Message boards : Number crunching : Good for Rosetta@home?
Author | Message |
---|---|
Tiago Send message Joined: 11 Jul 06 Posts: 55 Credit: 2,538,721 RAC: 0 |
NEC Launches World's Fastest Vector Supercomputer, SX-9 World's first vector performance exceeding 100 GFLOPS per single CPU core http://www.nec.co.jp/press/en/0710/2501.html What do you think of this "little monster"? |
The_Bad_Penguin Send message Joined: 5 Jun 06 Posts: 2751 Credit: 4,271,025 RAC: 0 |
Yeah, I saw a similar article: Just one core is capable of over 102 gigaflops (billion floating-point math operations per second); a single node of 16 CPUs generates up to 1.6 teraflops (trillion operations) while a full cluster of nodes produces 839 teraflops, NEC claims. The SZ-9 is also able to handle especially large data sets with a shared memory of 1TB as well as 128GB per second transfers all processors in the system. |
M.L. Send message Joined: 21 Nov 06 Posts: 182 Credit: 180,462 RAC: 0 |
Checks loose change to see if there is enough! OR...thinks Rosie should issue all crunchers with some of these. |
jaxom1 Send message Joined: 5 Jun 06 Posts: 180 Credit: 1,586,889 RAC: 0 |
of course, the app would have to be written to support it. Could you get it to work on my BlackBerry while you are at it? It's always on not doing anything..... except.. for.. sending me unwanted e-mails. :-) |
Luuklag Send message Joined: 13 Sep 07 Posts: 262 Credit: 4,171 RAC: 0 |
of course, the app would have to be written to support it. well if any of you guys has the money for it, to rent it 1 month will cost you 1,8 million dollar i thought, or euro, so both ways a lot of money ;) |
The_Bad_Penguin Send message Joined: 5 Jun 06 Posts: 2751 Credit: 4,271,025 RAC: 0 |
let's see, F@H calculates 20 gflops per PS/3 @ $399 (40gb hdd version). So, 5 PS/3's ($2k) should equal one core of the sx-9 (102 gflops). $2k * 16 cpu's = $32k = a single node of the sx-9 (1.6 tflops). $32k * ~500 (approx # of nodes in a full sx-9 cluster) = $16m (~800 tflops). Wonder if Sony would give us a price break on an order of 40,000 PS/3's... <checking for loose change under the sofa cushions> well if any of you guys has the money for it, to rent it 1 month will cost you 1,8 million dollar i thought, or euro, so both ways a lot of money ;) Rosetta currently has 160,000+ users, if only 25% of them own (were to purchase) a PS/3, we'd be near the petaflop barrier. Although, given that Micro$oft has "donated" $$$ to Rosetta, I wonder if there's some quid pro quo that Rosetta not code for Sony's PS/3, a M$ competitor in the gaming console wars... of course, the app would have to be written to support it. |
jaxom1 Send message Joined: 5 Jun 06 Posts: 180 Credit: 1,586,889 RAC: 0 |
let's see, F@H calculates 20 gflops per PS/3 @ $399 (40gb hdd version). Microsoft did not donate. The Gates Foundation Did. Get your Tax Right Offs Straight Mister!... :-) Seriously though, They are two distinct pools of money. Bill's Left or Right pants pockets.. ;-) |
Michael G.R. Send message Joined: 11 Nov 05 Posts: 264 Credit: 11,247,510 RAC: 0 |
Isn't the whole point of distributed computing NOT to have to pay for supercomputers? Not that it would hurt to have access to one :) In the meantime, cheap quad-cores and RAM should help the project gain teraflops. |
The_Bad_Penguin Send message Joined: 5 Jun 06 Posts: 2751 Credit: 4,271,025 RAC: 0 |
Bill's left pocket, right pocket, under his sofa cushions, all the same, lol ! Anywhose, just visiting F@H, seems they've done better with PS/3, getting ~35 gflops per PS/3. To reach NEC SX-9 performance, it'd take "only" 24,000 (not 40,000) PS/3's @ $400, or $9.6m (not $16m). Guess we just "lost" that Sony discount with the decreased size of the order... Microsoft did not donate. The Gates Foundation Did. Get your Tax Right Offs Straight Mister!... |
itayperl Send message Joined: 20 Mar 06 Posts: 5 Credit: 168,038 RAC: 0 |
Bill's left pocket, right pocket, under his sofa cushions, all the same, lol ! Don't forget power costs ;) PS3 takes about 200W while folding, so 24,000 * 200W = 4800 kW, for one year of non stop crunching @ $0.12 / kWh: 4800 * 24 * 365 * 0.12 = $5,045,760 |
dcdc Send message Joined: 3 Nov 05 Posts: 1832 Credit: 119,821,902 RAC: 15,180 |
Bill's left pocket, right pocket, under his sofa cushions, all the same, lol ! surely you've multiplied by 24 twice there? Should be $210 using your figures. |
Luuklag Send message Joined: 13 Sep 07 Posts: 262 Credit: 4,171 RAC: 0 |
Bill's left pocket, right pocket, under his sofa cushions, all the same, lol ! his math is correct. 24000 ps3's use 200W each an hour, so 4800 kW. times 24 times 365 is the hours in a year. times the 4800 times the cost of 1 kW/h makes that figure correct |
dcdc Send message Joined: 3 Nov 05 Posts: 1832 Credit: 119,821,902 RAC: 15,180 |
ah - thought you meant per PS3! i read it as $5k... -00- |
Natronomonas Send message Joined: 11 Jan 06 Posts: 38 Credit: 536,978 RAC: 0 |
But surely the NEC power costs are also substantial - you'd have to do a 3,5,8 year cost comparison or something to work out which worked out best overall. (Obviously, the PS3, you can play super-tournament games on it on the weekend! : P ) Crunching Rosetta as a member of the Whirlpool BOINC Teams |
The_Bad_Penguin Send message Joined: 5 Jun 06 Posts: 2751 Credit: 4,271,025 RAC: 0 |
OR..... if Rosie isn't under M$ influence to not work with Sony... you COULD get the donors of 24,000 PS3's to themselves pay for the consoles and energy expenditures... but that WOULD require Rosie to make the effort to reach out to Sony... (seems to have worked for F@H) and so far that hasn't been done by Rosie because ?????? (seems to be worth Rosies time and effort - our quick calculations show that the value of the consoles and 1 year of energy is worth $15m to her) and... the PS3 is standardized, no Mac vs IBM, no PowerPC vs Intel, no OS-X vs various flavors of Windows (XP / Vista / 32 bit / 64 bit) vs various flavors of linux, varying amounts of RAM, etc. AND.... you can play super-tournament games on it on the weekend! But surely the NEC power costs are also substantial - you'd have to do a 3,5,8 year cost comparison or something to work out which worked out best overall. |
The_Bad_Penguin Send message Joined: 5 Jun 06 Posts: 2751 Credit: 4,271,025 RAC: 0 |
Since we're fantasizing (24,000 PS/3's for $9.6m)... PS/3 was designed as an open platform PS/3 price has decreased from $599 to $399 PS/3 performance (at F@H) has increased from 20 Gflops to 35 Gflops (and may yet increase again) and now... PS/3 power usage decreases from 200 watts to 135 watts !!! From the Inquirer 40GB PS3 to include 65nm Cell processor A German tech site (via Gamesindustry.biz) is reporting that the new 40GB PS3 console will include a 65nm version of the Cell processor, thus reducing the system's power usage. The new system should therefore output less heat and noise, according to the tech site's report. The site also claims that the new console features a smaller heat pipe and a new motherboard. Computer Base quotes 135W at full load, down from 200W. The site says the console's noise levels are down from 1.3 to 0.5-0.8 sones. Engadget has shots of the 40GB packaging here. |
The_Bad_Penguin Send message Joined: 5 Jun 06 Posts: 2751 Credit: 4,271,025 RAC: 0 |
Also from the Inquirer Gaurav Khanna, an astrophysicist working at the University of Massachusetts, has been building his own supercomputer using Playstation 3 consoles. The professor has been renting time on supercomputers at NASA and the National Science Foundation to run highly complicated calculations on the amount of radiation emitted when a black hole swallows a star, which doesn't come cheap. In an average year, he rents about 30,000 hours, which costs between $20,000 and $30,000, a significant chunk of his grant money. To ease his supercomputing plight, Khanna turned to the cell chip inside the PS3. By linking eight of them together, he said he gets the same processing power as a supercomputer with 200 processors. More at Computerworld: For $4,000 or so, I can get eight PS3s that can do the same task that I'd do on a supercomputer," he said. "For a one-time cost, I have this resource I can use privately. I can use it indefinitely over and over again. That's hugely attractive. That's why I considered the project. I have my own supercomputer right here. There's no elaborate process for getting time. There's no waiting. It's just mine." "It has a unique design. It has a lot more potential -- a better processor in my opinion," he added. "What's unique is that they made it an open platform. Normally with a game console, the maker controls who can run what on it. What Sony did was make the PS3 an open platform. They let you run whatever you want on it. It has the full capabilities of a normal computer. You can run Firefox or whatever you want. It gave me the possibility of doing whatever I want with it." It sounds like a real wild thing and people will wonder why the hell it works," said Olds. "It works because that particular video game processor is a supercomputer processor. It will scream on the right kinds of workloads. The things that make it really good for video game processing make it good on certain processing-intensive workloads. This would be good for doing highly parallel processor-intensive analysis, taking on a big bunch of numbers, big calculations. Olds pointed out that IBM already picked up on this capability and integrated the cell chip into its QS20 BladeCenter product. And from wikipedia: IBM's new planned supercomputer, IBM Roadrunner, will be a hybrid of General Purpose CISC as well as Cell processors. It is reported that this combination will produce the first computer to run at petaflop speeds. It will use an updated version of the Cell processor, manufactured using 65 nm technology and enhanced SPUs that can handle double precision calculations in the 128 bit registers, reaching double precision 100 GFLOPs So tell me again, why doesn't Rosie think the PS/3 is worth the effort ??? |
Natronomonas Send message Joined: 11 Jan 06 Posts: 38 Credit: 536,978 RAC: 0 |
that can handle double precision calculations in the 128 bit registers, reaching double precision 100 GFLOPs[/u][/b] It may be those double-precision FP units. I know at SETI that was considered an issue, it could be the same here. (The regular Cell only have single-precision). Now that F@H has got 'the market', as it were, it might be hard for R@H to capture the gains f@h did... perhaps better to aim for xbox or something, especially when to the casual observer the two projects seem quite similar. Crunching Rosetta as a member of the Whirlpool BOINC Teams |
dcdc Send message Joined: 3 Nov 05 Posts: 1832 Credit: 119,821,902 RAC: 15,180 |
i think the problem is that F@H got in first with the PS3, so they've got their flagship project now which shows off the PS3's power. However, I believe Rosetta changes more rapidly than the folding software, and the cell each have a limited amount of memory which might restrict rosetta. MS were talking with the bakerlab about the 360 but there's no way MS are gonna release something that maxes the CPU when there's already so many overheating problems! Also, it would look to the uninformed observer that the xbox isn't anywhere near as powerful as the PS3 when compared to the number of PCs it replaces. Of course Cell is responsible for graphics duties in the PS3 so the xenon would need to be maxed to make a fair (but pointless!) comparison, and I imagine coding for xenon would be as difficult as for any GPU... |
Jmarks Send message Joined: 16 Jul 07 Posts: 132 Credit: 98,025 RAC: 0 |
i think the problem is that F@H got in first with the PS3, so they've got their flagship project now which shows off the PS3's power. However, I believe Rosetta changes more rapidly than the folding software, and the cell each have a limited amount of memory which might restrict rosetta. Folding may be the first in the USA not the world. There is a Spanish project that had it long before Folding. If you go to the Sony PS3 page they have whites papers and video links to them and even Ibm hacking instructions. Jmarks |
Message boards :
Number crunching :
Good for Rosetta@home?
©2024 University of Washington
https://www.bakerlab.org