warp functional unit overlap, and variable SM occupancy.</li>
232
+
</ol>
233
+
AccelWattch uses microbenchmarking and quadratic programming for dynamic power modeling. The microbenchmarks selectively stress all GPU hardware components and the quadratic programming solver tunes the weight of each component to match hardware power measurements. To model the power consumption of a kernel, AccelWattch couples the tuned per-component power with per-component activity factors obtained from the kernel's execution.
234
+
Depending on the AccelWattch configuration, activity factors can come from:
235
+
<ol>
236
+
<li><b>AccelWattch SASS SIM</b>: Native ISA performance simulation</li>
237
+
<li><b>AccelWattch PTX SIM</b>: Virtual ISA performance simulation</li>
238
+
<li><b>AccelWattch HW</b>: Hardware performance counters from execution on real silicon</li>
239
+
<li><b>AccelWattch HYBRID</b>: Hardware performance counters for some components; performance simulation for others</li>
Our <ahref="http://paragon.cs.northwestern.edu/talks/2021-MICRO-AccelWattch-Kandiah-talk.mp4">MICRO 2021 video</a> presents the AccelWattch framework in more detail.
247
+
<br><br>
248
+
<hr>
249
+
250
+
<h3>AccelWattch Manual</h3>
251
+
<aname="manual"></a>
252
+
<ul>
253
+
<li> Introduction: The AccelWattch MICRO 2021 paper [<ahref="http://paragon.cs.northwestern.edu/papers/2021-MICRO-AccelWattch-Kandiah.pdf">paper</a>, <ahref="http://paragon.cs.northwestern.edu/talks/2021-MICRO-AccelWattch-Kandiah-slides.pptx">slides</a>, <ahref="http://paragon.cs.northwestern.edu/talks/2021-MICRO-AccelWattch-Kandiah-talk.mp4">video</a>]</li>
254
+
<li> Beginner guide and how to use: <ahref="https://github.com/accel-sim/accel-sim-framework/blob/release-accelwattch/README.md"> AccelWattch beginner manual</a>
0 commit comments