reminder that this beauty is ruined because devs were consta

>>56947576
also CELL was infinitely extendable due to its nature. you had to stage your own pipeline to distribute tasks into the SPU cores so you were not limited by hardware pipelining that might require special optimizations for specific situations.

>>56947576
>Last of Us
>30fps semi ugly garbage with nearly everything scripted
wow buddy that shit didn´t make proper use of the cell cpu in any way and that shit was first party

but yeah it had potential

>>56947576
But isn't ps4 also 8 weak cores?

>>56947618
it's x86-64 general purpose cores. nearly identical to standard PC.

>>56947625
And?

It's still 8 cores and op is bitching about multicore programming, not x86 vs cell.

>>56947576
what was it and why should I care?
If they couldn't into programing so bad that they went extinct, I really doubt it was a "beauty" and it probably shouldn't be missed.

>>56947638
completely different architectures. SIMD matters.

>>56947647
So is op (you?) complaining about multicore programming or CPU architectures?

>>56947638
It's an entirely different concept. PS4's CPU just has 8 AMD64 cores, the Cell has a single PowerPC64 core with 8 special cores. They're more like GPU cores, actually probably a part of why the Cell isn't really a thing now. Cell came out right around the time GPGPUs started making leaps and bounds.

Fun fact: Sony initially intended for the PS3 to not have a GPU and just use the Cell for everything.

>>56947658
both

GPUs are way more parallel than Cell and people manage to program them somehow.

>>56947673
it was around 2005 when people complained. there were barely dual core cpus around for consumers.

Good instruction set for HPC and graphics tasks.
Shit implementation due to small amounts of local store and terrible C/C++ support.
Also the EIB was not that great as a bus.

Best instruction:
SHUFB
Close second:
ROTQBII

>>56947669
No it isn't.
You've gone from bitching about people not taking advantage of multicore to bitching about a change in cpu architecture.

Is this damage control from you not realising that ps4 also has 8 cores when you made this thread?

>>56947700
I'm not OP you fucking retard

If clang&LLVM was around when cell was released, it would all be different. because devs mostly suffered from lack of proper compiler optimization and debugger tools. it would all be different if a specialized compiler with strong backend like LLVM was used.

>>56947707
Then stop defending that idiot.

I'm not having a go at cell or calling it inferior so don't get defensive either. I'm pointing out that op bitches about the lack of multicore processing when in reality modern consoles also have many cores.

>>56947683
So I guess it all turned out all right in the end?

Now we all have quad core CPUs and GPUs with hundreds of cores, and software is halfway decent at using it. It's just like you wanted, just not with the Cell in particular, and it took us a few extra years to get there. Hell even our phones are quad core.

>>56947731
My phone has 8 cores :D

>>56947669
Right, in 2005 GPU compute wasn't really a thing yet, so having massively powerful CPUs was still important in console design. Sony still dropped the ball on not providing documentation, splitting the memory for the CPU and GPU, even though both were used for graphics generation, and not having enough RAM. Had only they initially dumped useless "home media" features for a larger pool of unified RAM and actually provided dev tools, history would've gone way differently. Otherwise it's practically a Sega Saturn that managed to succeed.

>>56947750
D:<

>>56947771
>:D:<

File: 400px-AmdahlsLaw.svg.png (22KB, 400x300px) Image search: [Google]

22KB, 400x300px

>>56947673
Most workloads aren't highly parallel. This is the problem with Cell and more cores in general.

Do people actually believe that the cell was some magic chip that if programmed right was equal to ten quantum computers?

>>56947821
Of course not.

>>56947821
people literally (

L I T E R A L L Y

) built clusters based on ps3s just for pure power on streaming data.

>>56947802
why isn't 2 cores, 50% parallel portion x2 speedup? all the parallel code should be running on the second core, leaving only the unparallel part of the workload on the first core, effectively halving the time taken
>>56947821
at th time it had massive theoretical throughput considering the price of a PS3

>>56947765
>Otherwise it's practically a Sega Saturn that managed to succeed.

Sony thought they were on to a winning formula.
The PS3 is bascally an extension of the "spirit" of the PS2 in programming terms.

Relatively underpowered CPU and GPU combined with massive vector units and a asynchonous architecture based on DMA.

>>56947833

https://en.wikipedia.org/wiki/PlayStation_3_cluster

File: 1466536908700.jpg (498KB, 1200x900px) Image search: [Google]

498KB, 1200x900px

>>56947821
>>56947833

You're not realizing that companies have an absolute lack of concern for quality and technology, they'll literally do the slowest and buggiest piece of shit if they think it'll sell and not get people dissatisfied enough to not buy

never hire pajeets.

>>56947700
>You've gone from bitching about people not taking advantage of multicore to bitching about a change in cpu architecture.
>Is this damage control from you not realising that ps4 also has 8 cores when you made this thread?

>YOUR USERNAME IS "Anonymous" THEREFORE YOU ARE THE SAME PERSON

back reddit you halfwit

>>56947765
>Right, in 2005 GPU compute wasn't really a thing yet, so having massively powerful CPUs was still important in console design. Sony still dropped the ball on not providing documentation, splitting the memory for the CPU and GPU, even though both were used for graphics generation, and not having enough RAM. Had only they initially dumped useless "home media" features for a larger pool of unified RAM and actually provided dev tools, history would've gone way differently. Otherwise it's practically a Sega Saturn that managed to succeed.
Didn't they plan on using the cell as the GPU originally? The original plan was multiple cells could connect together and share the workload. Back when it was supposed to do 1080 60fps there was some plan to include cells in Sony TVs. In theory they would network together and you would get better graphics.

Then the entire plan fell apart and they had to run to nvidia for a GPU at the last minute.

>indians
>game development
Anon, you're a fucking retard. Indians think vidya is for babbies and want nothing to do with it.

>linking to MIT
>implying you're not an MIT nigger

>>56948879
because IBM fucking shits failed so hard and hyped the CPU so much. also they gave this supposedly secret project to microsoft for xbox project. so microsoft got their 3 general purpose cores without investing a single dollar for R&D. IBM deserves a huge punishment. both for failing to deliver and breaking NDA.

>>56947833
Thats because what CPUs designed around signal processing are good for. Know what they aren't good for? Vidya games

>>56949030
PS2 clusters were built too

the only thing that is wrong with cell is ibm never really provide a good low level API to properly utilize its SPE, unlike intel or amd that actively developed directx and vulkan to get the most of their hardware

>>56949036
Ps2 was also a giant pain in the ass to get normal software/games to have performance and feature parity with Xbox/GCN. The only reason devs bothered to invest money into doing it was the PS2s massive install base

>>56949046
this.

>shitty compiler support
>shitty optimization
>shitty and VERY shitty documentation
>shitty api
>failing to introduce developers the new architecture and patterns&pitfalls for it

it took years for companies to wake up but it was too late for cell.

i once read a post on anandtech which sadly gets edited/remove his criticizing post about how cell is actually a single core cpu that is only manage to score 1005 score on geekbench, about the same score as the score of a 1,6 ghz powermac g5 or 2,8 ghz pentium4. the comments in that article are attacking how geekbench arent legit benchmark since it doesnt recognize SPE as integer cores

File: playstations.jpg (42KB, 380x253px) Image search: [Google]

42KB, 380x253px

>tfw one of the top 100 supercomputers is 1760 ps3s wire together running linux

you don't even need a CS education, just the ability to put two and two together

>>56949109
>geekbench
there's your problem
>>56949112
the reason they're probably not more common is reliability, you can't rely on your calcuations being correct because >no ECC etc.

>>56947883
50% of the code is sequential
50% of the code is parallelisable

You run the parallelisable section of the code on two cores so it runs twice as fast. The total runtime is now 75% of the original runtime.

Speed-up = 1/0.75 = 1.33 which is what is shown (correctly) on that plot.

>>56949202
>You run the parallelisable section of the code on two cores so it runs twice as fast. The total runtime is now 75% of the original runtime.
what no, you run the sequential part of the code on 1 core and the parallelisable part on the other

>>56947802
This guy gets it.

>>56947673
You never used CUDA, have you? It is a fucking pain to program stuff to run on GPUs.

Most graphic processing peoples are embarrassingly parallel so it is worth effort of programming GPUs to do graphical operations (hence the name). The vast majority if general purpose computing problems are not embarrassingly parallel so it is not with the effort.

>>56949237
Bait? You cannot run anything on the second core while the sequential part of the code is running... that is why it is called sequential.

http://www.giantbomb.com/videos/an-uncharted-tech-retrospective/2300-11203/

might be of interest

>>56949265
the 2 cores are independent.
sequential, to me, implies that it's single-core
parallelised code is code that can distribute its workload across multiple cores.

File: pixel1.gif (1MB, 480x270px) Image search: [Google]

1MB, 480x270px

>>56947669
> special cores

>>56947576
But the ps4 also has 8 cores?

>>56949380
read the thread 74gg0t

>>56949387
It's too long and I don't understand this op. He's complaining about multi core processing not taking off when it clearly did. I don't he understands the CELL very well or what made it so difficult to develop for.

>>56949112
How do you even wire multiple PS3s together? What cable and what port?

>>56949335
The sequential portion of the code is the portion of the code which can only run on one core at a time (ie cannot be parallelised). The parallel proportion of the code cannot be run while the sequential portion is executing.

>>56949438
probably connecting them all into a single ethernet switch and use cluster tools like spark or something that does task balancing across network. not sure tho, I need to be enlightened.

>>56949438
They use the Bluetooth obviously.
What the fuck do you think, numbnuts?
It obviously uses the gigabit ethernet connection.

>>56949438
Back when Linux was still available for install it's all just patch cables connected to routers and switches.

>>56949464
So Bluetooth or Ethernet?

>>56949464
>They use the Bluetooth obviously.
Nothing stopping you

>>56949019
MEME 8/9
4 times the price of overpirced intel xeons.
no thanks

File: 1450607694977.gif (415KB, 307x277px) Image search: [Google]

415KB, 307x277px

>>56949335
But since the total workload both share is being distributed to the two cores, the sequential code is technically parallelised, which is a contradiction.

>>56949453
Isn't the old ps3 100mb only?

>>56949453
>>56949526
consider the following:
processor 1 starts a parallelisable workload on all other processor(i. e. processor 2 in a 2 processor system, processors 2-4 in a 4 processor system etc.). Now we start a workload that cannot be parallelised on the main processor. After everything is done the program is finished.
Let's say our system has 2 cores. core 1 executes the sequential workload in 5 seconds and core 2 executes the parallelisable workload in 5 seconds.
Total running time is 5 second.

http://www.wsj.com/articles/SB123069467545545011
read this, it's interesting

>>56947700
>Is this damage control from you not realising that ps4 also has 8 cores when you made this thread?

Not him but, I understand the point, nowadays multi cores are standard, back then they weren't, so the CPU got shat on because "eh, this will never be a thing" and nowadays those developers are jobless or don't want to remember that they said that.

Nothing to do with PS4

>>56949484
Can you imagine though. WE BUILT A SUPERCOMPUTER WITH BLUETOOTH. Someone do it just for the absurdity.

>>56949717
>To Read the Full Story, Subscribe or Sign In

Jesus why is this even allowed?

>>56949717
run into a wall faggot

>>56949697
What the fuck are you not getting about the concept of sequential?

By definitions the sequential portion runs on one core and only one core. The second core lies idle while the sequential portion is executing.

If the second core is able to run while the first is running them that section of the task is by definition a parallelisable section of code and hence not sequential.

>>56948879
>Didn't they plan on using the cell as the GPU originally
Yes
>The original plan was multiple cells could connect together and share the workload. Back when it was supposed to do 1080 60fps there was some plan to include cells in Sony TVs. In theory they would network together and you would get better graphics.
No

>>56949747
The PS4 and XBONE processor gets shat on because only a moron would chose a 1.6Ghz 8 core processor over a 3.2GHz 4 core.

>>56949814
cell is technically a single core cpu, with 8 co processor, and x360 cpu are three cores not 4, and no, the 8 core jaguar will significantly outmuscle those gimped cpu of a console from more than a decade ago

>>56949112
I'm genuinely curious how the contract for those handles replacements.
They can't just pull a load of retail units from the warehouse and send them because they killed linux.
Which leads me to wonder if they resigned an old firmware with new keys, have ancient stock on standby with the means to do the above if it runs out or they just give them bespoke firmware variants of the latest.
Either way really I want a look at what firmware they've got.

>>56949783
>If the second core is able to run while the first is running them that section of the task is by definition a parallelisable section of code and hence not sequential.
what happens if i keep adding cores to the system? according to amdahl's law, the runtime of the parallel workload decreases?

>>56949905
Did you see where I mentioned Cell or Xenon because I sure don't.

>>56947884
>sony thought they had a winning formula

Didn't they though? On the whole, the PS4 was more successful than the 360 iirc, and the products released on it seemed to perform slightly better on average. Seems like a relative success to me.

Unless you're talking about the engineering paradigm on a whole.

>>56947802
Most workloads yes, but it just so happens that many aspects of modern 3D games are independent from the engine. Physics, sound, lighting, drawcalls, etc can all be spun off onto different threads that run independently from each other.

>>56949112
https://therearenosunglasses.wordpress.com/2009/12/11/why-does-the-air-force-want-thousands-of-playstations/

>>56949335
> you can compute x+y before neither x or y are known

>>56947601
Daily reminder that a game like that could run on a cheap 90nm cpu+gpu built in 2006

>>56947576
This CPU is useless for Gaymen.

>>56950037
you can certainly compute x^2 and y1+z1 y2+z2 y3+z3 etc.
who says that both have to be dependant on another?

>>56949972
sony got cocky as fuck after having the best selling home console of all time, which is why the ps3 was a disaster. it's a miracle that they were even able to salvage it.

>>56948816
>can't refute the point so resorts to ad hom

>>56950069
This, in the original plans for the PS3 (the one where they would use multiple Cell processors as GPU and CPU) it was aimed as a multimedia device like the PSX before it, a DVD player that could play PS games. After the reveal of the PS3 devs like Square Enix went to Sony and said they wouldn't make games for PS3 if it wasn't being sold on gaming capabilities.

>>56950068
> you can compute operations on values that arent even fetched yet
Yeah sure and how expect you for it to happen? Prediction? A 1/2^32 chance of hitting the right value just for single integer sounds very good.

I get it you are from /v/ right?

>>56949697
No, you don't understand. If the sequential code of a process can run parallel to any other code, by definition that code is parallelised and not sequential.

CELL WAS THE BEST THING OF ITS TIME

FUCKING PAJEET DEVS WERE TOO LAZY BECAUSE THEY WERE STRANGER TO MANUAL CPU PIPELINING, TASK DIVIDING AND WORKLOAD BALANCING, COMPLETELY STRANGER TO MULTITHREADED ARCHITECTURES

NOW THEY ARE ALL UNEMPLOYED AS MULTITHREADING WAS REQUIRED MORE AND MORE IN THE FOLLOWING YEARS

HA HA HA

File: PaintDotNet_2016-10-06_16-26-49.png (22KB, 607x220px) Image search: [Google]

22KB, 607x220px

>>56950215
>I get it you are from /v/ right?
nice ad hominem

>>56950829
what the actual fuck is that shit

File: Screen Shot 2016-10-06 at 11.12.38 AM.png (2MB, 1904x1394px) Image search: [Google]

2MB, 1904x1394px

>>56950829
>BCU-100
well I'll be damned. That's pretty interesting.

>>56947647
SMD

>>56950978
What the fuck am I even looking at, are those little CPU looking deals what RAM used to be?

>>56950988
The XDR? Nigga can you read? It is RAM.
>>56950930

>>56950930
Damn do I ever want one of these

>>56951728
I'm sure most of them have either been retired or scrapped by now.

Check Ebay, I bet you could find a few for sale pretty cheap.

>>56951728
>>56951740

It was sadly never put into production

>>56952109
Damn. I had hoped with the fact that people built clusters out of the PS3, that there was someone crazy enough to have rigged up a super-computer using these things.

>>56947576
This desu.
I don't know why people don't just use erlang for everything.

File: HSA-enabled_integrated_graphics.svg.png (51KB, 1024x576px) Image search: [Google]

51KB, 1024x576px

>>56947576
Cell programmability problem wasn't being multi-core, it was that SPUs weren't even real cores and didn't have transparent cache coherency.

You basically had to figure out how to chop tasks into pieces at the high level then juggle DMA transfers into and out of the fuckers all day long.

It's not a coincidence that MS and Sony went for coherent HSA designs this generation.

whats awesome about the cell is how it manage to run gta v with only slightly less fidelity compared to the pc and ps4 port, like how the fuck they manage to cram a games that eats 4gb vram on my pc into a system with 512 system ram and 32mb vram

>>56947576
I still remember when the PS3 used to hold the best scores on Folding@Home

File: 6a0120a85dcdae970b0120a86d7ec8970b-pi.png (5KB, 639x434px) Image search: [Google]

5KB, 639x434px

>>56952859
Forgot pic

>>56949046
they have pretty good reason to not waste R&D on a proper API for cell, the market for cell outside the ps3 is pretty much non existent. the HPC market that is originally the target market for cell are refusing it. sure they can argue which one first, egg or chicken, but with the looming intel core architecture, its very hard to convince the target market to choose cell since they will spend another dollaridoos to learn how cell works and optimize it