First PS3 "Supercomputer" ...

[Jesus2006]

There is some 80 GFLOP number thrown around but the translation broke down here, perhaps refering to this (http://sc06.supercomputing.org/schedule/pdf/pap225.pdf).

This is some "realworld experienced" value ("...in der Praxis", in practice), but there is no source where they might have gotten that from. Possibly avg you get in various applications.

Maybe i should completely translate such articles next time

 

[Capeta]

Since only 6 SPEs are active in PS3 Linux, they estimated linearly that they can get 150 GFLOPs.

With 8 PS3s, assuming 100% scaling, they claimed 8 * 150 GFLOP = 1.2 TFLOP.

And that is EXACTLY what I'm talking about. It is Single Precision so why are they comparing 1.2TFLOPS of SP to 2.7TFLOPS of DP? Doesn't compute. Why not use DP figures instead if you're going to be comparing it DP figures for a Top500 placing? The article is sensationlizing, there's really no other way to describe it. To get into the Top500, you would need at least 90 PS3s.

 

[patsu]

And that is EXACTLY what I'm talking about. It is Single Precision so why are they comparing 1.2TFLOPS of SP to 2.7TFLOPS of DP? Doesn't compute. Why not use DP figures instead if you're going to be comparing it DP figures for a Top500 placing? The article is sensationlizing, there's really no other way to describe it.

Your arguments are all over the place (e.g., claiming mysterious 2.4TFLOP number when nothing like that appears in both articles).

The most active area of work for the current Cell implementation is in SP. Most "real world" numbers quoted for Cell are therefore in SP. What's so surprising about it ?

In fact, the article highlighted the single precision focus as a weakness.

Similarly heavily another weakness weighs: The processor used in the console, the Cell Broadband engine (Cell) with a 64-bittigen power PC core and eight so-called Synergistic processing of element (SPE), is correctly fast only with floating decimal point computations in simple accuracy. Therefore it is suitable only conditionally for scientific computations, which take place as a rule with double accuracy. Eight SPEs can implement than 200 billion floating decimal point operations (GigaFlop/s) per second with the current clock of 3,2 GHz theoretically together more. In practice one comes on approximately 80 GigaFlop/s. into the PS3 only seven SPEs is used, under Linux is only to six at the disposal.

They went on to highlight again...

To the introduction into the Top500-Liste of the supercomputers is not enough still for a long time. At place 500 at present a Xeon cluster with a maximum floating decimal point achievement (Rmax) of approximately 2.7 TeraFlop/s ranks - and in double accuracy.

You're just grasping at straws.

 

[Capeta]

To get into the Top500 you need 90 PS3s. There's a big difference bewteen 90 and 8. Yes I'm grasping at straws.

1.2TFLOPS..meaningless.

Why didn't they quote DP numbers for the PS3 cluster??? Maybe because the WOW factor would drop an order of magnitude?

 

[patsu]

So ?

With Folding@Home, we will soon have a worldwide network of PS3 "grid".

The next Cell will go into RoadRunner, overshadowing the current champion, BlueGene.

Cell is already in the supercomputing/scientific-computing business today. The article merely highlighted some isolated effort for creating a small PS3-cluster under 5K, without extensive cooling and power generators.

EDIT:

To get into the Top500 you need 90 PS3s. There's a big difference bewteen 90 and 8. Yes I'm grasping at straws.

1.2TFLOPS..meaningless.

If the SP performance is meaningless, why did the military, universities and national labs buy Cell hardware ?

Why didn't they quote DP numbers for the PS3 cluster??? Maybe because the WOW factor would drop an order of magnitude?

Because the PS3 cluster is for doing mainly SP problems and SP real world numbers are readily available ?


BTW, you have not explained the 2.4TFLOP number you magically dreamt up. How is it relevant to the article ? You were attacking it based on that number too.

 

[Rainbow Man]

To get into the Top500 you need 90 PS3s. There's a big difference bewteen 90 and 8. Yes I'm grasping at straws.

Nobody's claiming 8 PS3s creates a top500 super computer. You're trying to kick in open doors. It just looks silly.

Peace.

 

[Capeta]

Nobody's claiming 8 PS3s creates a top500 super computer.

I didn't claim the article did.

So ?

With Folding@Home, we will soon have a worldwide network of PS3 "grid".

The next Cell will go into RoadRunner, overshadowing the current champion, BlueGene.

Cell is already in the supercomputing/scientific-computing business today. The article merely highlighted some isolated effort for creating a small PS3-cluster under 5K, without extensive cooling and power generators.

If the SP performance is meaningless, why did the military, universities and national labs buy Cell hardware ?

Because the PS3 cluster is for doing mainly SP problems and SP real world numbers are readily available ?

BTW, you have not explained the 2.4TFLOP number you magically dreamt up. How is it relevant to the article ? You were attacking it based on that number too.

You're now entering strawman territory and going off on a tangent. I've already showed why that article is a sensationlist piece, nothing more. Again answer the question if you want to continue to beat a dead horse, why did they quote SP figures when comparing to a Top500 supercomputer instead of quoting DP figures? It is utterly pointless to compare SP figures from a PS3 "supercomputer" cluster to DP figures from a Top500 supercomputer.

 

[Jesus2006]

I didn't claim the article did.

No, the article did not - read again.

All they said was that it's still a long way to get into the Top500, with a DP System at 2,7TFlops DP accuracy at place 500!

They are not comparing SP with DP figures at all.

There really is nothing in there to be upset about. This is just to show that you can get a system with quite a bit of computing power for a very low price. All flaws were explicitly mentioned in this very articel.

 

[patsu]

I didn't claim the article did.

Which statement in the article says you can create a top 500 supercomputer using 8 PS3s ?

You're now entering strawman territory and going off on a tangent. I've already showed why that article is a sensationlist piece, nothing more. Again answer the question if you want to continue to beat a dead horse, why did they quote SP figures when comparing to a Top500 supercomputer instead of quoting DP figures?

Because SP performance is Cell's strength ? And people buy Cell today primarily for its SP performance ? The article is always clear about the DP and SP differences. It also commented that it will take a long time to make it into top 500.

I don't think the article is trying to be sensational.

The 500th supercomputer consists of 800 processors. You optimistically mentioned that 90 PS3s can break in (probably assuming 100% efficiency). You sound more hopeful than the article.

 

[Capeta]

Which statement in the article says you can create a top 500 supercomputer using 8 PS3s ?

"I didn't claim the article did" means I didn't claim the article made the claim. It doesn't mean the article made the claim.

Because SP performance is Cell's strength ? And people buy Cell today primarily for its SP performance ? The article is always clear about the DP and SP differences. It also commented that it will take a long time to make it into top 500.

I don't think the article is trying to be sensational.

I thought it was sensational, otherwise they would've used DP figures instead which are widely available. Using DP figures wouldn't have had the same WOW factor so it's not surprising they used 1.2TFLOPS SP figures to mislead readers into thinking it could nearly reach 2.7TFLOPS DP figures of a Top500 supercomputer.

The 500th supercomputer consists of 800 processors. You optimistically mentioned that 90 PS3s can break in (probably assuming 100% efficiency). You sound more hopeful than the article.

I'm not being optimistic/hopeful at all since the 500th supercomputer is probably using old off-the-shelf CPUs. I'm just using numbers given out by STI on DP for CELL. I try to avoid massaging numbers.

 

[patsu]

"I didn't claim they did" means I didn't claim the article made the claim. It doesn't mean the article made the claim.

>_<

I thought it was sensational, otherwise they would've used DP figures instead which are widely available. Using DP figures wouldn't have had the same WOW factor so it's not surprising they used 1.2TFLOPS SP figures to mislead readers into thinking it could nearly reach 2.7TFLOPS DP figures of a Top500 supercomputer.

You didn't read the article initially, didn't you ? If you have read it, you would have the context to interpret the numbers. Quoting the SP numbers does sound more impressive, but SP is also Cell's focal point today. It is natural for people to quote them. It does not necessarily mean that they are trying to pull a fast one.

I'm not being optimistic/hopeful at all since the 500th supercomputer is probably using old off-the-shelf CPUs. I'm just using numbers given out by STI on DP for CELL. I try to avoid massaging numbers.

Still, the estimate is overly optimistic (even if you didn't intend to). If it was someone else, you might have jumped on him and claimed that he's being sensational.

In any case, I'll leave it as it is. There is no point to take this further.

 

[Capeta]

You didn't read the article initially, didn't you ? If you have read it, you would have the context to interpret the numbers. Quoting the SP numbers does sound more impressive, but SP is also Cell's focal point today. It is natural for people to quote them. It does not necessarily mean that they are trying to pull a fast one.

And if they didn't try to sensationlize the article by using 1.2TFLOPS of SP and instead used 200 or 240GFLOPS of DP, it would've been more credible and less BS.

Still, the estimate is overly optimistic (even if you didn't intend to). If it was someone else, you might have jumped on him and claimed that he's being sensational.

Actually there's a difference between using 1.2TFLOPS of SP from 8 PS3s vs using an "optimistic" 240GFLOPS of DP from 8 PS3s, the former has a much higher WOW factor when compared to a 2.7TFLOPS supercomputer. There is no supercomputer in the Top500 that is rated in GFLOPS, they're all rated higher than 1TFLOPS.

 

[patsu]

And if they didn't try to sensationlize the article by using 1.2TFLOPS of SP and instead used 200 or 240GFLOPS of DP, it would've been more credible and less BS.

If you say so, but the article is not wrong in claiming 1.2TFLOP SP given their stated assumptions.

Actually there's a difference between using 1.2TFLOPS of SP from 8 PS3s vs using an "optimistic" 240GFLOPS of DP from 8 PS3s, the former has a much higher WOW factor when compared to a 2.7TFLOPS supercomputer. There is no supercomputer in the Top500 that is rated in GFLOPS, they're all rated higher than 1TFLOPS.

And yet you're wrong in claiming 90 PS3s can break into Top 500. The article is correct in saying that they are a long way off.

 

[Capeta]

If you say so, but the article is not wrong in claiming 1.2TFLOP SP.

Yet completely meaningless when trying to compare it to 2.7TFLOPS of DP.

And yet you're wrong in claiming 90 PS3s can break into Top 500. The article is correct here.

The article is correct when it tried to make it seem as though 1.2TFLOPS SP is relevent when comparing to a Top500 supercomputer that gets 2.7TFLOPS of DP? Ok....>_<

 

[CEO]

Yet completely meaningless when trying to compare it to 2.7TFLOPS of DP.



The article is correct when it tried to make it seem as though 1.2TFLOPS SP is relevent when comparing to a Top500 supercomputer that gets 2.7TFLOPS of DP? Ok....>_<

Capeta calm down. You should have seen what jesus did there by now. Its ovious that the article didn't sensationalyze all the supercomputer crap as much as jesus2006 did.