<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
</head>
<body>
<div dir="ltr">The figure did not clearly say all cores share L3. Instead, we should look at p.16 of <a href="https://www.redbooks.ibm.com/redpapers/pdfs/redp5472.pdf">https://www.redbooks.ibm.com/redpapers/pdfs/redp5472.pdf</a>
<div><br>
</div>
<div>"The POWER9 chip contains two memory controllers, PCIe Gen4 I/O controllers, and an interconnection system that connects all components within the chip at 7 TBps. Each core has 256 KB of L2 cache, and all cores share 120 MB of L3 embedded DRAM (eDRAM)."<br>
</div>
<div>
<div>
<div dir="ltr" class="gmail_signature" data-smartmail="gmail_signature">
<div dir="ltr">--Junchao Zhang</div>
</div>
</div>
<br>
</div>
</div>
<br>
<div class="gmail_quote">
<div dir="ltr" class="gmail_attr">On Mon, Sep 23, 2019 at 11:58 AM Mills, Richard Tran via petsc-dev <<a href="mailto:petsc-dev@mcs.anl.gov">petsc-dev@mcs.anl.gov</a>> wrote:<br>
</div>
<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<div bgcolor="#FFFFFF">L3 and L2 are shared between cores, actually. See the attached 'lstopo' PDF output from a Summit compute node to see an illustration of the node layout.<br>
<br>
--Richard<br>
<br>
<div>On 9/23/19 9:01 AM, Zhang, Junchao via petsc-dev wrote:<br>
</div>
<blockquote type="cite">
<div dir="ltr">
<div><font face="arial, sans-serif">I also did OpenMP stream test and then I found mismatch between OpenMPI and MPI. That reminded me a subtle issue on summit: pair of cores share L2 cache. One has to place MPI ranks to different pairs to get best bandwidth.
See different bindings</font></div>
<div><a href="https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n2c21g3r12d1b21l0=" target="_blank">https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n2c21g3r12d1b21l0=</a> and <a href="https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n2c21g3r12d1b22l0=" target="_blank">https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n2c21g3r12d1b22l0=</a>.
Note each node has 21 cores. I assume that means 11 pairs. The new results are below. They match with we what I got from OpenMPI. The bandwidth is almost doubled from 1 to 2 cores per socket. IBM document also says each socket has two memory controllers. I
could not find the core-memory controller affinity info. I tried different bindings but did not find huge difference.</div>
<div> </div>
<div><font face="monospace">#Ranks Rate (MB/s) Ratio over 2 ranks<br>
1 29229.8 -<br>
2 59091.0 1.0<br>
4 112260.7 1.9<br>
6 159852.8 2.7<br>
8 194351.7 3.3<br>
10 215841.0 3.7<br>
12 232316.6 3.9<br>
14 244615.7 4.1<br>
16 254450.8 4.3<br>
18 262185.7 4.4<br>
20 267181.0 4.5<br>
22 270290.4 4.6<br>
<font color="#ff0000">24 221944.9 3.8<br>
26 238302.8 4.0</font></font><font color="#ff0000"><br>
</font></div>
<div><br>
</div>
<div><br>
</div>
<div>
<div dir="ltr">
<div dir="ltr">--Junchao Zhang</div>
</div>
</div>
<br>
</div>
<br>
<div class="gmail_quote">
<div dir="ltr" class="gmail_attr">On Sun, Sep 22, 2019 at 6:04 PM Smith, Barry F. <<a href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a>> wrote:<br>
</div>
<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<br>
Junchao,<br>
<br>
For completeness could you please run with a single core? But leave the ratio as you have with over 2 ranks since that is the correct model.<br>
<br>
Thanks<br>
<br>
Barry<br>
<br>
<br>
> On Sep 22, 2019, at 11:14 AM, Zhang, Junchao <<a href="mailto:jczhang@mcs.anl.gov" target="_blank">jczhang@mcs.anl.gov</a>> wrote:<br>
> <br>
> I did stream test on Summit. I used the MPI version from petsc, but largely increased the array size N since one socket of Summit has 120MB L3 cache. I used MPI version since it was easy for me to distribute ranks evenly to the two sockets.
<br>
> The result matches with data released by OLCF (see attached figure) and data given by Jed. We can see the bandwidth saturates around 24 ranks.<br>
> <br>
> #Ranks Rate (MB/s) Ratio over 2 ranks<br>
> ------------------------------------------<br>
> 2 59012.2834 1.00<br>
> 4 70959.1475 1.20<br>
> 6 106639.9837 1.81<br>
> 8 138638.6929 2.35<br>
> 10 171125.0873 2.90<br>
> 12 196162.5197 3.32<br>
> 14 215272.7810 3.65<br>
> 16 229562.4040 3.89<br>
> 18 242587.4913 4.11<br>
> 20 251057.1731 4.25<br>
> 22 258569.7794 4.38<br>
> 24 265443.2924 4.50<br>
> 26 266562.7872 4.52<br>
> 28 267043.6367 4.53<br>
> 30 266833.7212 4.52<br>
> 32 267183.8474 4.53<br>
> <br>
> On Sat, Sep 21, 2019 at 11:24 PM Smith, Barry F. <<a href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a>> wrote:<br>
> <br>
> Junchao could try the PETSc (and non-PETSc) streams tests on the machine. <br>
> <br>
> There are a few differences, compiler, the reported results are with OpenMP, different number of cores but yes the performance is a bit low. For DOE that is great, makes GPUs look better :-)<br>
> <br>
> <br>
> > On Sep 21, 2019, at 11:11 PM, Jed Brown <<a href="mailto:jed@jedbrown.org" target="_blank">jed@jedbrown.org</a>> wrote:<br>
> > <br>
> > For an AIJ matrix with 32-bit integers, this is 1 flops/6 bytes, or 165<br>
> > GB/s for the node for the best case (42 ranks).<br>
> > <br>
> > My understanding is that these systems have 8 channels of DDR4-2666 per<br>
> > socket, which is ~340 GB/s of theoretical bandwidth on a 2-socket<br>
> > system, and 270 GB/s STREAM Triad according to this post<br>
> > <br>
> > <a href="https://openpowerblog.wordpress.com/2018/07/19/epyc-skylake-vs-power9-stream-memory-bandwidth-comparison-via-zaius-barreleye-g2/" rel="noreferrer" target="_blank">
https://openpowerblog.wordpress.com/2018/07/19/epyc-skylake-vs-power9-stream-memory-bandwidth-comparison-via-zaius-barreleye-g2/</a><br>
> > <br>
> > Is this 60% of Triad the best we can get for SpMV?<br>
> > <br>
> > "Zhang, Junchao via petsc-dev" <<a href="mailto:petsc-dev@mcs.anl.gov" target="_blank">petsc-dev@mcs.anl.gov</a>> writes:<br>
> > <br>
> >> 42 cores have better performance.<br>
> >> <br>
> >> 36 MPI ranks<br>
> >> MatMult 100 1.0 2.2435e+00 1.0 1.75e+09 1.3 2.9e+04 4.5e+04 0.0e+00 6 99 97 28 0 100100100100 0 25145 0 0 0.00e+00 0 0.00e+00 0<br>
> >> VecScatterBegin 100 1.0 2.1869e-02 3.3 0.00e+00 0.0 2.9e+04 4.5e+04 0.0e+00 0 0 97 28 0 1 0100100 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >> VecScatterEnd 100 1.0 7.9205e-0152.6 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 1 0 0 0 0 22 0 0 0 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >> <br>
> >> --Junchao Zhang<br>
> >> <br>
> >> <br>
> >> On Sat, Sep 21, 2019 at 9:41 PM Smith, Barry F. <<a href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a><mailto:<a href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a>>> wrote:<br>
> >> <br>
> >> Junchao,<br>
> >> <br>
> >> Mark has a good point; could you also try for completeness the CPU with 36 cores and see if it is any better than the 42 core case?<br>
> >> <br>
> >> Barry<br>
> >> <br>
> >> So extrapolating about 20 nodes of the CPUs is equivalent to 1 node of the GPUs for the multiply for this problem size.<br>
> >> <br>
> >>> On Sep 21, 2019, at 6:40 PM, Mark Adams <<a href="mailto:mfadams@lbl.gov" target="_blank">mfadams@lbl.gov</a><mailto:<a href="mailto:mfadams@lbl.gov" target="_blank">mfadams@lbl.gov</a>>> wrote:<br>
> >>> <br>
> >>> I came up with 36 cores/node for CPU GAMG runs. The memory bus is pretty saturated at that point.<br>
> >>> <br>
> >>> On Sat, Sep 21, 2019 at 1:44 AM Zhang, Junchao via petsc-dev <<a href="mailto:petsc-dev@mcs.anl.gov" target="_blank">petsc-dev@mcs.anl.gov</a><mailto:<a href="mailto:petsc-dev@mcs.anl.gov" target="_blank">petsc-dev@mcs.anl.gov</a>>> wrote:<br>
> >>> Here are CPU version results on one node with 24 cores, 42 cores. Click the links for core layout.<br>
> >>> <br>
> >>> 24 MPI ranks, <a href="https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n6c4g1r14d1b21l0=" rel="noreferrer" target="_blank">
https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n6c4g1r14d1b21l0=</a><br>
> >>> MatMult 100 1.0 3.1431e+00 1.0 2.63e+09 1.2 1.9e+04 5.9e+04 0.0e+00 8 99 97 25 0 100100100100 0 17948 0 0 0.00e+00 0 0.00e+00 0<br>
> >>> VecScatterBegin 100 1.0 2.0583e-02 2.3 0.00e+00 0.0 1.9e+04 5.9e+04 0.0e+00 0 0 97 25 0 0 0100100 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>> VecScatterEnd 100 1.0 1.0639e+0050.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 2 0 0 0 0 19 0 0 0 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>> <br>
> >>> 42 MPI ranks, <a href="https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n6c7g1r17d1b21l0=" rel="noreferrer" target="_blank">
https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n6c7g1r17d1b21l0=</a><br>
> >>> MatMult 100 1.0 2.0519e+00 1.0 1.52e+09 1.3 3.5e+04 4.1e+04 0.0e+00 23 99 97 30 0 100100100100 0 27493 0 0 0.00e+00 0 0.00e+00 0<br>
> >>> VecScatterBegin 100 1.0 2.0971e-02 3.4 0.00e+00 0.0 3.5e+04 4.1e+04 0.0e+00 0 0 97 30 0 1 0100100 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>> VecScatterEnd 100 1.0 8.5184e-0162.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 6 0 0 0 0 24 0 0 0 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>> <br>
> >>> --Junchao Zhang<br>
> >>> <br>
> >>> <br>
> >>> On Fri, Sep 20, 2019 at 11:48 PM Smith, Barry F. <<a href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a><mailto:<a href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a>>> wrote:<br>
> >>> <br>
> >>> Junchao,<br>
> >>> <br>
> >>> Very interesting. For completeness please run also 24 and 42 CPUs without the GPUs. Note that the default layout for CPU cores is not good. You will want 3 cores on each socket then 12 on each.<br>
> >>> <br>
> >>> Thanks<br>
> >>> <br>
> >>> Barry<br>
> >>> <br>
> >>> Since Tim is one of our reviewers next week this is a very good test matrix :-)<br>
> >>> <br>
> >>> <br>
> >>>> On Sep 20, 2019, at 11:39 PM, Zhang, Junchao via petsc-dev <<a href="mailto:petsc-dev@mcs.anl.gov" target="_blank">petsc-dev@mcs.anl.gov</a><mailto:<a href="mailto:petsc-dev@mcs.anl.gov" target="_blank">petsc-dev@mcs.anl.gov</a>>> wrote:<br>
> >>>> <br>
> >>>> Click the links to visualize it.<br>
> >>>> <br>
> >>>> 6 ranks<br>
> >>>> <a href="https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n6c1g1r11d1b21l0=" rel="noreferrer" target="_blank">
https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n6c1g1r11d1b21l0=</a><br>
> >>>> jsrun -n 6 -a 1 -c 1 -g 1 -r 6 --latency_priority GPU-GPU --launch_distribution packed --bind packed:1 js_task_info ./ex900 -f HV15R.aij -mat_type aijcusparse -vec_type cuda -n 100 -log_view<br>
> >>>> <br>
> >>>> 24 ranks<br>
> >>>> <a href="https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n6c4g1r14d1b21l0=" rel="noreferrer" target="_blank">
https://jsrunvisualizer.olcf.ornl.gov/?s4f1o01n6c4g1r14d1b21l0=</a><br>
> >>>> jsrun -n 6 -a 4 -c 4 -g 1 -r 6 --latency_priority GPU-GPU --launch_distribution packed --bind packed:1 js_task_info ./ex900 -f HV15R.aij -mat_type aijcusparse -vec_type cuda -n 100 -log_view<br>
> >>>> <br>
> >>>> --Junchao Zhang<br>
> >>>> <br>
> >>>> <br>
> >>>> On Fri, Sep 20, 2019 at 11:34 PM Mills, Richard Tran via petsc-dev <<a href="mailto:petsc-dev@mcs.anl.gov" target="_blank">petsc-dev@mcs.anl.gov</a><mailto:<a href="mailto:petsc-dev@mcs.anl.gov" target="_blank">petsc-dev@mcs.anl.gov</a>>> wrote:<br>
> >>>> Junchao,<br>
> >>>> <br>
> >>>> Can you share your 'jsrun' command so that we can see how you are mapping things to resource sets?<br>
> >>>> <br>
> >>>> --Richard<br>
> >>>> <br>
> >>>> On 9/20/19 11:22 PM, Zhang, Junchao via petsc-dev wrote:<br>
> >>>>> I downloaded a sparse matrix (HV15R) from Florida Sparse Matrix Collection. Its size is about 2M x 2M. Then I ran the same MatMult 100 times on one node of Summit with -mat_type aijcusparse -vec_type cuda. I found MatMult was almost dominated by VecScatter
in this simple test. Using 6 MPI ranks + 6 GPUs, I found CUDA aware SF could improve performance. But if I enabled Multi-Process Service on Summit and used 24 ranks + 6 GPUs, I found CUDA aware SF hurt performance. I don't know why and have to profile it.
I will also collect data with multiple nodes. Are the matrix and tests proper?<br>
> >>>>> <br>
> >>>>> ------------------------------------------------------------------------------------------------------------------------<br>
> >>>>> Event Count Time (sec) Flop --- Global --- --- Stage ---- Total GPU - CpuToGpu - - GpuToCpu - GPU<br>
> >>>>> Max Ratio Max Ratio Max Ratio Mess AvgLen Reduct %T %F %M %L %R %T %F %M %L %R Mflop/s Mflop/s Count Size Count Size %F<br>
> >>>>> ---------------------------------------------------------------------------------------------------------------------------------------------------------------<br>
> >>>>> 6 MPI ranks (CPU version)<br>
> >>>>> MatMult 100 1.0 1.1895e+01 1.0 9.63e+09 1.1 2.8e+03 2.2e+05 0.0e+00 24 99 97 18 0 100100100100 0 4743 0 0 0.00e+00 0 0.00e+00 0<br>
> >>>>> VecScatterBegin 100 1.0 4.9145e-02 3.0 0.00e+00 0.0 2.8e+03 2.2e+05 0.0e+00 0 0 97 18 0 0 0100100 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>>>> VecScatterEnd 100 1.0 2.9441e+00133 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 3 0 0 0 0 13 0 0 0 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>>>> <br>
> >>>>> 6 MPI ranks + 6 GPUs + regular SF<br>
> >>>>> MatMult 100 1.0 1.7800e-01 1.0 9.66e+09 1.1 2.8e+03 2.2e+05 0.0e+00 0 99 97 18 0 100100100100 0 318057 3084009 100 1.02e+02 100 2.69e+02 100<br>
> >>>>> VecScatterBegin 100 1.0 1.2786e-01 1.3 0.00e+00 0.0 2.8e+03 2.2e+05 0.0e+00 0 0 97 18 0 64 0100100 0 0 0 0 0.00e+00 100 2.69e+02 0<br>
> >>>>> VecScatterEnd 100 1.0 6.2196e-02 3.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 22 0 0 0 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>>>> VecCUDACopyTo 100 1.0 1.0850e-02 2.3 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 5 0 0 0 0 0 0 100 1.02e+02 0 0.00e+00 0<br>
> >>>>> VecCopyFromSome 100 1.0 1.0263e-01 1.2 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 54 0 0 0 0 0 0 0 0.00e+00 100 2.69e+02 0<br>
> >>>>> <br>
> >>>>> 6 MPI ranks + 6 GPUs + CUDA-aware SF<br>
> >>>>> MatMult 100 1.0 1.1112e-01 1.0 9.66e+09 1.1 2.8e+03 2.2e+05 0.0e+00 1 99 97 18 0 100100100100 0 509496 3133521 0 0.00e+00 0 0.00e+00 100<br>
> >>>>> VecScatterBegin 100 1.0 7.9461e-02 1.1 0.00e+00 0.0 2.8e+03 2.2e+05 0.0e+00 1 0 97 18 0 70 0100100 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>>>> VecScatterEnd 100 1.0 2.2805e-02 1.5 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 17 0 0 0 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>>>> <br>
> >>>>> 24 MPI ranks + 6 GPUs + regular SF<br>
> >>>>> MatMult 100 1.0 1.1094e-01 1.0 2.63e+09 1.2 1.9e+04 5.9e+04 0.0e+00 1 99 97 25 0 100100100100 0 510337 951558 100 4.61e+01 100 6.72e+01 100<br>
> >>>>> VecScatterBegin 100 1.0 4.8966e-02 1.8 0.00e+00 0.0 1.9e+04 5.9e+04 0.0e+00 0 0 97 25 0 34 0100100 0 0 0 0 0.00e+00 100 6.72e+01 0<br>
> >>>>> VecScatterEnd 100 1.0 7.2969e-02 4.9 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 1 0 0 0 0 42 0 0 0 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>>>> VecCUDACopyTo 100 1.0 4.4487e-03 1.8 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 3 0 0 0 0 0 0 100 4.61e+01 0 0.00e+00 0<br>
> >>>>> VecCopyFromSome 100 1.0 4.3315e-02 1.9 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 29 0 0 0 0 0 0 0 0.00e+00 100 6.72e+01 0<br>
> >>>>> <br>
> >>>>> 24 MPI ranks + 6 GPUs + CUDA-aware SF<br>
> >>>>> MatMult 100 1.0 1.4597e-01 1.2 2.63e+09 1.2 1.9e+04 5.9e+04 0.0e+00 1 99 97 25 0 100100100100 0 387864 973391 0 0.00e+00 0 0.00e+00 100<br>
> >>>>> VecScatterBegin 100 1.0 6.4899e-02 2.9 0.00e+00 0.0 1.9e+04 5.9e+04 0.0e+00 1 0 97 25 0 35 0100100 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>>>> VecScatterEnd 100 1.0 1.1179e-01 4.1 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 1 0 0 0 0 48 0 0 0 0 0 0 0 0.00e+00 0 0.00e+00 0<br>
> >>>>> <br>
> >>>>> <br>
> >>>>> --Junchao Zhang<br>
> >>>> <br>
> >>> <br>
> <br>
> <SummitNode.png><br>
<br>
</blockquote>
</div>
</blockquote>
<br>
</div>
</blockquote>
</div>
</body>
</html>