The rise and fall of the PlayStation supercomputers • Hacker News Board

Re: The rise and fall of the PlayStation supercomputers

#51

Post by zyztem » Sat, Dec 07, 2019, 8:58 PM UTC

ascagnel_ wrote: post #13

ChuckNorris89 wrote: post #11

Earlier quoted context omitted.

To play the devil's advocate, Sony is in the business of selling games, their cosnoles were just a means to that and since they were selling the hardware at a loss, selling more of it to people who will never buy games was bad for their business model.

There was no reason they couldn’t have sold a slightly less locked-down system to researchers at a higher price point —- Microsoft did exactly that with unsold X360 Kinect units (and bundled in a USB adapter) for medical research.

Cell processor shipped for commercial installation in the IBM Blades QS20/QS21/QS22, which was based on Roadrunner supercomputer.

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

Re: The rise and fall of the PlayStation supercomputers

#52

Post by acdha » Sat, Dec 07, 2019, 8:59 PM UTC

clay_the_ripper wrote: post #3

> the PS4 can’t easily be turned into a cog for a supercomputing machine. “There’s nothing novel about the PlayStation 4, it’s just a regular old PC,” Khanna says. “We weren’t really motivated to do anything with the PlayStation 4.” 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?

Semiconductor pricing is largely determined by volume. The Cell processor had mediocre normal performance inline with other Power derivatives but really good vector performance for the target price (this was before GPU computing matured), so people with vector-amenable tasks could theoretically win by adopting a processor shipping by the million rather than using lower-volume chips targeted at traditional HPC uses. Since Cell was a Sony/IBM project, IBM was interested in reselling the same work to new customers and (IIRC) needed the volume to have their chip foundry business be healthy.

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).

Re: The rise and fall of the PlayStation supercomputers

#53

Post by aorth » Sat, Dec 07, 2019, 8:59 PM UTC

Seriously surprised that the article didn't mention Iraq buying 4,000 PlayStation 2's in the year 2000. I think Saddam was doing some supercomputing.

https://www.theregister.co.uk/2000/12/19/iraq_buys_4000_play...

Re: The rise and fall of the PlayStation supercomputers

#54

Post by masklinn » Sat, Dec 07, 2019, 9:02 PM UTC

krilly wrote: post #5

I would expect this to be more viable than ever, now that the cost of console hardware is so heavily subsidised by online subscription services.

The CPUs of modern console are pretty standard general-purpose CPUs: the PS4 and XB1 use customised AMD APU, and the Switch uses a Tegra X1. That's not useful for supercomputers.

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.

Re: The rise and fall of the PlayStation supercomputers

#55

Post by lozaning » Sat, Dec 07, 2019, 9:02 PM UTC

ianai wrote: post #38

kube-system wrote: post #31

Earlier quoted context omitted.

They did. The article linked above mentions that it took them a while to convince Sony to agree to it. 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.

Definitely a completely different support contract than a consumer console. That sounds more like an amortization of how many consoles they’d need to support that large of a system for the life of the contract.

Could also be the rake you pay to the company who will sell you hardware on terms with a PO.

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.

Re: The rise and fall of the PlayStation supercomputers

#56

Post by lozaning » Sat, Dec 07, 2019, 9:10 PM UTC

walrus01 wrote: post #33

That wire rack shelf (sans wheels) is still how some low budget colocation/dedicated server operators set up their hardware. Minitower or mid-tower microatx/ATX sized cases with low cost motherboards, CPU, power supplies. A cheap microatx sized tower case with one rear exhaust fan is $35. Each system with one or two cat5e cables and a single power cable. In really low cost colocation facilities the limiting factor is…

If you're not running 7+ cards on a mining rig you're probably doing it wrong. The higher the number of cards you run in a rig the more you can distribute the cost of the non mining hardware (HD, mobo, cpu, memory) across cards.

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.

Re: The rise and fall of the PlayStation supercomputers

#57

Post by LaserToy » Sat, Dec 07, 2019, 9:15 PM UTC

mschuster91 wrote: post #17

ChuckNorris89 wrote: post #14

Earlier quoted context omitted.

The reason was cost. Have fun convincing management to allocate resources to develop and support another SKU tailored for researchers.

So what? Don't need much difference in hardware, it was only a firmware change.

It is time and not a priority.

A lot of things can be done given unlimited time and resources, but unfortunately we don’t have that.

Re: The rise and fall of the PlayStation supercomputers

#58

Post by walrus01 » Sat, Dec 07, 2019, 9:16 PM UTC

lozaning wrote: post #56

walrus01 wrote: post #33

That wire rack shelf (sans wheels) is still how some low budget colocation/dedicated server operators set up their hardware. Minitower or mid-tower microatx/ATX sized cases with low cost motherboards, CPU, power supplies. A cheap microatx sized tower case with one rear exhaust fan is $35. Each system with one or two cat5e cables and a single power cable. In really low cost colocation facilities the limiting factor is…

If you're not running 7+ cards on a mining rig you're probably doing it wrong. The higher the number of cards you run in a rig the more you can distribute the cost of the non mining hardware (HD, mobo, cpu, memory) across cards. 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 t…

The last time I spent 15 minutes doing a possible ROI calculation on a GPU based Ethereum mining system, about a year ago, I came to the conclusion that the upfront cost of buying the GPUs alone, plus the electricity (entirely ignoring the cost of a dedicated mining purpose motherboard and cpu/ram, power supply, etc) would not have a reasonable payback period. Even if the electricity was 1.5 cents a kWh, and cooling was free.

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.

https://www.techradar.com/news/best-mining-motherboards

Re: The rise and fall of the PlayStation supercomputers

#59

Post by forrestthewoods » Sat, Dec 07, 2019, 9:16 PM UTC

DCKing wrote: post #49

The Playstation 1, 2 and 3 are each in their own way Sony's take on an alternate history of computer graphics. It was Sony's intent with all three consoles to put the "special effects" (that in 2019 are performed using shaders) in the CPU , and not the GPU. 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 p…

> 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.

Re: The rise and fall of the PlayStation supercomputers

#60

Post by vvanders » Sat, Dec 07, 2019, 9:19 PM UTC

zanny wrote: post #43

ChuckNorris89 wrote: post #11

Earlier quoted context omitted.

To play the devil's advocate, Sony is in the business of selling games, their cosnoles were just a means to that and since they were selling the hardware at a loss, selling more of it to people who will never buy games was bad for their business model.

To play devils advocate, you can sell games without building perfectly viable general purpose compute hardware and then locking it down with DRM to make it useless as such. Microsoft, Sony, and Nintendo are not developing game consoles out of anything but an incredibly evil greed - they are wasting insurmountable amounts of raw materials making computers that cannot be used as computers. The consoles that existed in…

If consoles were sold at a profit what you say may make sense. However building an open platform means you need to devote resources to it both from a capital perspective and opportunity cost.

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.