The rise and fall of the PlayStation supercomputers
theverge.com
theverge.com
As a result, the PS1, but especially the PS2 and PS3 were engineered to have amazing vector capabilities in the CPU. The PS2's Emotion Engine had extremely high performance vector processing units, and was so optimized for compute throughput it used two different floating point formats, neither of which are compatible with the IEEE754 standard that every other piece of hardware you've ever touched uses. The PS3 was the culmination of that philosophy, with the Cell being explicitly developed to put as much raw number crunching hardware in something that can still be called a CPU. It was just a single unimpressive POWER-compatible CPU core, and 60% of its die area went to specifically engineered number crunching units. This is what made it "supercomputer grade".
There's a persistent rumor that the Cell was so powerful that Sony had intended for the Playstation 3 to be equipped with two Cell chips, and no true GPU at all. But it seems Sony realized during development that their philosophy had no future. Sony compromised and introduced a standard Geforce GPU. Compared to the more conventional architecture of the Xbox 360, the compromised design still had slightly more potential but it was a lot more difficult to work with for games, and more expensive to produce as well.
Starting with the last generation of consoles, homegrown graphics architectures aren't tried anymore. You can only get performance and especially good developer buy-in on commodity hardware. As a result, the PS4 and Xbox One have PC hardware. Their successors will too. The Nintendo Switch shares a chip with the Nvidia Shield TV and Google Pixel C tablet. Console hardware is not really interesting anymore if you ask me.
Agreed. And that’s a good thing.
PS1, PS2, and PS3 were all underpowered. I’ll never forgive the PS2 for holding back games for 7 years. Grumble grumble.
PS3 may technically not have been slower. But if it takes 2x the effort to get a 10% improvement it is in practice.
Everything’s a PC now. Which means the focus can be on making good games instead of optimizing for specialized hardware to serve a fraction of the audience.
The PS2 for sure was less powerful than the more conventional architectures of the GameCube and Xbox, but it was released 18 months before either of them in a time when Moore's law was still extremely relevant. Yet it could still hold its own against the competition. All pieces of impressive hardware.
PS2 dominated sixth gen sales. By being the winning console but releasing 18 months earlier it held back everyone. And since, like you said, Moore's law was very much alive this represented a significant reduction in overall quality for an entire generation. :(
PS3 at it's peak was great. But it took waaaay too much effort to get there making it a very poor value proposition for developers. And 7th gen was a fully multi-platform era so the end result is the average game was held back and only first party prestige titles could justify the extra cost.
It’s funny that this is always a criticism levelled at the CPU and not the developers.
Assume you have a developer who is one of the best in the world at writing hyper-optimized SPU code. Imagine this dev WANTS to write micro-optimized assembly. (And there are many such devs). This developer's time would far more efficiently spent if they were writing "normal" code for a "normal" processor.
So yeah. The criticism belongs squarely on the processor. :)
Powerful custom hardware can be great for the developers that invest in it, until it isn't anymore and now you've got a bunch of sunk R&D. And then there's the need to ship on more than one platform...
And even if everyone were a super smart 10x-developer like you surely are, their time would still be better spent on a platform that is more productive. The idea that some technology is difficult but rewards proficiency with incredible productivity is almost entirely wrong, which is why nobody does web development in Assembly. LaTeX vs Word would be the only example I would have agreed with in the past, but if you search you'll find a study showing even the most proficient practitioners are slower and make more typesetting mistakes using the former.
That's not called for. They never implied they were even above average. Just that a bunch of game devs were somewhat lazy.
That's not called for. You don't know those developers, you don't know what work was on their plate or what constraints they were toiling under.
But they still weren't calling themselves some kind of better-than-everyone dev.
Though if you're using "developers" to mean that entire section of the company, including management, then what was on their plate and the constraints are almost all self-inflicted.
(The lack of ram was definitely the biggest issue, and the thing that held back PC games the most as levels had to be made a lot smaller than they could have been otherwise - Destiny and Final Fantasy XIV are two concrete examples, both released expansions with much larger playable areas as soon as PS3/360 support was dropped).
Compared to other consoles of the same generation they were on par though.
Compared to top-of-the-line PCs of the same era yes. But the GPU alone of "PCs of the same era" would set you back as much as a 6th gen console, and you'd have a shiny graphics adapter sitting in a box.
The 7th gen is where consoles started really falling behind (though somewhat ironically also where MS decided to go with somewhat more bespoke hardware than the "PC in a box" Xbox).
The most serious competition at the time was the Nintendo 64, and it lacked storage.
From your comment, it seems like Sony tried exactly this but failed. Is there an inherent flaw in such a design or just poor execution?
Edit: the biggest problem I see with such a design is RAM IO, but I would think that could be mitigated. Maybe even eliminate support for DDR, instead embedding 4 or 8 GB in the package.
From what I understand this is still different from the Sony approach where you'd have a weaker GPU, but would offload shaders etc to the CPU which could do this due to it's powerful dedicated vector processing units. Of course this had the flaw of GPUs eventually develping shader capabilities etc and making it pointless - and harder to program for if you have to target PC/xBox etc as well.
BTW the latest AMD GPU architecture (RDNA) is specific for gaming, in that they get around 40% better framerates for the same raw GFLOPS of their previous architecture (GNC).
It was a different world though, PowerPC was still alive and strong back then.
The PS4 does continue the trend but I think that's more a question of time and investment rather than pure tech considering they have a steady trickle of top notch exclusives.
The different ARMs don't even share a common instruction sets, I think.
One thing that made it practical to work with was that we could install Linux on a PS3.
When Sony retroactively killed the ability to install Linux, they did more than make PS3 irrelevant to supercomputing. They made me deeply skeptical of relying on a technology with a single-vendor gatekeeper.
I'd love to poke into it myself. The only barrier is the language you have to use to speak with it.
But on benchmark levels Cell would be competing on different grounds to an X64 and a few PCIe slots right? I've not a clue, I'm curious but I know I don't really know what I'm talking about after this point.
the cell really taught them the trade off between latancy and throughput. That by batching up computations they could do orders of magnitude more than naive pipelining. But, to get that good batching, they had to compromise on framerate.
Which gets to the second big takeaway, which was that standard deviation in the framerate was more important than the framerate and by keeping the framerate low they could also drastically reduce standard deviation in the framerate (Incidently I understand that this was something the amiga demoscene really understood, and low S.D. was a must for any text marquee to look completely smooth).
> There's a persistent rumor that the Cell was so powerful that Sony had intended for the Playstation 3 to be equipped with two Cell chips, and no true GPU at all. But it seems Sony realized during development that their philosophy had no future.
The original plan was for no GPU at all, relying on the SPUs as pixel processor units. When they realized that that wasn't feasible for a few reasons, they quickly partnered with Toshiba to make a custom GPU, and that ended poorly too, so Toshiba scrambled and licensed an NVIDIA core. The final "RSX" chip was still a Toshiba design, but with NVIDIA at the center.
As games got more and more complicated, the amount of developer support & tooling did as well. The Xbox 360 had incredibly strong tooling (ask someone who worked on the device about 360's PIX, and they'll gush for days), and the PS3 had very little. The PS3 shader compiler infamously had a random search for an optimizer, and Sony's recommendation was to just run it continually with different random seed inputs in hopes it found a better optimization for your shader.
MSX, PC98. The Nintendo uses a standard 6502 CPU with a PPU and the Mega Drive used the uberknown M68k.
Sounds like machine learning with linear regression.
TBF until the early aughts bespoke hardware was the only way to hit a suitable price target and get the performances and graphics snazz you wanted.
The first "consumer" PCs which were competitive with consoles in terms of gaming (late 90s, starting with the Voodoo 2 or so) had the graphics adapter alone costing as much as a current-gen console.
and Hallelujah to that
Praise Talos
We really do not need a Wakanda with an independent history of processor technology. Maybe 50 years ago the Cambrian explosion of possibilities was relevant, but this century?
Consoles had to compromise based on heating and costs while being competitive in graphical capabilities, those things haven't been issues that the market actually cares about in 15-20 years. For the market, the graphics pale to the actual experience, they're good enough and would never attract enthusiasts that are willing to use more electrical energy for bragging rights.
General processors and off the shelf graphics processors are good enough.
Interchangeable parts wins again.
Does this mean the Xbox was pretty revolutionary for showing that you really can make a console with commodity hardware?
No Windows Phone iteration ever had more than a few percentage points of share. iOS and Android both clearly destroyed it in the market. Had WP seen a few years of sustained success it would still be around today, sort of like how BlackBerry the brand is still around, albeit for Android-based devices.
It was likely more about making it easy to port existing windows directx based games quickly and cheaply. Similarly, it granted a development stack that a great many developers would already be somewhat familiar with. All good things for a new entrant to a tough marketplace.
Presumably it also allowed them to get to market faster? Much less custom silicon?
Meanwhile off the shelf hardware is cheap and has support from developers. And any kind of enhancement to consumer products involves cheap hardware to keep costs down.
I couldn't help but wonder if it wasn't a functional and yet a bit of a corporate vanity project that had missed the boat and had nonsensical ambitions.
As such... its sad as we will never see any new crazy ass hardware architectures going forward.
Instead they cowardly remove it due to a 'hack' by Geohot and locked it down for 'security purposes' and potential researchers, scientists and those without access to supercomputers are the ones to suffer from this decision.
Sony had to appeal to their gaming community over the science community so that's that.
Also it's 'OtherOS' not 'OpenOS'.
SONY is infamous for going overboard on DRM.
A lot of things can be done given unlimited time and resources, but unfortunately we don’t have that.
Sony also demoed rack-mount with Cell (https://en.wikipedia.org/wiki/Zego) but it is not clear that it was available in open market
I'm sure that military personnel enjoy games as much as the next person, but it's probably not why they bought the machines.
https://phys.org/news/2010-12-air-playstation-3s-supercomput...
Also, if you run the math in the article you linked, it may suggest they paid more than list price.
> core made of 1,760 Sony PlayStation 3 (PS3) consoles
> the PS3s for the supercomputer's core cost about $2 million
that's $1,136 per unit.
Company I worked for in the past had a significant line of business with a major post prod company in the valley that was just buying hundreds of DVD players off ebay at a 20% mark up and then drop shipping them.
There's tons of companies and government agencies that cant just send someone to bestbuy with a visa to make purchases. Has to be a preapproved vendor and has to be on at least net 30 terms and purchased via PO. There's a premium to be paid to companies/vendors that can work within your companies processes to quickly get you what you need to not block projects.
I do wonder if PS3 have GSA pricing though.
The consoles that existed in the 70s and 80s were largely justified in being specialized hardware specifically for realtime rendering which had no realistic application outside the narrow scope of fixed function realtime interactive rendering they produced. In the 90s the PS1, N64, and Dreamcast were all computer adjacent with the next geration - PS2 / Gamecube / Xbox - all being perfectly functional general purpose computers that were simply crippled with proprietary software into being toys.
Since then their hardware has only gotten more generic and general purpose and the degree of their draconian platform lockdown grown to an insane degree to resist emulation and unlocking. Its a shame Valve only ever saw SteamOS as a hedge against Microsoft - if they had taken it seriously and put in the legitimate ridiculous amount of money it would have taken to breakout a console platform without the horrifying amount of proprietary bullshit going into the competitions products we might have seen that incredible amount of waste become as taboo as it deserves to be.
Honestly the worst offender is the Nintendo Switch - it is verbatim just the successor device to the Nvidia Shield Tablet, just with a grotesque proprietary operating environment and impenetrably locked bootloader. Every time I see one of those things I am struck by how absurdly wasteful it is to see someone carrying that thing and a regular tablet when Nintendo, Sony, and Microsoft could all just be shipping their own proprietary DRM laden store apps on top of a freedom respecting software foundation. If you want to exploit people by depriving software freedoms while they pay you for it... I'd argue its a product of propagandized deception and manipulation, but its still people considered informed consumers buying that garbage. Its a lot less bad than entire computers, fit for a multitude of purposes, being reduced to toys.
I think this is a bit harsh. Is artwork a waste of good canvas?
(I guess to be fair to you here, at least canvas is a renewable resource)
I was trying to come up with some convoluted analogy about making art on a gold plated easel that self destructs every decade and burns any canvas besides a specific size that is put on it, but just trying to formulate a physical analog hints at the absurdity of it all.
Actually, wait. You buy a tractor but you can only use it on certain days of the week and it permanently shuts off after 3 years and stops working if you try to attach a snow plow to it. Except that is already the reality of John Deere products being DRM locked and not owned by the buyers.
Driving demand towards the hardware, instead of software and licensing costs on a negative margin product is a pretty sure way to run a company out of business.
They take the air out of the industry with how much money they make operating the way they do. That doesn't make it right, ethical, or reasonable.
But it wasn't, mostly due to the games. Many of them would require emulating a mouse just to get into the game, which meant emulating a desktop. Then the game would start and it would switch over to the controller mode.
Many games just didn't work well under it, and so forth.
Are you saying we should force them with laws?
If so why?
The Nintendo DS, at the time, was unparalleled in terms of software quality, and, while not as portable as a Game Boy Advance, was compatible with my existing GBA collection, allowed me to wirelessly play; and even eventually play online, with friends, and offered an almost dauntingly large library of quality software.
This was before the days of the iPhone, and while I did have a Windows Phone at the time, it never could have, and certainly did not, have the library of software, or hardware design, to consider it a valid console. Being a PDA and a phone was more than enough at the time.
I played ‘Halo’, of all things, on my PowerBook G4 back in the day, saw that as the driver to XBOX, and couldn’t have cared less.
Is it worse to try and fail than to never try at all?
Linux support was not removed due to Geohot. It was a driver and business priority issue. Geohot just sealed the deal.
Having worked there during the kerfuffle, I can confirm their reason was valid for not supporting Linux going forward. It was a non trivial effort and they wanted those engineers to make it support the new hardware turns. It was a company wide goal to hit break even at retail pricing at that time (Sony was amazing at running hardware coat forecasting internally and iirc hit their 7 year Target by a week). This required the new hardware. So they dropped Linux. That part is real.
Then they effed up by not messaging it well and by dropping promised back support.
And then a bunch of the community responded as petulant children and attacked sonys main business and released products that allowed others to infringe Sony and other game studios property with impunity. These actions killed any chance Sony would engage with this community for the next decade or so.
Sony was also backed into a corner technically as I recall as the thing hot hacked (by realizing it was a null crypto key on the one processor), and thus allowed Linux was also the thing that did the drm. Sony had to choose between another year or so of drm or Linux support.
As in many cases not parties are at fault. Contracts can and are broken. Sony paid multiple prices for it. So did the Linux/hacker community.
I recall they offered a "Linux" kit with mouse/keyboard, network/HDD adapter (so limited for fat models only) and a VGA cable to use with a computer monitor with sync-on-green support...
As someone who was on the dev side of the fence at that time one of the areas where Sony was pretty poor was the devtools. Other than some hardware capabilities of the profiler Microsoft blew away everything else that Sony had.
While running Linux on it was pretty neat, I wonder what the opportunity cost was to other parts of the development experience on the platform.
And then a bunch of the community responded as petulant children and
Sony supported linux in public, people bought their hardware for it, and then, without warning, sony janked the rug from under them. Of course people tried to take back their hardware, for which they paid. The blame here is 100% on Sony.Sony has a massive internal conflict of interest by being both a content creator and a hardware provider. They sabotage their own products to the detriment of their customers again and again.
Sony worked with the community for a few years, and had the longest unhacked console to show for it. Then They slapped the community in the face and paid dearly for it, as the should. Sony should beg the community for another chance.
You can apply this to any other infringement on freedom.
Extorting a company by threatening them with release of an exploit if you don't get your way is not legitimate, that's blackmail or extortion.
All of the hackers attacking Sony for months, disabling PSN, is not legitimate, and should never have been supported by anyone.
Just let the courts handle it like they should and did.
What's illegal about reverse engineering a physical product you sell to someone?
> Disabling psn
Are you kidding me? The PSN was notriously insecure. Sony didn't give a crap until they were embarassed.
Reverse engineering legal and to me moral.
Extorting a company with a threat of release of a thing to get what you want: extortion and illegal. This was done by people around hots.
Agreed: psn was insecure. Hacking is still illegal.
By taking these actions we lost the moral high ground as hackers. The class action by itself would be just fine
What sony did is illegal.
I hadn't bought any Sony hardware at that moment so IIRC I just decided right there and then that there was no reason to send money Sonys way or feel sorry for them.
First Sony console I bought was a used PS3 sometime after PS4 was released. And I was enthusiastic about PS when they first released the console with Linux support.
I guess I'm not the only one.
Lack of support for new hardware does not explain why they dropped support for early hardware.
You no longer have the thing you paid for.
Honestly, your attitude is rather shocking.
Which sony regulary acts with impunity of being toxic towards it's users. (One thing to note: you can never get a refund on something you "bought" or something that was fraudulently bought on your PSN account. Do a charge back on a psn purchase, you loose all of your entitlements you purchased before)
Selling a Linux box isn't. One can argue that selling a Linux box at a loss is a dumb idea. One can argue doing so helped super computing to the masses and was a good gesture. However when the reason the box is subsidized is threatened then it's pretty rational to remove the thing threatening it.
This was definitely a lesser of 2 evils for Sony business choice. They misjudged how much people cared about Linux. As others have misjudged other ads like no man's sky.
The PlayStation 3 was cracked wide open [1] without OtherOS and because people were pissed at its removal. On Playstation 3s released in the first 3.5 years of its lifespan you can get full hypervisor compromise without any hardware mods, and you can yet again run OtherOS on the latest firmware. Now with less restrictions than before. Newer PS3s (starting with CECH-25xx) have 'only' been partially compromised. They can't run OtherOS, but they can still run homebrew and piracy despite Sony taking OtherOS away.
Maybe it's a bit of solace that this stuff only punished Sony. But inevitably, the lesson they learnt is "we'll never experiment with this stuff again".
[1]: This video is really worth watching if you're interested in this stuff: https://www.youtube.com/watch?v=DUGGJpn2_zY
This is OT, but I think people here might enjoy it: Geohot doing hackerrank problems for 5 hours straight: https://www.youtube.com/watch?v=EecfVsdQMcM
Wonder what made the writer originally assume that? Mid-generation console refreshes would be the one occurence where much care would be taken to keep the performance not faster, not slower, but completely identical against the original hardware. I remember reading somewhere that the Xbox 360 refreshes had some hardware to intentionally add latency for inter-chip communications to keep performance identical.
https://www.theregister.co.uk/2000/12/19/iraq_buys_4000_play...
To be a fly on the wall during that discussion. I'm sure money wasn't an issue for the Air Force, or that Sony had some balls of steel.
Notice how there's very few military aircraft companies in the USA?
That's because the little fish got a visit from Washington saying, "that was your last contract unless you merge."
Northrop tried to resist, for a while anyway.
Regarding foreign companies, the Avro Arrow was cut up after the US offered Canada a northern missile defense network. The Arrow was an early Mach 2 interceptor - think of the foreign sales possibilities.
Many of the designers ended up at NASA.
The article doesn’t really explain in detail the difference that made the ps3 an attractive buy for making super computers over regular PCs. Can someone elaborate?
> At the time of its inception, its main streaming Cell processor delivered a 20 times speed increase over PCs for some calculations, processing power which could not be found on other systems such as the Xbox 360.[36][190] The PS3's high speed and efficiency introduced other opportunities for worthwhile optimizations according to Amdahl's law
> The PS3's uniform console environment made technical support easier and made Folding@home more user friendly.[36] The PS3 also had the ability to stream data quickly to its GPU, which was used for real-time atomic-level visualizing of the current protein dynamics
This meant you could buy much more hardware for the same cost verse buying the parts individually.
DISCLAIMER: I am not a chip expert. I threw together this summary after reading several articles.
It was crap for creating games (and a pain in the ass to develop for), but it was great for pure compute.
By comparison the 360 used a trio of a similar general-purpose core without the SIMD processors on the side[1] had significantly less pure compute power.
[1] but with better built-in SIMD: the PS3's general-purpose core had a single internal SIMD unit while each of the 360's had 2, the 360's SIMD units were also improved with games-relevant operations and 4x the number of SIMD registers
6 usable for games but 8 physically in the silicon, with one disabled and one reserved for the OS. So a dual thread PPU and 6 game usable SPEs for the advertised "9 thread" (sometimes just referred to as 9-core) CPU.
Indeed, I initially put in the disabled one (for yield) and one reserved (especially as IBM's server version had all 8 enabled) but I figured that was of low enough interest I could just list the actually usable ones.
The problem was similar to Intel’s flop with IA-64: you could get a few amazing benchmarks but code which couldn’t be perfectly scheduled like that performed poorly, and that tuning was hard even if it could be done. The scientists I supported at the time were interested but the numbers made no sense: some code was inherently branch-heavy and couldn’t be tuned much and for the portions which could it was cheaper to buy more regular hardware using the money which would have gone to a high-end programmer, especially when you looked at the long-term maintenance cost of that heavily optimized code (everyone of that era remembered ripping out previous assembly optimizations which had become worse than C on newer CPU & compilers).
It's literally there: the PS4 is amd based pc (bulldozer architecture iirc) packaged in a fancy case and with a nice gpu.
Each system with one or two cat5e cables and a single power cable. In really low cost colocation facilities the limiting factor is often total kW of electricity and cooling available, not square footage of floor space.
It's been a common thing in the datacenter/dedicated server business for about 18-20 years.
The same general idea was adopted by Bitcoin miners using ASIC hardware, and later on with Ethereum mining again with systems using two or four GPUs connected to a low-cost ATX motherboard.
A mobo that supports 8 cards might be double the cost of a board that supports 4 cards, but it prevents you from having to buy a second cpu, a second hard drive, second psu, and more memory to put in your second cheapo mobo.
Now if you can buy random cheap ATX boxes for less than $27 dollars per PCI-E slot you are better off initially, but then management across all your various junk boxes becomes a PITA quick.
I've seen the various motherboards from Taiwanese board manufacturers which are designed for mining, with many small pci-express 3.0 slots intended for use with riser ribbon/interface cards, such as for use as you describe with 8 GPUs attached to one motherboard.
I bought surplus Core2Duo systems for $10 and slapped in a cheap SSD and a 1060 6GB. The CPU would sit there at 99% idle and the power supply never broke a sweat.
ROI is more or less impossible now. By the time you'd get close the reward will halve again. Also back then the diff kept rising increasingly quickly as well. I was doing this back in like 2016 with free power and cooling and made a grip of money.
The PS3 was using a pretty weird CPU, which importantly had a single general-purpose core but 6 SIMD co-processors, making it very useful for compute tasks on supercomputers.
These days you'd use off-the-shelf compute-oriented FPGA or stream / GPGPU units paired with general-purpose CPUs e.g. OLCF-4 has 9216 general-purpose CPUs driving 27648 Tesla V100.
One thing these new consoles are doing that seems interesting is that they use a huge pool of unified RAM (rather than two independent pools of main+video memory).
Something like a $190 Ryzen 3600 on a $100 motherboard. Add RAM and a small low cost SSD, or just do network boot/PXE. Total cost might be somewhat more than buying a PS4 Pro but it would run circles around it in any cpu intensive application.
See their 33c3 presentation on how they did it. It's filled with great details on how PS4 differs from PCs: https://youtu.be/-AoHGJ1g9aM
No hardware manufacturer does this any more.