

        <cite>Bryce</cite>
      <time>0:00</time>
      <p>
          Okay,   but   buddy,   but   what   trained   Chat   GPT?
      </p>

        <cite>Conor</cite>
      <time>0:03</time>
      <p>
          Uh   that   the   compute   that   trains   NVIDIA   GPUs.
      </p>

        <cite>Bryce</cite>
      <time>0:07</time>
      <p>
          Yeah,   but   that   that   doesn&#39;t   affect   the   answers.   I&#39;m   telling   you,   that   that   large   language   model,   it   almost   certainly   knows   how   it   was   trained.   Do   you   know   the   address?   I   I   can&#39;t   get   the   address   right   now   because   Connor   and   I   are   in   the   middle   of   recording.
      </p>

        <cite>Conor</cite>
      <time>0:30</time>
      <p>
          Welcome   to   ADSP   the   podcast,   episode   166,   recorded   on   January   24th,   2024.   My   name   is   Connor,   and   today   I   quiz   my   co-host   Bryce   about   what   the   top   20   GPU   SDK&#39;s   libraries   and   tools   are.
      </p>

        <cite>Bryce</cite>
      <time>0:51</time>
      <p>
          Oh,   I   have   some   bad   news.   My   my   little   thing   that   my   microphones   go   in.   Look   at   this.   Come   on,   Rode.   What&#39;s   this   quality   control?   This   is   starting   to   fall   apart.
      </p>

        <cite>Conor</cite>
      <time>1:00</time>
      <p>
          It&#39;s   not   acceptable.
      </p>

        <cite>Bryce</cite>
      <time>1:02</time>
      <p>
          Yeah.   Oh,   I   wanna   speak   of   quality   control.   Speaking   of   quality   control.   This   is   potentially   relevant   content   for   the   episode.   So   you   know   about   how   that   that   door,   the   door   plug   fell   like   off   off   of   that   uh   that   Alaska   Airlines   flight   like   a   couple   weeks   ago.
      </p>

        <cite>Conor</cite>
      <time>1:21</time>
      <p>
          Oh   yeah,   yeah,   I   I   heard   about   that.
      </p>

        <cite>Bryce</cite>
      <time>1:24</time>
      <p>
          Yeah.   So   they   um   uh   there   were   some   reports   out   now   that   um   indicate   uh   what   the   root   of   the   problem   was.   And   the   root   of   the   problem   seems   to   be   that   the   the   four   bolts   that   like   were   supposed   to   hold   the   door   in   place   were   just   not   installed   at   all.   And   um   and   I&#39;ve   been   reading   because   I&#39;m   very   uh   tuned   into   the   aviation   world,   I&#39;ve   been   reading   some   some   things   about   this.   And   and   the   the   process   failure   that   led   to   uh   to   this   issue   is   actually   kind   of   interesting.   Um   so   so   they   they   thought   that   there   was   something   defective   with   the   door   plug,   uh,   because   there   were   a   lot   of   issues   from   the   people   who   provided   the   door   plug.   And   so   because   like   when   they   were   doing   their   QA   on   the   on   the   plane   before   they   shipped   it   out   of   the   factory,   they   um   they   they   noted   that,   hey,   we   gotta   investigate   this   door   plug.   We   think   the   door   plug&#39;s   a   problem.   So   they   made   a   note   on   the   record   to   go   check   the   door   plug,   and   then   they   went   and   checked   the   door   plug.   And   there&#39;s   a   note   on   the   record   that   said,   all   right,   we&#39;ve   we&#39;ve   taken   the   door   out   and   we&#39;ve   investigated   it,   and   now   we&#39;re   deciding   whether   or   not   we&#39;re   going   to   uh   uh   like   replace   the   door   or   not.   And   then   the   note   says,   and   if   we   do   replace   the   door   with   a   new   one,   if   we   we   get   rid   of   this   one   and   put   in   a   new   one,   we   will   put   a   note   in   the   system   for   the   door   to   be   rechecked,   to   do   QA   on   the   on   the   new   door   again.   And   see,   that&#39;s   where   the   mistake   was,   because   they   had   already   taken   the   door   off   of   the   plane.   And   so   because   they&#39;d   taken   the   door   off   of   the   plane,   they   needed   to   redo   the   QA   anyways,   regardless   of   whether   they   replaced   the   door.   They   needed   to   check   that   the   door   was   properly   reinstalled.   But   they   didn&#39;t   end   up   replacing   the   door.   And   because   they   didn&#39;t   end   up   replacing   the   door,   they   didn&#39;t   mark   that   the   door   needed   to   be   rechecked   by   QA.   And   because   of   that,   the   door   wasn&#39;t   rechecked   by   QA,   and   the   plane   shipped   out   of   the   factory   without   any   bolts   to   hold   the   door   in   place.   And   uh,   you   know,   uh   my   partner   being   a   process   engineer,   I   have   now   some   interest   in   the   field   of   uh   uh   process   and   quality   control   and   what   leads   to   the   root   cause   of   failures.   And   I   just   thought   that   this   was   quite   interesting.
      </p>

        <cite>Conor</cite>
      <time>3:41</time>
      <p>
          You   know,   I&#39;d   say   that   our   listeners   don&#39;t   find   it   interesting,   but   you   know,   we&#39;ve   le   we&#39;ve   recently   learned   that   I   don&#39;t   have   my   finger   on   the   pulse   of   what   our   listeners   do   and   do   not   find   interesting.   Well,   yeah,   hopefully   my   planes   that   I   get   on   don&#39;t   have   a   door   taken   off   and   then   put   back   on.   If   it   gets   taken   off,   let&#39;s   hope   it   gets   replaced.
      </p>

        <cite>Bryce</cite>
      <time>4:03</time>
      <p>
          Well,   I   mean,   I   no,   I   I   I   think   there&#39;s   some   there&#39;s   some   deep   lesson   to   learn   here.   Um   like,   you   know,   uh,   you   know,   if   you   if   you   fix   some   problem,   or   if   you   decide   that   it   wasn&#39;t   a   problem   at   all,   like   if   you   fix   some   problem   in   your   code,   you   need   to   do   something   to   test   that   the   fix   is   good,   and   and   you   also   need   to   add   some   sort   of   regression   test   to   make   sure   that   you   don&#39;t   run   into   the   problem   again.   I   I   could   imagine   ways   in   which   there   is   a   lesson   from   this   that   could   be   learned   uh   by   the   software   industry.
      </p>

        <cite>Conor</cite>
      <time>4:35</time>
      <p>
          Yeah,   don&#39;t   work   on   planes.   Alright,   here&#39;s   my   episode   topic   for   the   day.   We   were   recently   chatting,   was   it   a   week   ago?   Was   it   pre   or   post   a   recording?   I   can&#39;t   remember,   about   the   landscape   of   SDKs,   libraries,   languages   that   are   used   to   target   GPUs.   Specifically   in   our   case,   we   care   about   NVIDIA   GPUs.   And   I   I   showed   you   some   diagram   that   listed   off   a   few   of   these   things,   and   I   think   it   was   just   today,   I   decided   to   ask   ChatGPT   to   be   specific,   ChatGPT   4,   which   I   signed   up   for   in   the   last   couple   weeks,   for   a   specific   task,   which   it   was   not   able   to   complete   successfully,   but   now   I&#39;ve   paid   for   it   for   a   month,   so   we   may   or   may   not   keep   it.   And   the   question   that   I   asked   was   so   now   we&#39;re   gonna   do   a   little   family   feud,   this   little   quiz   time.   But   I   think   if   you   can   get   five   out   of   the   top   20   answers   that   ChatGPT   gave,   I&#39;ll   be   impressed.   Um   and   and   don&#39;t   worry.   I   think   you&#39;re   gonna   do   well.   But   uh,   you   know,   time   on   hang   on.
      </p>

        <cite>Bryce</cite>
      <time>5:44</time>
      <p>
          I   just   need   to   wait   long   enough   to   log   into   ChatGPT.
      </p>

        <cite>Conor</cite>
      <time>5:48</time>
      <p>
          Don&#39;t   be   honest,   be   honest,   Bryce.   I   can   see   I   can   see   the   light   changing   on   your   face.   I   know   you&#39;re   navigating   to   different   websites.   Do   not   I   want   to   see   your   hands   above   your   head   for   the   for   the   next   five   minutes.   You   can   rest   them   on   your   head   if   you   want.   So   the   question   I   asked   was   Hey,   hang   on,   I   gotta   get   back   to   seeing   your   beautiful   face.   Yeah,   I   want   hands   above   your   head   for   the   for   the   next   five   minutes   while   we   play   this   little   game.   And   a   listener,   you   can   play   along   as   well.   And   uh   and   then   after   we&#39;re   done   this   little   top,   you   know,   trying   to   get   five   out   of   20,   I&#39;m   curious   to   see   how   many   Okay,   Google,   open   Chat   GPT.   This   question   comes   in   two   parts   because   my   first   question,   ChatGPT   4,   went   going   on   ad   nauseum   and   detail,   and   I   was   like,   I   had   to   correct   it.   Anyway,   so   and   I&#39;ll   give   you   the   beginning   of   Chat   GPT&#39;s   answer,   uh,   which   is   before   the   list   starts.   So,   what   are   the   main   SDKs,   libraries,   and   ways   developers   utilize   GPUs?   Then   it   started   spitting   out   a   bunch   of   garbage,   and   I   said,   Stop,   stop,   stop.   And   I   said,   Don&#39;t   give   me   a   description,   just   list   me   20.   And   then   ChatGPT   responded,   sure,   here   is   a   list   of   20   SDKs,   libraries,   and   tools   for   GPU   programming.   And   I   will,   and   then   it   goes   one   through   20.   So   the   the   question   is,   how   many   can   Bryce   get?   And   I   will   give   you   a   hint.   The   first   one   is   extremely   obvious.   It   might   be   so   obvious   that   you   might   not   guess   it,   uh,   but   it&#39;s   like   it&#39;s   like   the   most   obvious   one   when   you   ask   how   do   you   target   a   GPU?
      </p>

        <cite>Bryce</cite>
      <time>7:17</time>
      <p>
          I&#39;m   gonna   go   with   um   OpenGL.
      </p>

        <cite>Conor</cite>
      <time>7:20</time>
      <p>
          All   right,   so   CUDA.   Actually,   wait,   wait,   wait.   Let   me   get   let   me   get   a   little   notepad   out.   So   your   first   answer   was   OpenGL.   That   is   number   five   answer   on   the   board.   So   you&#39;re   one   for   one.   Yeah.   What   was   your   second   one?   CUDA.   CUDA   is   the   number   one   answer   on   the   board.   So   you&#39;re   two   for   two.   Vulcan.   Vulcan   is   the   number   three   answer   on   the   board.   You&#39;re   three   for   three.   You&#39;re   crushing.   So   you   you&#39;re   definitely   Did   I   say   Open   CL?   Open   CL   is   number   two   on   the   board.   Four   for   four.   Not   a   single   miss   yet.
      </p>

        <cite>Bryce</cite>
      <time>7:55</time>
      <p>
          Oh,   so   I   wait,   can   you   just   I&#39;ve   gotten   number   one,   number   two,   number   three.
      </p>

        <cite>Conor</cite>
      <time>7:58</time>
      <p>
          And   number   five.
      </p>

        <cite>Bryce</cite>
      <time>7:59</time>
      <p>
          But   and   number   five.   Uh   which   one   was   number   five?
      </p>

        <cite>Conor</cite>
      <time>8:03</time>
      <p>
          Uh   so   the   order   was   CUDA,   open   CL,   Vulcan,   and   then   we   skip   number   four   and   then   open   GL.
      </p>

        <cite>Bryce</cite>
      <time>8:08</time>
      <p>
          Gosh.   I   I   don&#39;t   know   what   could   be   number   four.   Um   well,   see,   so   so   the   the   way   that   you   program   GPUs   on   on   uh   this   is   not   my   answer,   but   I&#39;m   gonna   explain   why   I&#39;m   not   gonna   give   it   as   an   answer.   Um   I&#39;m   not   gonna   say   metal,   in   the   which   is   the   the   API   for   programming   GPUs   on   Apple&#39;s   platform.   The   reason   I&#39;m   not   gonna   say   it   is   because   metal   is   newer.   Um,   and   so   I   think   it&#39;s   gonna   appear   like   later   on   the   list.   Um   oh,   uh   uh   wait,   hang   on.   Direct   X.
      </p>

        <cite>Conor</cite>
      <time>8:39</time>
      <p>
          Number   four   is   Direct   X.   Crushing   it.   So   not   a   single   miss   yet,   folks.   And   I   was   gonna   say,   we&#39;ll   go   till   you   get   like   three   or   five   misses.   Uh,   but   at   this   rate,   you   know,   uh   you&#39;re   gonna   run   out   of   things   to   guess   uh   because   you&#39;re   you&#39;re   crushing   it   so   much.   Yeah.   And   I   think   I&#39;m   gonna   be   generous.   Uh   Metal   is   number   18   on   the   list,   and   that&#39;s   because   uh   Chat   GPT   4,   I   think   it   got   updated   to   like   2023   June,   and   it   also   has   access.   I   see   you   typing,   buddy.   I   see   you   typing.
      </p>

        <cite>Bryce</cite>
      <time>9:08</time>
      <p>
          No,   no,   it&#39;s   it&#39;s   Ramona&#39;s   calling   it.   Hang   on,   pause.   And   you   know,   the   thing   about   that   Hands   above   your   head.
      </p>

        <cite>Conor</cite>
      <time>9:16</time>
      <p>
          Hands   above   your   head.
      </p>

        <cite>Bryce</cite>
      <time>9:17</time>
      <p>
          The   thing   about   that,   the   thing   about   that   is   that   she   knew   that   I   was   recording.
      </p>

        <cite>Conor</cite>
      <time>9:22</time>
      <p>
          Uh   you&#39;d   think   that   as   time   goes   on,   like   the   professional   quality   of   this   would   like   just   steadily   increase,   but   it   seems   it   seems   like   the   opposite.
      </p>

        <cite>Bryce</cite>
      <time>9:31</time>
      <p>
          That   one   was   not   my   fault.   I   I   can&#39;t   ignore   uh   a   phone   call   from   the   girlfriend   because   it   could   be   something   urgent.   Um   and   I   would   assume   that   if   she   was   making   a   phone,   if   she   was   calling   me   on   the   phone   while   she   knew   I   was   recording   the   podcast,   that   it   was   something   like,   oh,   I   get   hit   by   a   bus   or   something   like   that.   Not   oh,   I   can&#39;t   find   the   pet   go   order   that   I&#39;m   supposed   to   pick   up.
      </p>

        <cite>Conor</cite>
      <time>9:52</time>
      <p>
          All   right,   so   to   recap,   Bryce   has   got   his   hands   above   his   head   now.   He&#39;s   gone   five   for   five,   technically   six   for   six,   because   we&#39;re   giving   him   metal,   which   is   number   18   on   the   list.   And   not   as   not   a   single   miss   yet.   And   we&#39;ve   still   got   four   14   to   guess.
      </p>

        <cite>Bryce</cite>
      <time>10:04</time>
      <p>
          Well,   I&#39;m   gonna   say   thrust.
      </p>

        <cite>Conor</cite>
      <time>10:06</time>
      <p>
          Thrust   is   not   on   the   list.   Your   first   miss.   That   is   the   parallel   algorithms   library   that   we   have   talked   about   from   time   to   time   on   this   podcast.
      </p>

        <cite>Bryce</cite>
      <time>10:15</time>
      <p>
          Um,   how   about   HLSL?
      </p>

        <cite>Conor</cite>
      <time>10:17</time>
      <p>
          HLSL.   That   is   not   on   the   list.
      </p>

        <cite>Bryce</cite>
      <time>10:21</time>
      <p>
          Yeah,   that&#39;s   the   that&#39;s   like   uh   it&#39;s   the   high-level   shader   language.   Um   I   for   think   it&#39;s   like   a   part   of   opening.   Hands   above   your   head.
      </p>

        <cite>Conor</cite>
      <time>10:30</time>
      <p>
          Hands   above   your   head.
      </p>

        <cite>Bryce</cite>
      <time>10:31</time>
      <p>
          Hang   on,   I&#39;m   looking   at   no,   it   it&#39;s   like   a   DirectX   thing.   Um   let&#39;s   see.
      </p>

        <cite>Conor</cite>
      <time>10:36</time>
      <p>
          You   want   to   hint?   Think   about   uh   libraries.   Because   remember,   ChatGPT   responded,   sure.   Here   is   a   list   of   20   SDKs,   libraries,   and   tools.   And   a   couple   of   these   were   actually   on   the   diagram   that   I   showed   you,   if   you   recall.   The   diagram?
      </p>

        <cite>Bryce</cite>
      <time>10:48</time>
      <p>
          When   did   you   show   me   a   diagram?
      </p>

        <cite>Conor</cite>
      <time>10:49</time>
      <p>
          A   week   ago,   I   ended   up   showing   you   the   stuff   I   was   working   on,   and   then   mentioned   that   uh   was   having   a   conversation,   and   my   question   was,   you   know,   what&#39;s   the   most   PyTorch?   PyTorch   is   number   12   on   the   list.   We&#39;re   s   we&#39;re   what   is   that?   So   now   seven   for   nine.   If   PyTorch   is   on   the   list,   that   means   that   Numba?   What?   Numba?   Numba   is   a   great   guess.   I   was   surprised   to   not   see   Numba   on   this   list.   Uh,   but   numba   is   not   on   this   list.
      </p>

        <cite>Bryce</cite>
      <time>11:19</time>
      <p>
          If   Py   well,   uh   are   you   suggesting   that   torch   is   on   the   list   because   PyTorch   is   on   the   list?
      </p>

        <cite>Conor</cite>
      <time>11:23</time>
      <p>
          That   is   not   what   I   was   suggesting.
      </p>

        <cite>Bryce</cite>
      <time>11:25</time>
      <p>
          Okay.
      </p>

        <cite>Conor</cite>
      <time>11:26</time>
      <p>
          Then   continue   with   your   There   is   a   massive   competitor   that   used   to   be   more   popular   than   PyTorch.
      </p>

        <cite>Bryce</cite>
      <time>11:32</time>
      <p>
          Oh,   TensorFlow?
      </p>

        <cite>Conor</cite>
      <time>11:33</time>
      <p>
          That   is   correct.   Number   11   on   the   list.
      </p>

        <cite>Bryce</cite>
      <time>11:36</time>
      <p>
          Sorry   to   everyone   I   know   who   works   on   TensorFlow.   So   I   still   haven&#39;t   gotten   six,   seven,   eight.
      </p>

        <cite>Conor</cite>
      <time>11:44</time>
      <p>
          You   haven&#39;t   gotten   six   through   ten.   At   this   point,   you&#39;ve   got   the   top   five,   number   11,   TensorFlow,   number   12   PyTorch,   and   number   18   metal,   which   is   a   grand   total   of   eight.   You&#39;ve   had   three   misses,   so   we&#39;ll   give   you   two   more   misses.   And   definitely,   oh,   here&#39;s   another   hint.   So   number   eight,   number   nine,   number   seventeen   are   all   uh   by   NVIDIA.
      </p>

        <cite>Bryce</cite>
      <time>12:04</time>
      <p>
          Are   all   by   NVIDIA.
      </p>

        <cite>Conor</cite>
      <time>12:06</time>
      <p>
          And   actually   number   20   as   well,   although   they   don&#39;t   have   the   name   of   NVIDIA   in   the   title   of   it.   Number   And   actually   number   number   nine   doesn&#39;t   as   well.   So   but   anyways,   number   nine,   number   ten,   uh,   number   seventeen,   and   number   twenty.
      </p>

        <cite>Bryce</cite>
      <time>12:18</time>
      <p>
          Is   one   of   them   something   that   you&#39;ve   worked   on?
      </p>

        <cite>Conor</cite>
      <time>12:20</time>
      <p>
          One   of   them   is   something   that   I&#39;ve   worked   on,   yes.
      </p>

        <cite>Bryce</cite>
      <time>12:23</time>
      <p>
          Rapids?
      </p>

        <cite>Conor</cite>
      <time>12:23</time>
      <p>
          Rapids   comes   in   at   number   20.   And   it   actually,   it&#39;s   interesting.   It   says   rapid   that&#39;s   surprising.   Yeah,   it   says   rapid   AI,   and   then   in   parentheses,   formerly   known   as   Rapids,   a   suite   of   libraries   for   data   science   on   GPUs.   I   checked.   I   don&#39;t   think   it&#39;s   rapid   AI.   I&#39;m   pretty   sure   it&#39;s   called   the   the   URL   is   rapids.ai,   and   at   least   the   marketing   stuff   on   the   main   landing   page   still   says   rapids.   So   uh   rapids,   which   is   the   GPU   accelerated   version   of   the   popular   Python   library   pandas.   So   now   we&#39;ve   got   nine.   We&#39;re   nine   for   twelve.
      </p>

        <cite>Bryce</cite>
      <time>12:51</time>
      <p>
          Who   else   do   we   have   that   programs   GPUs?
      </p>

        <cite>Conor</cite>
      <time>12:54</time>
      <p>
          Yeah,   so   number   eight,   number   nine,   and   number   seventeen   are   all   NVIDIA   made.
      </p>

        <cite>Bryce</cite>
      <time>12:58</time>
      <p>
          I   know   two   of   them   are   Are   they   compute   or   graphics?
      </p>

        <cite>Conor</cite>
      <time>13:02</time>
      <p>
          One   of   them   is   compute,   two   of   them   is   compute,   and   I   think   the   third   one   is   leans   more   towards   graphics.   Yeah.
      </p>

        <cite>Bryce</cite>
      <time>13:11</time>
      <p>
          Oh   god,   what&#39;s   the   name   of   the   what&#39;s   the   name   of   the   RTX   thingy?   The   library.   But   I   don&#39;t   think   it   could   be   that   one.   What&#39;s   the   name   of   the   RTX   library?
      </p>

        <cite>Conor</cite>
      <time>13:23</time>
      <p>
          I   don&#39;t   know.   And   I   don&#39;t   think   it&#39;s   one   of   these   three,   so.
      </p>

        <cite>Bryce</cite>
      <time>13:26</time>
      <p>
          DLSS?
      </p>

        <cite>Conor</cite>
      <time>13:28</time>
      <p>
          It   is   not   DLSS.   And   that&#39;s   not   the   RTX   thing,   is   it?   That&#39;s   the   uh   super   sampling   thing   that   like   upscales   your   1080p   to   4k.
      </p>

        <cite>Bryce</cite>
      <time>13:38</time>
      <p>
          Yeah.
      </p>

        <cite>Conor</cite>
      <time>13:38</time>
      <p>
          One   wrong   guess   left.
      </p>

        <cite>Bryce</cite>
      <time>13:39</time>
      <p>
          Wait,   hang   on.   Okay,   okay,   I&#39;m   gonna   take   another   look.   Is   Unreal   Engine   on   there?
      </p>

        <cite>Conor</cite>
      <time>13:44</time>
      <p>
          Unreal   is   not   on   there.   So   in   total,   you   got   this   pretty   good,   man.   I   thought   you&#39;d   get   roughly   five,   because   that&#39;s   what   I   thought   I   would   get.   Just   because   some   of   these   are   are   off   in   the   weeds.   Uh,   but   you   got   nine   correct,   five   incorrect   guesses,   but   numba,   thrust,   I   think,   are   fantastic   guesses.   I   actually   don&#39;t   know   much   about   HLSL,   but   here,   let&#39;s   go   through.   So,   to   recap   what   Bryce   got,   number   one,   CUDA,   number   two,   open   CL,   number   three,   Vulcan,   number   four,   direct   X,   number   five,   OpenGL,   number   11,   TensorFlow,   number   12,   PyTorch,   number   18,   Apple   Metal,   and   number   20,   Rapids.   We&#39;ll   go   through   now   the   ones   that   you   missed.   And   this   is   gonna   be   entertaining.   Yeah.   Number   six   was   WebGo.   I   definitely   wouldn&#39;t   have   gotten   that.   Uh   yeah.   Just   because   it&#39;s   not   top   of   mind.
      </p>

        <cite>Bryce</cite>
      <time>14:26</time>
      <p>
          I   swear,   I   swear   to   God,   if   Cubs   on   this   list,   I&#39;m   gonna   be   picked   up.
      </p>

        <cite>Conor</cite>
      <time>14:29</time>
      <p>
          Cub   is   not   on   the   list.   Cub&#39;s   the   you   know,   sibling   library   to   to   thrust.   Yeah.   Number   seven   from   our   competitor.   Can   you   guess   based   on   that?   AMD.   Uh   Sickle?   Nope.   Hip?   Nope.   I   actually   only   learned   about   this.   One   API?   Like   uh   a   few   months   ago.   One   API   is   number   19,   but   that&#39;s   by   Intel,   not   AMD.
      </p>

        <cite>Bryce</cite>
      <time>14:50</time>
      <p>
          CUDA&#39;s   number   one   and   one   API   is   number   19.   That&#39;s   the   world   I   want   to   live   in.   Sorry   to   all   of   our   missed   listeners   who   work   on   Intel,   but   uh   number   seven   though   is   AMD&#39;s.
      </p>

        <cite>Conor</cite>
      <time>15:01</time>
      <p>
          What&#39;s   their   version   of   CUDA,   do   you   know?   Rock&#39;em?   Yeah,   Rock&#39;em.
      </p>

        <cite>Bryce</cite>
      <time>15:05</time>
      <p>
          Okay,   but   it&#39;s   it&#39;s   called   Hip   now.   But   I   I   fun   fact,   I   was   actually   at   the   launch   event   for   Rock&#39;em.   Wait,   Rock&#39;em   is   called   Hip?   Since   when?   The   hip   would   be   the   more   proper   thing   to   call   it.   Rock&#39;em&#39;s   like   the   name   for   the   like   overarching   tweet.   Back,   but   when   I   worked   at   Berkeley   Lab,   um,   I   think   his   name   was   Greg.   He   was   like   the   product   manager   for   um   uh   Rock&#39;em,   and   I   got   invited   to   go   down   to   Santa   Clara.   This   is   when   I   lived   in   Berkeley,   he   invited   me   to   come   down   to   Santa   Clara   for   their   launch   of   Rock&#39;em.   And   uh   he   was   really   into   uh   like   rock   music.   And   I   I   don&#39;t   recall   if   he   played   a   guitar   there,   but   there   was   some   like   rock,   like   there   was   definitely   like   a   rock   music   theme   to   the   launch   event.   But   yeah,   I   was   there   at   the   launch   event   for   Rock&#39;em.   And   there   that   was   where?
      </p>

        <cite>Conor</cite>
      <time>15:52</time>
      <p>
          What   city?
      </p>

        <cite>Bryce</cite>
      <time>15:53</time>
      <p>
          That   was   it   was   in   San   It   was   in   San   Jose,   I   think.
      </p>

        <cite>Conor</cite>
      <time>15:56</time>
      <p>
          Okay.
      </p>

        <cite>Bryce</cite>
      <time>15:56</time>
      <p>
          Yeah.
      </p>

        <cite>Conor</cite>
      <time>15:57</time>
      <p>
          Interesting.
      </p>

        <cite>Bryce</cite>
      <time>15:57</time>
      <p>
          So   wait,   okay,   so   what   is   uh   just   to   just   to   pause   on   this,   you   know,   uh   NVIDIA   has   you   know,   HIP   is   I   think   Rockham   is   the   name   for   the   overall   software   stack,   but   HIP   is   specifically   their   like   CUDA-like   clone.
      </p>

        <cite>Conor</cite>
      <time>16:13</time>
      <p>
          I   see,   I   see.   Yeah,   so   the   Wikipedia   says   Rocam   is   an   AMD   software   stack   for   GPU   programming,   and   it   spans   several   different   domains,   blah,   blah,   blah.   But   then   it   says   it   offers   several   programming   models   and   then   proceeds   to   list   three   or   four   of   them,   which   is   hip.
      </p>

        <cite>Bryce</cite>
      <time>16:28</time>
      <p>
          Yeah,   yeah,   yeah.   I   I&#39;m   sorry,   I   was   confused   early.   Hip   is   one,   but   like   I   think   when   Rokam   launched,   um,   it   was   supposed   to   be   like   their   like   CUDA,   like   CUDA   for   their   platform,   and   I   think   later   the   the   hip   branding   got   introduced.
      </p>

        <cite>Conor</cite>
      <time>16:41</time>
      <p>
          Oh,   interesting.
      </p>

        <cite>Bryce</cite>
      <time>16:42</time>
      <p>
          So   I   think   that   that&#39;s   why   that&#39;s   why   I   thought   that   that   Rakham   had   been   renamed   to   hip,   because   I   think   originally   what   Rockham   was   was   supposed   to   be   like   their   CUDA,   but   I   I   could   be   I   could   be   wrong   about   that.   Anyways,   but   yeah,   Rocam   is   like   the   overall   like   like   um   uh   plat   like   in   the   same   sense   that   like   like   CUDA   is   like   the   overall   platform,   but   like   CUDA   C   ⁇   is   one   way   to   program   that.   Like   Rockham   is   the   overall   platform,   and   uh   Hip   is   their   CUDA   C   ⁇   equivalent.
      </p>

        <cite>Conor</cite>
      <time>17:06</time>
      <p>
          Gotcha.
      </p>

        <cite>Bryce</cite>
      <time>17:07</time>
      <p>
          Anyways,   give   me   the   rest   of   the   list.
      </p>

        <cite>Conor</cite>
      <time>17:08</time>
      <p>
          Alright,   so   that   was   number   seven,   number   eight,   and   number   nine   are   both   from   NVIDIA.   It   starts   with   coup   and   ends   in   three   letters.   Under   under   underpins   both   PyTorch   and   TensorFlow.
      </p>

        <cite>Bryce</cite>
      <time>17:22</time>
      <p>
          Coup   DNN?
      </p>

        <cite>Conor</cite>
      <time>17:23</time>
      <p>
          Coup   DNN   is   correct,   yes.   I   was   I   expected   you   to   get   that   one   as   well.   Number   nine,   a   little   bit   more   esoteric,   but   we&#39;ve   been   getting   emails   I   think   every   week,   and   it&#39;s   been   called   Insert   Tuesday.   Have   you   have   you   been   reading   your   NVIDIA   emails   closely   enough?   And   that   being   said,   we   get   literally   like   thousands   of   emails,   and   that&#39;s   even   after   like   filtering   them.
      </p>

        <cite>Bryce</cite>
      <time>17:43</time>
      <p>
          Yeah.   Tensor   RT.   Yeah,   and   video   Oh,   Tensor   RT.   Although   actually   Yes,   I   forgot   about   that.   I   forgot   about   Tensor   RT.   Yeah,   that&#39;s   not   true.
      </p>

        <cite>Conor</cite>
      <time>17:52</time>
      <p>
          Now   that   I&#39;m   actually   thinking   about   this   though,   uh   is   it   is   it   Tensor   RT   Tuesdays?   Now   that   I   think   about   it,   I   don&#39;t   think   it&#39;s   Tensor   RT   Tuesdays.   No,   it   is   Tensor   RT   Tuesdays.   Hopefully   this   isn&#39;t   like   proprietary   stuff.   It&#39;s   alright   to   mention   that.   In   public,   right?   The   name   of   a   NVIDIA   internal   email.
      </p>

        <cite>Bryce</cite>
      <time>18:08</time>
      <p>
          I   mean,   I   think   I   think   this   whole   podcast   episode   is   sketchy   because   we&#39;re   uh   we&#39;re   talking   about   uh   competitors   SDKs,   and   I&#39;m   making   uh   and   I   I   seem   to   recall   earlier   I   made   a   crude   joke   at   Intel&#39;s   expense.   Uh   but   you   know   what?
      </p>

        <cite>Conor</cite>
      <time>18:24</time>
      <p>
          Hey.
      </p>

        <cite>Bryce</cite>
      <time>18:25</time>
      <p>
          It   is   what   it   is.
      </p>

        <cite>Conor</cite>
      <time>18:26</time>
      <p>
          This   is   uh   not   the   opinions   of   NVIDIA,   it&#39;s   the   opinions   of   us,   and   this   is   just   what   ChatGPT   lists.   We   didn&#39;t   put   them   number   seven.
      </p>

        <cite>Bryce</cite>
      <time>18:34</time>
      <p>
          But   buddy,   okay,   but   buddy,   but   what   trained   Chat   GPT?
      </p>

        <cite>Conor</cite>
      <time>18:40</time>
      <p>
          Uh   that   the   compute   that   trains   NVIDIA   GPUs.
      </p>

        <cite>Bryce</cite>
      <time>18:44</time>
      <p>
          Yeah,   but   that   that   doesn&#39;t   affect   the   answers.   I&#39;m   telling   you,   that   that   large   language   model,   it   almost   certainly   knows   how   it   was   trained,   right?
      </p>

        <cite>Conor</cite>
      <time>18:53</time>
      <p>
          I   don&#39;t   think   so.
      </p>

        <cite>Bryce</cite>
      <time>18:56</time>
      <p>
          I&#39;m   sure   that   in   the   uh   the   input   data,   there&#39;s   a   lot   of   information   that   says   lar   that   the   that   that   open   AI   uses   NVIDIA   GPUs.   And   so   of   course   you&#39;re   asking   you   ask   that&#39;s   basically   asking   her   like   how   were   you   born?
      </p>

        <cite>Conor</cite>
      <time>19:08</time>
      <p>
          It   is   funny   actually   because   number   13   on   the   list   is   open   AI   GPT.
      </p>

        <cite>Bryce</cite>
      <time>19:14</time>
      <p>
          Oh,   that&#39;s   interesting.   That&#39;s   interesting.
      </p>

        <cite>Conor</cite>
      <time>19:16</time>
      <p>
          Which   uh   I   don&#39;t   know,   I   didn&#39;t   to   call   open   AI   GPT   like   an   SDK,   I   guess   maybe,   because   there   are   SDKs   attached   to   like   uh   Yeah.
      </p>

        <cite>Bryce</cite>
      <time>19:26</time>
      <p>
          I   mean   I   like   I&#39;ve   used   the   OpenAI   API,   so   um,   yeah.
      </p>

        <cite>Conor</cite>
      <time>19:30</time>
      <p>
          Not   shocked   by   that   one.   Continuing   down   the   list,   number   10   is   PyCuda,   which   I   am   actually   not   familiar   with.   Number   11,   we   already   covered,   so   we&#39;ll   skip   to   number   four.
      </p>

        <cite>Bryce</cite>
      <time>19:40</time>
      <p>
          Oh   yeah,   yeah,   I   think   PyCuda   is   not   actually   a   thing   that   we   uh   officially   support.   I   think   PyCuda   is   like   the   Python   CUDA   um   interface   thingy.
      </p>

        <cite>Conor</cite>
      <time>19:51</time>
      <p>
          Okay.   Number   14,   Microsoft   CNTK,   the   cognitive   toolkit,   never   heard   of   it.   Number   15,   Theano   with   GPU   support,   which   I&#39;m   surprised   that&#39;s   up   there   because   from   what   I   know,   like   Cafe   and   Theano   are   like   pretty   old   technologies,   but   hey,   probably   there&#39;s   a   lot   of   stuff   out   there.   Number   16,   AMD   Fire   Pro.   Never   heard   of   it.   Have   you   heard   of   it?
      </p>

        <cite>Bryce</cite>
      <time>20:17</time>
      <p>
          Uh   long,   long   time   ago.
      </p>

        <cite>Conor</cite>
      <time>20:21</time>
      <p>
          And   number   17   is   Nvidia   FizzX,   which   I   which   I   assume   is   a   physics   SDK   or   library.
      </p>

        <cite>Bryce</cite>
      <time>20:29</time>
      <p>
          Yeah.   But   okay,   so   AMD   Fire   Pro,   I   don&#39;t   think   it   was   their   name   of   like   their   software.   I   think   it   was   just   the   name   for   their   like   their   brand   of   graphics   cards.
      </p>

        <cite>Conor</cite>
      <time>20:42</time>
      <p>
          It   was   like   for   their   uh   it   says   uh   in   parentheses   next   to   AMD   Fire   Pro   that   it   was   an   STK   for   AMD   GPUs.
      </p>

        <cite>Bryce</cite>
      <time>20:50</time>
      <p>
          Uh   maybe   it   was.   But   I   think   it   was   I&#39;m   pretty   sure   it   was   old.   I&#39;m   pretty   sure   it   was   like   the   2000s.
      </p>

        <cite>Conor</cite>
      <time>20:58</time>
      <p>
          Yeah.   Well,   I&#39;m   I&#39;m   not   surprised.   I   mean,   Theano,   let&#39;s   let&#39;s   let&#39;s   check   in   on   Theano.   Ooh,   that&#39;s   not   what   I   meant   to   do.   Just   closed   all   my   tabs.
      </p>

        <cite>Bryce</cite>
      <time>21:06</time>
      <p>
          Nice,   Connor.   Closing   all   your   tabs.
      </p>

        <cite>Conor</cite>
      <time>21:09</time>
      <p>
          Well,   Ctrl   Shift   T,   baby.   Theano   is   a   Python   library   for   optimizing   compiler.   I&#39;m   gonna   do   it.   It   opens   up   your   most   recently   closed   tabs,   and   if   you   close   the   whole   window,   it&#39;ll   reopen   everything.   So   whenever   your   computer   shift   uh   restarts,   I   mean   it   always   asks   if   you   want   to   restore,   but   if   you   uh   need   to   and   you   like   accidentally   clicked   no   to   that,   you   can   just   go   Ctrl   Shift   T   a   bunch   of   times.   I   mean,   yeah,   Theano   came   out   in   2007.   So   yeah,   it&#39;s   been   a   while   for   a   row.   Been   a   while   been   around   for   a   while.   Anyway,   so   I   guess   to   wrap   up   this   little   I   mean,   first   of   all,   you   did   fantastic.   Definitely   better   than   I   would   have   done.
      </p>

        <cite>Bryce</cite>
      <time>21:46</time>
      <p>
          You   know   what?
      </p>

        <cite>Conor</cite>
      <time>21:56</time>
      <p>
          Well,   it&#39;s   more   important   is   that   you   crushed   top   five,   you   know,   right.   Off   the   bat   without   even   getting   a   single   one   wrong,   which   is   pretty   impressive.   And   I   guess   you   know,   to   put   a   little   bow   on   this   little   30-minute   episode   is   the   question   that   you   know   Bryce   and   I   were   talking   about   a   week   ago   at   OffPod   was   you   know,   what   is   the   main   the   main   way   that   developers   target   GPUs?   Like   what   path   are   they   taking?   Is   it   is   the   number   one   you   know   and   and   that   includes   things   like   PyTarch   and   TensorFlow,   where   even   though,   you   know,   potentially   it&#39;s   not   like   you&#39;re   doing   quote   unquote   GPU   programming,   you&#39;re   just   doing   you   know   machine   learning   stuff   that   happens   to   at   the   end   of   the   day   target   accelerated   compute,   which   could   be   in   the   form   of   a   GPU.   Like,   is   that   where   most   of   like   if   you   were   to   create   a   pie   chart   of   developers   that   target   GPU   compute   and   for   our   concern,   it&#39;s   you   know   mostly   like   I   said   before,   NVIDIA   GPUs,   what   does   that   pie   chart   look   like?   And   uh   that&#39;s   something   that   I   don&#39;t   really   have   the   answer   to.   I   don&#39;t   know,   I   haven&#39;t   seen   any   GTC   talks   or   keynotes   that   talk   about   that   because   I   think   uh   most   investors   and   people   they   care   about   NVIDIA,   it&#39;s   not   something   they   care   about,   but   as   like   a   developer   um   and   someone   that   works   on   libraries,   uh   it&#39;s   something   that   uh   I&#39;ve   been   thinking   about   recently.   Anyways,   thoughts,   comments?   I   know   I   know   your   initial   response   was,   yeah,   I   haven&#39;t   heard   uh,   you   know,   I   haven&#39;t   heard   of   a   slide   that   shows   that   data.
      </p>

        <cite>Bryce</cite>
      <time>23:28</time>
      <p>
          Well,   yeah,   it&#39;s   I   feel   like   I   feel   like   uh   maybe   this   is   just   my   corporate   experience,   but   I   feel   like   so   often   data   like   this   that   would   totally   help   us   make   better   decisions.   I   feel   like   there   are   so   many   software   organizations   that   don&#39;t   really   know   the   answers   to   these   things.   And   then   but   because   you   don&#39;t   actually   have   data   on   how   your   thing   is   used,   you   just   like   you   make   decisions   based   on   how   people   not   even   think   things   are   used,   but   how   people   feel   things   are   used.   And   I   make   the   distinction   because   thinking   is   like   a   logical   process,   whereas   feeling   is   an   emotional   process.   And   that&#39;s   how   I   I   think   we   uh   we   we   often   approach   this   question.   You   know,   I   I   I&#39;m   I&#39;m   sure   that   there   is   some   data   that   somebody   has,   but   um   the   fact   that   uh   that   I   that   it&#39;s   not   that   it&#39;s   not   something   that   like   is   in   front   of   me   is   like   a   bad   sign,   right?   Like   that&#39;s   something   that   that   I   should   be   that   like   every   every   like   senior   engineer   at   the   company   should   have   a   should   have   a   good   sense   of   that,   right?   Um   but   it&#39;s   also   when   you   have   uh   a   product   in   an   SDK   it&#39;s   as   widely   used   as   ours,   it   is   difficult   to   truly   survey   and   understand   all   of   your   users.   You   know,   and   and   also   like   how   do   you   how   do   you   assign   importance?   Is   it   by   the   the   the   the   number   of   developers   that   that   work   on   the   thing?   Is   it   by   the   amount   of   revenue   that   they   generate?   Is   it   by   like   like   uh   you   know,   is   it   by   the   the   amount   of   code   that&#39;s   written   by   a   particular   domain,   or   is   it   by   how   often   that   code   is   run?   Um   I   think   it&#39;s   very   difficult   to   actually   quantify   how   your   software&#39;s   used.   You   know,   make   maybe   maybe   there&#39;s   like   one   10-line   program   somewhere   that&#39;s   like   run   on   like   every   computer   in   the   world,   and   like   that&#39;s   how   your   software   is   most   frequently   used,   um,   versus   all   the   like   all   the   other   like   thousands   of   lines   that   are   code   of   code   that   are   written   that   use   your   library   are   barely   used.   It&#39;s   just   that   one   place   where   it&#39;s   used   a   bunch   of   times.
      </p>

        <cite>Conor</cite>
      <time>25:38</time>
      <p>
          I   mean,   I   completely   agree   that   this   is   an   extremely   hard   thing   to   measure.   I   I   was   I   just   while   you   were   chatting,   I   Googled   you   know,   CUDA   developer   survey,   because   I&#39;m   pretty   sure   that   doesn&#39;t   exist,   and   sure   enough,   it   does   not.   We   have,   I   think,   some   uh   byproduct   surveys,   like   I   think   the   fourth   or   third   top   thing   was   a   coup   DNN   stakeholder   survey.   But   I   mean   it&#39;s   if   there&#39;s   a   you   know   a   C,   you   know,   uh   what   do   they   call   it?   Um   community   survey,   there&#39;s   a   couple   of   them,   right?   One&#39;s   done   by   the   ISO   Foundation,   one&#39;s   done   by   JetBrains.   I   think,   you   know,   we   should   try   to   collect   some   of   that   data.
      </p>

        <cite>Bryce</cite>
      <time>26:18</time>
      <p>
          I   mean,   it&#39;s   it&#39;s   kind   of   sad   that   like   saying   I&#39;ve   been   saying   that   for   years.   I&#39;ve   been   trying   to   put   together   a   CUDA   survey   for   years.
      </p>

        <cite>Conor</cite>
      <time>26:24</time>
      <p>
          Yeah,   I   mean,   why   can&#39;t   we?   I   think   we   should   we   should   commit.
      </p>

        <cite>Bryce</cite>
      <time>26:27</time>
      <p>
          I   mean,   we&#39;ll   you   know   laziness.   What?
      </p>

        <cite>Conor</cite>
      <time>26:29</time>
      <p>
          Laziness   mostly.   Yeah,   I   mean   we   can   make   it   happen   if   we   wanted   to.
      </p>

        <cite>Bryce</cite>
      <time>26:32</time>
      <p>
          Um   and   I   think   actually,   like   the   the   you   know   Honestly,   if   you   if   you   wanted   to   ask   me   the   reason   why   um   uh   why   I&#39;ve   never   like   completed   plans   to   put   that   through.   And   like   I&#39;ve   I&#39;ve   written   it   on   my   to-do   list   like   four   or   five   times   over   the   the   six   years   I&#39;ve   been   here   in   NVIDIA.   Um   it&#39;s   just   it&#39;s   not   as   exciting   a   thing   to   do.   It&#39;s   important,   but   it&#39;s   not   as   exciting   or   cool.   And   also   it&#39;s   it   always   seems   like   it&#39;s   a   lot   of   work.   There&#39;s   a   lot   of   like   privacy   and   concerns   and   whatnot,   like   that.   So,   see,   if   we   were   gonna   do   this,   I   would   want   to   do   it   correctly.   And   the   way   that   I   would   want   to   do   it,   it   it   this   brings   up   something   that&#39;s   actually   been   on   a   recent   episode,   unless   Connor&#39;s   gotten   rid   of   it.   The   JetBrains   Ecos   like   developer   ecosystem   survey.   You   know   what   JetBrains   did?   They   one,   they   didn&#39;t   try   to   go   and   run   the   survey   themselves.   They   went   and   like   hired   people   who   are   like   professional   survey   runners   and   they&#39;re   like,   tell   us   how   to   do   this.   And   two,   they   do   the   survey   and   then   they   publish   the   results,   and   they   not   only   publish   the   results,   they   don&#39;t   publish   all   the   data,   but   they   publish   the   results   and   they   publish   a   description   of   their   statistical   methods.   Um   and   like   they   they   100%   do   this   the   correct   way.   So   like   if   I   was   gonna   do   this   sort   of   survey   at   Nvidia,   I   would   not   like   get   together   a   group   of   people   to   come   up   with   a   set   of   questions   and   make   like   a   Google   Drive   form.   Um   I   would   want   to   get   like   a   pile   of   money   and   go   to   some   company   who   are   professionals   and   doing   surveys   and   be   like,   you   go   figure   this   out.   Here&#39;s   my   requirements,   you   go   figure   this   out   for   me.
      </p>

        <cite>Conor</cite>
      <time>27:59</time>
      <p>
          I   mean,   I   completely   agree.   And   I   think   whatever   piles   of   money   um   it   costs   to   make   that   happen   probably   is   worth   it.   I   mean,   like   it&#39;s   sad.   You   Google   CUDA   Developer   Survey,   and   the   one,   two,   three,   four,   five,   six,   seventh   result   is   CUDA   versus   Rock   M,   the   ongoing   battle   for   GPU   computing.   And   it&#39;s   a   medium   article   by   I   don&#39;t   know   who   this   is,   uh   doesn&#39;t   actually   say.   Um   we&#39;ll   leave   it.   I   mean,   I   haven&#39;t   even   read   it,   I   just   glanced   at   it.   Uh,   but   the   point   is,   is   like   we   should   be   collecting   this   kind   of   stuff.   Like,   we   have   a   GTC,   you   know,   we   have   more   than   one   of   them,   but   like   the   main   one   happens,   whatever,   March,   April.   How   hard   would   it   be   like   if   we   contact   one   of   these   surveying   companies   a   couple   months   out,   say,   here&#39;s   our   criteria,   and   we   want   to   like,   you   know,   provide   this   for   every   single   person   that   either   virtually   attends   or   in-person   attends,   has   the   ability.   Plus,   we&#39;ll   also   like   put   the   link   on   online   for   anyone   that   just   wants   to   go   and   fill   it   out.   And   I   think   we   could   we   would   get   thousands   of   responses.   Like   supposedly   there   are   like   four   to   five   million   registered   CUDA   developers   or   something   like   that   in   the   world.   Like,   it&#39;d   be   pretty   easy   to   get,   I   think,   several   thousand,   you   know.   Uh   anyways,   we&#39;ll   put   it   on   our   list.
      </p>

        <cite>Bryce</cite>
      <time>29:03</time>
      <p>
          I   mean,   look,   and   if   and   if   it&#39;s   too   expensive,   we   can   always   just,   you   know,   take   take   that   out   of   your   pay,   right?
      </p>

        <cite>Conor</cite>
      <time>29:08</time>
      <p>
          I   mean,   not   my   pay,   take   it   out   of   your   pay.   Anyways,   that&#39;ll   wrap   what   it   what   what   are   we   on?   80s   ADSP,   episode   166.   Be   sure   to   check   the   show   notes   either   in   your   podcast   app   or   at   adspthepodcast.com   for   links   to   anything   we   mentioned   in   today&#39;s   episode,   as   well   as   a   link   to   a   GitHub   discussion   where   you   can   leave   thoughts,   comments,   and   questions.   Thanks   for   listening.   We   hope   you   enjoyed   and   have   a   great   day.
      </p>

        <cite>Bryce</cite>
      <time>29:34</time>
      <p>
          Low   quality,   high   quality.   That   is   the   tagline   of   our   podcast.
      </p>

        <cite>Conor</cite>
      <time>29:40</time>
      <p>
          That&#39;s   not   the   tagline.   Our   tagline   is   chaos   with   sprinkles   of   information.
      </p>