AMD does reverse GPGPU, announces OpenCL SDK for x86

http://v12extreme.com/ipost/facto300.png
300 done ...
http://v12extreme.com/ipost/facto3000.png
done ...
41493596034378540855568670930866121709511191949318 09917689467657697558565123531950086000765217800342 00751846353836171184957508711140459077945534021610 68339611621037904199177522062663390179682805164719 69749596884245772876609710300372611109534024112711 88331577388153284389297376130211063129303744014853 78725446079610290429491049793888120762511625132917 00464166896211759020357517548898065357786891528509 37824699946746991908320935110683638242870635222685 44339213775150488588104036818809099292912497141900 50893899440471535147315453158744150996017426787508 74603679741170723687472771439889206836916185036081 98459718093784453523958505377611086511162363145920 88610855745087451394530543621371189815084719209442 63742032750299963337849440147756714146808242074999 14714878359669720638954670589960178569480263388767 11287106800495082740071712481947638640136919354435 41203127866014347925499591435301206531034066255032 31020738351502195103148673612338739395096551462159 34901578994994407231100442692483814014145548787273 80458560235615832043179459530558306933512468907212 46151468485308724031267967089113548982733475375756 89936517639642478173346251087901574343739892049226 70983170339321071763439833524445760404765654004144 14699479984354554597799386702839428513413188913165 69531084851352509400614777404700733140654179442800 44366919036854692708572717016480115120574524486079 68773784803660653009109815639091294110633715621540 90380013505867162426233390243416662871652122859027 45688335048979268693697928783768948414365738664369 55075473964882256222183380014600761196859217603234 80846745521633041173800433114422592624369055878291 49079738857587845857398286953903023838372658824276 54306437517757897215045071361801730051628424476294 22748575562782876349876719528136891358391882449928 47415916831303340321999467520829148857643458638323 13545205075955912062067273296951386122994658607527 31788445244986534816416923884488906149585093437344 28898148844273218171312725338915345065811438233812 05875379808605080889761753882896252933633750454549 16860026722959122552885458448268665532431301135375 48124095612376860780077007079395418489071494673778 54407528307872988103912945121929864793703451257436 44558145975714082270598632516535290658457112358527 02119334529811055683988098840949803461850780252730 38736784042169427237980464304250045030806637032760 01634192144280570880243085056789210864697745513953 91198386361671903002781463801369324823327715951805 96193069504237836082620570887209297929797429404576 87733831987744468554429480032174105668942371054502 88704196119150727390000316420144742133232938716180 29555614004602867400422885389854650328028428515122 29602879574180162182323609832097144104701253306731 48961532367887349845539496043970503523477662113959 14519270422122231426998692087463520980686224354813 37619439513194286811348653156222817321497648170538 18461553265961875302964786011608722636404439222576 01926494610916885151013143945574398303192557154162 15144246912237051914909786184943615096310993363959 45617965933968519586053386311763241470668422571923 94742531726479559749993283247279807896470753054014 19409020060971267475318636552540321275775785393069 75300565952082074574994718981444537722482078884433 35118545601568853708182892895218300139654376947286 41877666576281538973734015941054368143543734613424 46920670700827824236455574508825566701572427528103 17141640631410681384330924027281318960884813040665 22616955282563718386246494429568885939384672672369 41994755713205460182634257310291153535327288081827 73021596787088437293412117084511580629967697266601 66363527695996902150212210495425956727859318551626 84471003744346204220035353912037383930954206950214 86207390653190910821344334251497896284236198571674 77384812609744305503625086635472073097129808469719 65377227798931602005607250580075124074944481633922 14398118492748281978655178478547749198714138485042 29038395409057084203813727713566770356504108178052 06950321362335216927405310153409217618340788177356 74646749071616600653230438902639786065509005309872 43544568931560132994240711229501545377152105194244 55127953649712148722221937292891598330017423979775 92530501318837883494884232222507318816399438935627 81710287543258879455885774278039071716638125790379 81491484455268858716299310145107332155547732645760 35916184298708323237568837917135073006026738292294 68708103075194602037643813867710733377931258225735 64355345771628040304809257859097472334139329040722 39860005448269296110393640127539539899397420021925 26892862256495927913636954698324731449409429749421 32087169636628129638461913781146092107010330121199 34264941666449130310898493535366401831282683112506 57838642590653719701090727642933053475129733671692 94150478709492417781215349794994497323584451302100 29720359993576507730563696950539990891252004810120 09056963314436817919424796356338910248625077336724 93998017234516270488501494383437358264400534814749 57421328873648479589553843836378275601433377798816 12685446240649413441611910895265332676162766022113 08792116659243794965348380302360642949819855410143 11566601739518539426008673198564586684635442730180 02229260758976719219836752952836515871552188769831 79990058531215186910377766768836542912474198260994 34535671529412823837612115555686210454583810355154 40495371847072636321853277548650181100262133122842 98609261121595730660239320774767428009094626743221 38805290643067711276964013735906251051050623568241 31765153303077535897513456514742416740151747072083 91018699899932793649108926879247397058141528555439 65954222603919059265825637344676406359525838966981 51198395988660368375304201799032818594556941255051 90663028548695333776829846000318080938221300381022 14387057461181304251961916405970456035183121708151 65864735655654053292841174862895708285679230005352 58463770612805914520355463899321278759063496278379 75871352588618213252263577038396202737385324908353 68049799008570152248330343952519734465334299465256 52360967428345505237397339023742618088717992837222 85366293439240895762913154442106573609205481842139 36589386771554284247727510016673435774309363894844 45647643771840738743794710078671510704495546576262 81566137550730763768080600031844296233977808233311 35978757713698301281757162567168328728151193733668 57894371090977485812228681268241223172726811849752 07863453107495331708260153159440253645365524453587 95203474521342924891664450480435535228197772198197 18690548841768963987827047820661269214725486182478 59626434279190274503452994769367997217285165465591 79947178906788568727857447008428972377823476308074 09195129662383464278396538650173246658501921440916 94630371265581197700774682562035198318782913591013 99781730363517376470671438399281029122446084832051 89832483488551310255397215831849316536707322731729 95431750775475634748127320956655431851879586978172 49172170086576809890832783083824043773797445534252 56887128988555131809670124978594542906096273705906 59970784172738420721605576789060565167694565490120 38816577586193923092436298338954985727987452339809 04998584674848503995091093988342104246931136178759 78611803096108774362764990414655167545507613665725 91499337611434024376291029038413588853131259113254 48492258960071848511693901939854346494154837823383 02531368775990005443722332901462568184095998830522 52158532859983399033659541893269668016326589935823 46632470803240204297913574257554985493728961920916 50794671997121439832581553945835125648010889886887 05688271122262873403577241842480323117302733844222 06040156092420795694932049438094024655625303033288 24165302038006041288444384884189129393985971765670 21150161134012116935553586498480294156323827944757 63150426857342698631165628009321645781654104118990 78396210758605145091526528422433647230880469088426 41252512658472913405919517175429115262200222975698 69279591246209643630570521330992164222584376518891 93630329851223282950806126200573565554213183555838 28931813879594096230379277723034442343234156160355 85905023244752745026308698314141253963717544136118 97269158650716722308083435295578401087236027347001 11878614623318543943105705848377047480603500455688 50206027302222563976307389399850249781551826799169 94164145540329909813190506654358156657691529068908 18620413844409145635529124206490171743643047345519 13759229149532829881518087400767334869976953228714 50791584448703980405737673555777873593937891577147 95602334070845639231417011839255523461811977591567 33859559192652706240637342777602158465110353680579 63320714896942663358570375305829676608224208465464 55855666788922262761999026396179263745785165254091 87566085438596612219442487204249600000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 00000000000000000000000000000000000000000000000000 000000000000000000000000000000

the time I saved programming the thing for sure make up for the time the PhD you need to do for GPGPU trivial coding of Factorial ....

I am now calculation the 1000000! ... what about you?

Dude, I can download a factorial calculator too. You're still avoiding my questions.

Quote:

---

Originally Posted by trinibwoy

The entire premise of your argument is false. Factorial isn't a strictly iterative calculation as all mutliplications of every two numbers in the sequence can happen in parallel. For example, 10!, using 4 threads.

Clock #1
Thread 1: 10*9
Thread 2: 8*7
Thread 3: 6*5
Thread 4: 4*3
Clock #2
Thread 1: 90*56
Thread 2: 30*12
Clock #3
Thread 1: 5040 * 360
Clock #4
Thread 1: 1814400 * 2

Only 4 clocks on a theoretical 4 core processor. How is your naive iterative algorithm supposed to be faster when it takes 8 clocks? You really need to read up on parallel scans, there isn't even any branching in this algorithm so I'm really curious where you're getting all of this bad info from.

---

did you ever try to move result from one Stream unit to an other one in a GPU?
the thing that I know if that it toke me 1 hour to prove that it is easy to do it on a CPU ... and you are still trying to do the theory ...

What you do not realize is that the arguement that I am using is what make x86 the most formidable architecture, I can put software together at the speed of light, while you keep thinking about how to do it ...
I gave you the link about how to parallelize n!, I am fully aware of it ... just try to implement it, and you ll figure out that your "trivial" claim is 100% bogus. In the mean time, I am demonstrating how easy it was for me to calculate the trivial part.

It is always easy to say it is trivial, it is an other story to actually do the work ... and it is my entiere point ...

I am a Doer ... I do stuff, I don't do claims without knowing exactly what I am talking about.

AMD is right, on OpenCL, it is more important to start with the CPU side. :yepp:

Back to Topic, thanks for not using "trivial" anymore in your posting.
my 1000000! number is coming. Just for the fun of it, I ll thread it :)

i wonder what it is like to be brainwashed by intel?:rofl:

Quote:

---

Originally Posted by Drwho?

It is always easy to say it is trivial, it is an other story to actually do the work ... and it is my entiere point ...

---

Ummmm it is trivial, any high school student can understand how easily parallelizable it is. Don't know why that seems to escape you.

Quote:

---

I am a Doer ... I do stuff, I don't do claims without knowing exactly what I am talking about.

---

You haven't done anything except paste the result from a program you downloaded on the internet :rolleyes: What you have done though is demonstrate that you believe running something slowly on a CPU is good because it's easier to code. That's pretty backward thinking my friend.

'for' loop initial declaration used outside C99 mode .:(

Quote:

---

Originally Posted by trinibwoy

Ummmm it is trivial, any high school student can understand how easily parallelizable it is. Don't know why that seems to escape you.



You haven't done anything except paste the result from a program you downloaded on the internet :rolleyes: What you have done though is demonstrate that you believe running something slowly on a CPU is good because it's easier to code. That's pretty backward thinking my friend.

---

Dude ... I assemble it, made the MFC frame ... and prove that you did not do it :-P
again ... I showed you the link before you come up with you own // "idea".
here :http://www.luschny.de/math/factorial/Benchmark.html
I understand it is perfectly threadable ... TRY TO DO IT ON THE GPU AND YOU'LL FIGURE OUT THAT YOU ARE SLOWER THAN THE CPU!

CLEAR!??


:rofl: :rofl: :rofl: I don't know who is the one with 'understanding' problems here...

why are you arguing about lame factorials :rolleyes:

Quote:

---

Originally Posted by trinibwoy

Ummmm it is trivial, any high school student can understand how easily parallelizable it is. Don't know why that seems to escape you.

---

hey im in high school and i take that as an offense but i do see how easy that would be to parallelize and then paste into a forum. it looks like you can do this with a lot of sequences.

Quote:

---

Originally Posted by clonez

why are you arguing about lame factorials :rolleyes:

---

That is what trinibwoy picked on my posting to say that I was totally wrong ... His Big opinion, nothing to back it up ... That is the conclusion. :yepp:

You should probably tell ATI, Nvidia, and intel that their future plans of doing processing on the GPU was all wrong, and the the CPU really was faster all along.

Probably a tough sell, but you might make millions if you can convince them?

How long until all games are converted over to run just on the CPU for the added performance you have displayed here with your...
31765153303077535897513456514742416740151747072083 91018699899932793649108926879247397058141528555439 65954222603919059265825637344676406359525838966981 51198395988660368375304201799032818594556941255051 90663028548695333776829846000318080938221300381022 14387057461181304251961916405970456035183121708151 65864735655654053292841174862895708285679230005352 58463770612805914520355463899321278759063496278379 75871352588618213252263577038396202737385324908353 68049799008570152248330343952519734465334299465256 52360967428345505237397339023742618088717992837222 85366293439240895762913154442106573609205481842139 36589386771554284247727510016673435774309363894844 45647643771840738743794710078671510704495546576262 81566137550730763768080600031844296233977808233311 35978757713698301281757162567168328728151193733668 57894371090977485812228681268241223172726811849752 07863453107495331708260153159440253645365524453587 95203474521342924891664450480435535228197772198197 18690548841768963987827047820661269214725486182478 59626434279190274503452994769367997217285165465591 79947178906788568727857447008428972377823476308074 09195129662383464278396538650173246658501921440916 94630371265581197700774682562035198318782913591013 99781730363517376470671438399281029122446084832051 89832483488551310255397215831849316536707322731729 95431750775475634748127320956655431851879586978172 49172170086576809890832783083824043773797445534252 56887128988555131809670124978594542906096273705906 59970784172738420721605576789060565167694565490120 38816577586193923092436298338954985727987452339809 04998584674848503995091093988342104246931136178759 78611803096108774362764990414655167545507613665725 91499337611434024376291029038413588853131259113254 48492258960071848511693901939854346494154837823383 02531368775990005443722332901462568184095998830522 52158532859983399033659541893269668016326589935823 46632470803240204297913574257554985493728961920916 'ssss

I still think GPU computing is the future. :up:

at least my GPU is bloody fast at folding@home... I like that together with my 8 threads daily :)

Go out into the sun Sir Francois... !!

Quote:

---

Originally Posted by Talonman

You should probably tell ATI, Nvidia, and intel that their future plans of doing processing on the GPU was all wrong, and the the CPU really was faster all along.

Probably a tough sell, but you might make millions if you can convince them?
I still think GPU computing is the future. :up:

---

Your really do see things a as absolutes as black & white to the point that you put words into peoples mouths.

All i see is DRWho saying that the CPU is still faster at some things & the GPU is not suitable for everything.

And then you turn around to make out like he's been saying that nothing should be done on the GPU.:shakes:

I am actually sitting here hoping he post some more numbers....

They make for such a good read. ;)

Here is one...

10000000000000000000000000000000000000000000000000 000000000000000000000000000000000000

That's a big one too. :)

"the CPU is still faster at some things & the GPU is not suitable for everything."

That i agree with.

Let me say first of all I'm guessing some of this based on what I've read from the CUDA manuals (I'm no programming expert obviously). If I get this all wrong, then correct me politely; I'm always happy to be corrected as I don't want to be wrong as it were. That being said, the factorials calculation can be computed in parallel, but you'd have to break up the problem and be able to keep track of what has/hasn't been calculated, which would be more effectively done on the CPU. Issuing the commands to multiply x*x to each unit is really easy after that point (and would be faster on the GPU). Data flows one way in the GPU, so you can't really go back and effectively reference what has and hasn't been done. So, yes, if you try to calculate the factorials ONLY on the GPU, it could be slow due to the initial pre-processing. The actual calculation would indeed be fast on the GPU. By effectively balancing the use of the CPU and GPU in tandem, you can likely break down the problem in to chunks, maximizing the strengths of each processor types. The balancing of course being the key, and requiring the PhD that Francois speaks of...

Quote:

---

Originally Posted by Talonman

You should probably tell ATI, Nvidia, and intel that their future plans of doing processing on the GPU was all wrong, and the the CPU really was faster all along.

Probably a tough sell, but you might make millions if you can convince them?

......

I still think GPU computing is the future. :up:

---

why tell intel that they run x86 on the shader, and i think that amd is working to that also

I bet both of those GPU's will be faster than a CPU at Factorial calculating too.

I just don't think you can make a blanket statement that CPU's are faster than GPU's at Factorial Calculating...

What CPU, what GPU?

Can we say that Nvidia GPU's are faster than ATI GPU's and not take some flack? A review on performance should be more specific.

Yes, some CPU's might be faster than some GPU's. I can acknowledge that.
My Q6600 probably would beat my 8600 GT...

Quote:

---

Originally Posted by xVeinx

Let me say first of all I'm guessing some of this based on what I've read from the CUDA manuals (I'm no programming expert obviously). If I get this all wrong, then correct me politely; I'm always happy to be corrected as I don't want to be wrong as it were. That being said, the factorials calculation can be computed in parallel, but you'd have to break up the problem and be able to keep track of what has/hasn't been calculated, which would be more effectively done on the CPU. Issuing the commands to multiply x*x to each unit is really easy after that point (and would be faster on the GPU). Data flows one way in the GPU, so you can't really go back and effectively reference what has and hasn't been done. So, yes, if you try to calculate the factorials ONLY on the GPU, it could be slow due to the initial pre-processing. The actual calculation would indeed be fast on the GPU. By effectively balancing the use of the CPU and GPU in tandem, you can likely break down the problem in to chunks, maximizing the strengths of each processor types. The balancing of course being the key, and requiring the PhD that Francois speaks of...

---

That sounds correct to me, for what that's worth. ;)

"By effectively balancing the use of the CPU and GPU in tandem, you can likely break down the problem in to chunks, maximizing the strengths of each processor types."

That is some big truth right there... :up:

Quote:

---

Originally Posted by Talonman

I still think GPU computing is the future.

---

The GPU is going to become part of the CPU.

I think much of this push for GPU computing is just marketing on the GPU makers part as they desperately try to stay alive and be relevant.

A standalone GPU company like Nvidia won't even exist in 10 years time.

Quote:

---

Originally Posted by Chad Boga

The GPU is going to become part of the CPU.

---

i could see this becoming the norm in low power and mobile devices but with large caches and multicore cpu's they really dont have the die space for a gpu, especially a high performance nvidia gpu.:stick:

Quote:

---

Originally Posted by Chumbucket843

i could see this becoming the norm in low power and mobile devices but with large caches and multicore cpu's they really dont have the die space for a gpu, especially a high performance nvidia gpu.:stick:

---

That is the case today, but in three or four shrinks time?

Quote:

---

Originally Posted by trinibwoy

Clock #1
Thread 1: 10*9
Thread 2: 8*7
Thread 3: 6*5
Thread 4: 4*3
Clock #2
Thread 1: 90*56
Thread 2: 30*12
Clock #3
Thread 1: 5040 * 360
Clock #4
Thread 1: 1814400 * 2

---

Actually, after looking at the threadization of the factorial, it is far from being trivial. KNowing that you have to deal with numbers that are way behond what the CPU or GPU registers can deal with ... Close to a million digits ... It is not as simple as explain on the quote ...
you got to write many loops doing multiplication between partial thread results, and multiplying numbers with thousands of digits with each other ... managing "carry" at each step ...
I am telling you my friend, this is not going to be even 5% of what the CPU is doing ... :rofl: :rofl:

Think about it, your (100000-2)! is 8.2639316883315556986809062972533 E+5565701 using Stirling's approximation ... that is a lot of Digits ... and you have to multiply by 999,999 then by 1,000,000 ...


you don t have enough memory on the GPU ... say :welcome: to Memory Swapping though PCI express ...
The fastest way to do it is use the Int registers .. and threads reasonably.

So, today, I demonstrated that it is easy to claim that the GPU can do well on everything, the reality is that there is a majority of case where the GPU will NOT be even close to a regular CPU... with some memory :-P
AMD did the right choice.

:D :ROTF: :rofl: :up:

And Intel just shot itself on the foot. :rolleyes:

Quote:

---

Originally Posted by Calmatory

And Intel just shot itself on the foot. :rolleyes:

---

well, honnesty is great no?
Actually, I believe that x86 will outperform OpenCL, using the same software architecture, so AMD and Intel are almost doing the same thing. AMD for the moment is not going x86, I would not be surprise that they do it one day for the GPU.

Francois

Quote:

---

Originally Posted by Drwho?

well, honnesty is great no?
Actually, I believe that x86 will outperform OpenCL, using the same software architecture, so AMD and Intel are almost doing the same thing. AMD for the moment is not going x86, I would not be surprise that they do it one day for the GPU.

Francois

---

i dont think it is possible to make a general statement like "x86 will outperform opencl".

from what i understand, the unique feature of opencl is that it can run threads on the cpu and the gpu. so run iterative/branching/recursive stuff on the cpu and use the gpu for stuff like matrix multiplications.

of course in a commercial environment it is also important how quickly you can get things working. your factorial is a great example for that. 1 hour to write cpu code, weeks to write gpu code. coders cost money, time you often don't have, hardware is cheap, just get a few dozen boxes running the cpu implementation