First Cell demo (48 MPEG 2 Videos)

[randycat99]

In order to do that you have to assume a lot of things...

[GASP!]

Do you or do you not stand behind your numbers??? Is 20% telling you something useful or absolute garbage? Pick one- run with it. What say you?

 

[JF_Aidan_Pryde]

So you need (48*640*480*2 + 1920*1080)*4*25 = 3 Gb/s

And that's just for pointsampling downscale - add a simple bilinear, or something fancier like cubic and required bandwith goes way higher. Actually we COULD test exact speed requirements for just filtered resizing on a P4, and it'd probably be more meaningfull benchmark then anything else we can do since it could actually be measured clock precise and without overhead of multiple abstraction layers.

I don't think I follow the numbers.
Why 640 x 480 x 2? What's the x2?

To me, this decoding problem (without resizing) is like this:

48 MPEG2 input -> CPU -> 48 SDTV streams output

The above in terms of bandwidth:

48MB/s -> CPU -> 1.47GB/s

Where 1.47GB/s = 640 x 480 x (32/4) x 25 x 48

With resizing, don't you just have to read in the 48 streams again (1.47GB/s) and output at 1920x1024? With fancier filtering, the number of operations in the CPU will need to go up but bandwidth should still be no more than 1.47GB/s input, no?

 

[PC-Engine]

In order to do that you have to assume a lot of things...

[GASP!]

Do you or do you not stand behind your numbers??? Is 20% telling you something useful or absolute garbage? Pick one- run with it. What say you?

My numbers??? My numbers were to establish a basis to conclude CELL is not orders of magnitude more powerful. It's not for the purpose of concluding it's only 20% more powerful. Cell could be 2X, 4X, or even 6X more powerful than a dual core HT P4 at the same clock speed when running these kinds of streaming apps that favor it more than a P4, but even then it's still not a single order of magnitude faster. That was my point.

 

[randycat99]

...and there you have it folks- after all this, not even pce stands behind the ramifications of his own "numbers-based" assertion. This whole thing was a wild-goose chase (but we knew this, of course- we were just a bit bored )

Not quite getting it, yet, pce? If you cannot even stand behind "20%" to mean anything significant (perhaps, even granting being within 100% of the "real" value), then you can't very well submit it as anything indicative or meaningful wrt "orders of magnitude". It's all bunk, if you can't even agree that "20%" has any basis. Have a nice day!

 

[PC-Engine]

Uh read it again. I've edited for clarification. BTW the 2X ,4X, and 6X were just arbitrary examples. For all intense and purposes it could very well even be 1.5X.

 

[randycat99]

So pick one, and make a statement. If you cannot even stand by that figure, then the "orders of magnitude" assertion is equally ambiguous. Hence, your argument crumbles.

 

[nAo]

Why 640 x 480 x 2? What's the x2?

Decoded streams must be written (+1) and read (+1), that's why we have a 2x factor.

With resizing, don't you just have to read in the 48 streams again (1.47GB/s) and output at 1920x1024? With fancier filtering, the number of operations in the CPU will need to go up but bandwidth should still be no more than 1.47GB/s input, no?

Unfurtunately this is not the case.
Every filter with a domain that is larger than a single sample will require multiple samples per pixel.
This means sample values are going to be reused multiple times and since we don't have a cache that can hold a full decoded stream no matter what we do, some sample is going to be fetched multiple times from memory.
Obviously there are ways to improve this situation, like to store a decoded stream in some hierarchical tiled fashion, and filtering a tile at time, following a special tiling order to maximize sample reuse between tiles.

 

[PC-Engine]

So pick one, and make a statement. If you cannot even stand by that figure, then the "orders of magnitude" assertion is equally ambiguous. Hence, your argument crumbles.

If I had to guess, CELL is 2-3 times more powerful in this type of app with the assumption that it's running at less than 3.5GHz and it's also doing downsampling.

 

[randycat99]

There it is folks- get your sigs ready...pce goes on record that the Cell is only 3x more powerful than an HT, DC 3.5 GHz P4. Whereas before he cited that an answer could not be made, then faced with the premise that "orders of magnitude" becomes irrelevant w/o an answer, an answer then magically becomes possible! I guess MS opened itself up to some severe programming growing pains in a major codebase migration to a new ISA, just to get away from an x86 solution that will only be a bit behind what they are going with now for XB2. Does this get any better?

 

[Simon F]

Decided to delete this message. Mods considered that it was only encouraging the bickering.

 

[PC-Engine]

There it is folks- get your sigs ready...pce goes on record that the Cell is only 3x more powerful than an HT, DC 3.5 GHz

Don't quote out of context. He said "if I have to guess".

randycat99 has a history of this, he also forgot to quote the assumptions that were made.

It's pefectly clear now why things get twisted from one meaning to the next for the sole purpose of trying to prove another person's argument doesn't hold water and hence crumbling.

Someone just got smacked while jumping the shark...

It's just bridge under the water AFAIC.

 

[randycat99]

Do you not stand behind this "guess"? If not, then why bother offering it? Perhaps "20%" that you based on "hard numbers" felt pretty unlikely, as well? So essentially we have achieved zero progress with your assertions.

Stand behind something, or just sdastfu, already. Geez!

 

[PC-Engine]

Do you not stand behind this "guess"? If not, then why bother offering it? Perhaps "20%" that you based on "hard numbers" felt pretty unlikely, as well? So essentially we have achieved zero progress with your assertions.

No you have achieved zero progress...

Just because one doesn't get the point, doesn't mean a point doesn't exist...

BTW where did you get this magical 20% from?

Without context, anyone can claim anything and pretty much get away with it.

 

[randycat99]

Yeah, I got your point, and it turned out to be bunk. Deal.

I guess we can take your response to the effect that you do not stand behind your "guess", you do not stand behind your "numbers", you just expect others to take your word as if it was handed to you from God. Yes, I sure jumped the shark, alright!

 

[PC-Engine]

Yeah, I got your point, and it turned out to be bunk. Deal.

If it makes you feel better then go ahead and believe that just like believing the Graphics Sythesizer had magical untapped FLOPS to bring it to Flipper levels.

Like I said...water under the bridge...

 

[randycat99]

So a "competent programmer" tells you Cell is only 20% ((48-40)/40) more powerful than a HT, DC 3.5 Ghz P4, or did you take some basic measurements and work them up into an out-of-context solution?

 

[PC-Engine]

So a "competent programmer" tells you Cell is only 20% ((48-40)/40) more powerful than a HT, DC 3.5 Ghz P4, or did you take some basic measurements and work them up into an out-of-context solution?

Give it up dude. Nobody said CELL is 20% more powerful from using a simple 40 vs 48 + downsampling example.

 

[randycat99]

Well aren't we all glad you are coming to your senses, finally!

 

[PC-Engine]

Well aren't we all glad you are coming to your senses, finally!

Watch out for that shark!

Key phrase is orders of magnitude...

 

[pahcman]

wow thread turn crazy. calm down guys. i have to quote this again

My claim is that a single threaded P4 at 3.5GHz can do at least 12 DVD streams minus the downscaling. Then using that as a base, a hyperthreading P4 at 3.5GHz would be able to do 20 streams. Finally based on those two factors, a dual core hyperthreading P4 at 3.5GHz would probably be able to do ~ 40 DVD streams minus the downsampling of course.

I'm not claiming a dual core HT P4 is more powerful than that CELL that was used in that demo. This is plainly obvious because 1. We don't know what clock speed CELL was running at and 2. Even if we knew the clock speed we still don't know the % utilization.

My point was to show that CELL is not orders of magnitude better than a dual core HT P4 which ERP and aaaaa00 also agree. This can be extrapolated with the known information.

why are some getting upset over something i find very simple pce said? he never say cell is about as good as p4, just not orders magnitude bigger. which my english say that is really really really big..

i know many feel disgusted with x86 over a long time, but please do not let the so call " new broadband architutre " or " 1000x P4 " talk , make you think all chips are hapless to cell revolution.