will Sony reach its old goal of 18,000x PS1, in PS3?

[Guden Oden]

Deadmeat,
You got it all backwards - as per usual.

BANDWIDTH increases speed.

MORE RAM can only AVOID SLOWING DOWN when paging. It doesn't ACTUALLY increase speed of anything!

You can check this yourself if you do computations of identical data-sets that fit into both amounts of main memory (matrix muls on datasets larger than on-chip caches for example - there are many benches of this type) or are computation-bound, not I/O-bound (such as video encoding).

So you're definitely wrong here. As per usual.

 

[I.S.T.]

Exactly. Deadmeat, you seriously need to go learn the basics of PC hardware again.

 

[Deadmeat]

...

Guden

BANDWIDTH increases speed.

The whole point is that Bandwidth doesn't matter much for PC System RAM type access.

MORE RAM can only AVOID SLOWING DOWN when paging.

Which matters far far more than PC environment.

You can check this yourself if you do computations of identical data-sets that fit into both amounts of main memory

That's not what Paul was asking. You don't understand the question, so you give the wrong answer.

IST

Exactly. Deadmeat, you seriously need to go learn the basics of PC hardware again.

You too don't understand Paul's question.

Paul's question

what performs better... a PC with 1024mb PC-133 SDRAM, or a PC with 512mb DDR.

The answer is PC133 for most applications. Why? Because of heavy paing. Try to understand the freaking question before trying to answer, OK?? I noticed almost none of you were reading into the original question.

READ PAUL'S QUESTION OVER PLEASE!!!!!!

what performs better... a PC with 1024mb PC-133 SDRAM, or a PC with 512mb DDR?

 

[I.S.T.]

The speed difference is much too great for most aplications to run faster. Unless it's somethign like Photoshop, which needs more RAM, the PC will run faster with the faster but less RAM. It will page more, sure. But the parts that ain't paging will run much faster.

 

[Guden Oden]

Re: ...

BANDWIDTH increases speed.

The whole point is that Bandwidth doesn't matter much for PC System RAM type access.

That is a point you made up. It was true once, when CPUs were very slow overall, but on a multi-GHz chip memory bandwidth has a definite impact on most any application that uses large quantities of memory and thus won't fit into the CPU's cache.

Also note, cache lines on a P4 is not 128 BITS. They're 128 BYTES, thus giving FSB bandwidth a good workout when replacing a line.

MORE RAM can only AVOID SLOWING DOWN when paging.

Which matters far far more than PC environment.

Another unfounded assumption you make. Or, do you have any actual data backing up your position, other than your own words?

You can check this yourself if you do computations of identical data-sets that fit into both amounts of main memory

That's not what Paul was asking.

REGARDLESS, it is true! Not every application is limited by memory size, far from it, and in many of those, bandwidth will play a large factor. This includes media encoding like already mentioned.

You don't understand the question, so you give the wrong answer.

Perhaps it is you who do not understand the question. You see, I happen to support Paul's position, that in most instances the system with the smaller yet faster RAM will be faster. Within reason of course. Facts are though, for most tasks people use their PCs for today, a system with 512MB fast memory will outclass a system with 1024MB of slow memory. Even games do not spend much time paging if you got 512MB, during all the time the game does NOT page, it runs consistently much faster than it would on the system with the larger, slow memory.

Seriously, how much time would YOU say a not completely underpowered system spends paging? Most of it is done when you start the game, and/or load the first level, not during actual gameplay unless you completely ignored the system recommendations on the packaging.

what performs better... a PC with 1024mb PC-133 SDRAM, or a PC with 512mb DDR.

The answer is PC133 for most applications.

Haha, no.

Why? Because of heavy paing.

In your dreamworld, perhaps, where "Cell" (sic) will be a chip with two PUs and 9 APUs (8 used, one in reserve) and 16MB eDRAM and simultaneously have no APUs but instead 4 Power4 CPUs and no eDRAM and simultaneously several other variations you've created in your head these past few months, all manufactured in a .65nm process using 113nm design rules.

So before you go and accuse others of not understanding... Here's a nickel. Go buy yourself a clue, mmkay?

Edit: quote error... Gnnh.

 

[Deadmeat]

...

The speed difference is much too great for most aplications to run faster. Unless it's somethign like Photoshop, which needs more RAM, the PC will run faster with the faster but less RAM. It will page more, sure. But the parts that ain't paging will run much faster.

PC mag disagrees... You see a drastic performance improvement just by doubling memory size, and this has been proven time after time.

That is a point you made up. It was true once, when CPUs were very slow overall, but on a multi-GHz chip memory bandwidth has a definite impact on most any application that uses large quantities of memory and thus won't fit into the CPU's cache.

The memory access latency situation has worsened in the Ghz era and the CPU themselves are designed to overcome this.(ie hyperthreading). Again, it is the memory access latency that governs the direction of modern CPU engineering, not bandwidth.

Also note, cache lines on a P4 is not 128 BITS. They're 128 BYTES thus giving FSB bandwidth a good workout when replacing a line.

From the 3.2 Ghz CPU's point of view, any L2 cache miss will cause the pipe stall measuring in 200+ cycles. The priority of CPU engineering is to reduce this massive latency by all means(Intel's talking about something called "helper" thread just to handle access pattern prediction and prefetch in their future direction). Doubling of bandwidth won't do a thing to solve this problem.

Another unfounded assumption you make. Or, do you have any actual data backing up your position, other than your own words?

Not every application is limited by memory size, far from it

Small applications do not stress the memory subsystem at all!!! It is the large applications that do stress the memory subsystem, and large application benefits far more from reduced paging afforded by larger system RAM than increased bandwidth. Why is it so hard to understand?????

and in many of those, bandwidth will play a large factor. This includes media encoding like already mentioned.

Paul's link already proves that there is a 5(most)~15(little)% performance between SDR and DDR. While this proves that doubling memory capacity increases performance by large margin...

Within reason of course. Facts are though, for most tasks people use their PCs for today, a system with 512MB fast memory will outclass a system with 1024MB of slow memory.

If you are running Win98, sure.
If you are running XP, hell no.

Even games do not spend much time paging if you got 512MB, during all the time the game does NOT page, it runs consistently much faster than it would on the system with the larger, slow memory.

jvd already discussed about how he saw a 15% frame rate jump from Star Wars game he was benchmarking just by doubling of RAM to 1 GB...

Seriously, how much time would YOU say a not completely underpowered system spends paging?

More than you would expect.

WinXP allocates well over 1 GB in virtual memory. Most systems have 256~512 MB installed. Where do you think the remainder comes from?? Paging, of course.

 

[Paul]

Paul's link already proves that there is a 5(most)~15(little)% performance between SDR and DDR. While this proves that doubling memory capacity increases performance by large margin..

BOTH GPU'S WERE USING DDR, ONE WITH 64MB AND THE OTHER 128, THE ONE WITH 64 OUTCLASED THE RADEON 8500 BY OVER 1000 POINTS IN 3DMARK.

jvd already discussed about how he saw a 15% frame rate jump from Star Wars game he was benchmarking just by doubling of RAM to 1 GB...

While games like UT2003 would run better with 512mb DDR instead of a GB of SD???

If you are running Win98, sure.
If you are running XP, hell no.

Oh so now it's *if your using XP????* What a joke.

WinXP allocates well over 1 GB in virtual memory. Most systems have 256~512 MB installed. Where do you think the remainder comes from?? Paging, of course.

*Looks at laptop page file* 672MB, I didn't tinker with these settings. Laptop is a Athlon 2400+ 512mb DDR.

So... WRONG.


Insanely funny how you built up your complex house of contradictions thatched together with bullshit only to have everyone stomp it down.

 

[I.S.T.]

Re: ...

The speed difference is much too great for most aplications to run faster. Unless it's somethign like Photoshop, which needs more RAM, the PC will run faster with the faster but less RAM. It will page more, sure. But the parts that ain't paging will run much faster.

PC mag disagrees... You see a drastic performance improvement just by doubling memory size, and this has been proven time after time.

It's a PC maganize. Those are notorious for errors for Christ's sake. While I have read that this one isn't quite as bad as the others, I still don't trust it. Show me a web link to other than their site and I'll believe ya. Edit: Messed up the quoting. Edit2: Fixed a typo.

 

[Megadrive1988]

please lets get back on-topic before this thread gets locked.


I think Sony is aiming for "CGI-like" visuals with PS3. I think Sony will go a long way towards getting their, but fall short in a few areas. PS3 will be incredibly powerful, but like PS1 and PS2, it will have some flaws.

 

[I.S.T.]

Just like any other system, lol. IMO they won't hit their target of the PS3 being 18,000X more powerful than the PS1. That's incredibly high. They missed the PS2's target, as shown in this thread. I think that 5000X is a more realistic target.

 

[Vince]

Just like any other system, lol. IMO they won't hit their target of the PS3 being 18,000X more powerful than the PS1. That's incredibly high. They missed the PS2's target, as shown in this thread. I think that 5000X is a more realistic target.

How is that high? It's an open-ended number, it's probably highly realistic when you actually look at the progression companies like ATI have spearheaded. Stuff like filtering are incredibly expensive as opposed to doing it on a CPU, Bilinear is what, 28ops/pixel? At 1200MPixel/sec that's 33,600Mops/sec alone.

As Gary Tarolli of 3dfx once stated, these numbers are more fun than meaningful.

 

[I.S.T.]

It's just a feeling. Maybe I was a bit too concervitive about how much power the PS3 will have, lol.

 

[Deadmeat]

...

Paul

BOTH GPU'S WERE USING DDR, ONE WITH 64MB AND THE OTHER 128, THE ONE WITH 64 OUTCLASED THE RADEON 8500 BY OVER 1000 POINTS IN 3DMARK.

What's your point? Are we talking about CPU or GPU here???

While games like UT2003 would run better with 512mb DDR instead of a GB of SD???

Got a benchmark to prove your claim??? I look forward to it.

Oh so now it's *if your using XP????* What a joke.

The reason Win98 doesn't see much gain is that it had a poor memory management. NT/XP memory management is better.

*Looks at laptop page file* 672MB, I didn't tinker with these settings. Laptop is a Athlon 2400+ 512mb DDR.

Change the setting to desktop performance mode and see what you get.

IST

It's a PC maganize. Those are notorious for errors for Christ's sake. While I have read that this one isn't quite as bad as the others, I still don't trust it. Show me a web link to other than their site and I'll believe ya.

Well, if you don't trust the biggest mag in the industry, whom can you trust???

I think that 5000X is a more realistic target.

PSX2 saw a 45~50X performance jump over PSX it replaced(1.5 million vertices/s to 66 million vertices/s. 33 Mpixles to 1200 Mpixels), less in real world performance because of development problems associated with its poorly designed architecture. PSX3 will see half a leap(~20X) over PSX2 it will replace.

 

[Deadmeat]

...

http://www.buycomputermemory.com/how-much-do-i-need.html

Memory upgrade offers the biggest bang for your buck, your system runs faster with just more memory because of reduced paging demand(the real performance killer), while switching to faster memory type won't do a damned thing for reducing paging....

 

[I.S.T.]

Re: ...

Paul

BOTH GPU'S WERE USING DDR, ONE WITH 64MB AND THE OTHER 128, THE ONE WITH 64 OUTCLASED THE RADEON 8500 BY OVER 1000 POINTS IN 3DMARK.

What's your point? Are we talking about CPU or GPU here???

You said bandwidth doesn't matter. Only RAM size. He used the GPU benchmark to show that it does matter.

IST

It's a PC maganize. Those are notorious for errors for Christ's sake. While I have read that this one isn't quite as bad as the others, I still don't trust it. Show me a web link to other than their site and I'll believe ya.
Well, if you don't trust the biggest mag in the industry, whom can you trust???

You haven't heard of the Maximum PC controvercy, have you?

I think that 5000X is a more realistic target/
PSX2 saw a 45~50X performance jump over PSX it replaced(1.5 million vertices/s to 66 million vertices/s. 33 Mpixles to 1200 Mpixels), less in real world performance because of development problems associated with its poorly designed architecture. PSX3 will see half a leap(~20X) over PSX2 it will replace.

Will, eh? So you know this....How? We don't know the specs yet. What if say it has 512 megs of RAM instead of 256, like you are claiming. Or maybe the Cell direvtive has 64 APUs. We don't know. It's waaaaaaay too early to tell.

Sorry if the quoting looks rediculus, not used to quoting on phpBB.

 

[Fox5]

Re: ...

Ok, I screwed up because I originally typed 4th gen as having 64 MB from the Xbox example, then reversed to 32 MB to represent the typical 4th gen console configuration(PSX2) because this was supposed to be an illustration of general case. I forgot to update other numbers thereafter.

OK, SCEI specific example

2nd Generation(SNES) : 184 KB(I am throwing in SNES number for comparison)
3rd Generation(PS) : 3.5 MB : 19X jump
4th Generation(PS2) : 40 MB : 11.4X jump
5th Generation(PS3) : 274 MB : 6.85X jump

As you can see, the performance leap from previous generation is getting smaller... The graphical leap from SNES to PSX was groundshaking, while the graphical leap from PSX2 to PSX3 won't be as clear....

Thought ps2 was only 32MB, not 40. And where do you get 274 MB for ps3 from?

BTW, I guess deadmeat would have a point that 1024 MB of SDR would be faster than 512 MB of DDR......if the sdr was dual channel compared to single channel ddr or it had a higher bus bittage, or ran at twice the speed. He'd probably also be right for non-gaming applications(performing about the same as ddr), which includes benchmarks like 3dmark to some extent(I believe a 1 ghz p3 with sdr will outperform a 1.5 ghz p4 with ddr). It'd also be true if both the sdr and ddr had more than enough bandwidth for the cpu and system.(no need for an athlon xp to have 20 gigabytes of bandwidth....wait, actually yeah there would be, I think that's faster than the cache speeds of the cpu)

http://techreport.com/reviews/2001q1/p4-vs-athlon/index7.x
But here ddr can outperform sdr by a large margin, even when the computer shouldn't be limited by bandwidth that much yet.(the athlon 1.1 gig I believe has sdr, versus the athlon 1.2ghz with ddr) True, it isn't twice the sdr versus half the ddr, but I don't think that would matter. I happen to have a second computer set up with 640MB of ddr, and it has pretty much nothing on bootup except windows xp and whatever drivers it loads(so no memory hogging microsoft office booting up or something), and I can safely disable the page file and will run any game basically forever with no problems, yet if the computer had to make use of more than that 640MB of ram even once, the computer or program would crash. Thus I think I can safely say 512 MB ddr would maintain its large performance advantage over a gig of sdr, unless you somehow managed to use over 512MB of ram for a significant portion of time, which would probably require multiple applications running.
However, I would say if you're running a sub ghz computer, sdr ram is probably a safe choice, and even ok for athlons, but not for the bandwidth starved p4's or motherboards with lots of integrated stuff.

http://www.hardwareanalysis.com/action/printarticle/1499
If you look here, you'll see a p4 with rdram(hey, it still has bandwidth) greatly outperforming a p4 with sdram in games, and most other things as well.
"DRV-07 in particular performed no less than 50% better when paired with RDRAM! We double and triple checked these results, and they kept coming up the same. Suffice it to say that users interested in professional graphics should definitely look toward a high-bandwidth memory solution."

Come on, memory bandwidth has been one of the biggest limitations for comptuers for quite some time, or they wouldn't have caches on cpus, and wouldn't have to invent all sorts of bandwidth saving techniques.(I think that's what sse and hyperthreading pretty much are)

 

[I.S.T.]

Re: ...

Ok, I screwed up because I originally typed 4th gen as having 64 MB from the Xbox example, then reversed to 32 MB to represent the typical 4th gen console configuration(PSX2) because this was supposed to be an illustration of general case. I forgot to update other numbers thereafter.

OK, SCEI specific example

2nd Generation(SNES) : 184 KB(I am throwing in SNES number for comparison)
3rd Generation(PS) : 3.5 MB : 19X jump
4th Generation(PS2) : 40 MB : 11.4X jump
5th Generation(PS3) : 274 MB : 6.85X jump

As you can see, the performance leap from previous generation is getting smaller... The graphical leap from SNES to PSX was groundshaking, while the graphical leap from PSX2 to PSX3 won't be as clear....

Thought ps2 was only 32MB, not 40. And where do you get 274 MB for ps3 from?

BTW, I guess deadmeat would have a point that 1024 MB of SDR would be faster than 512 MB of DDR......if the sdr was dual channel compared to single channel ddr or it had a higher bus bittage, or ran at twice the speed. He'd probably also be right for non-gaming applications(performing about the same as ddr), which includes benchmarks like 3dmark to some extent(I believe a 1 ghz p3 with sdr will outperform a 1.5 ghz p4 with ddr). It'd also be true if both the sdr and ddr had more than enough bandwidth for the cpu and system.(no need for an athlon xp to have 20 gigabytes of bandwidth....wait, actually yeah there would be, I think that's faster than the cache speeds of the cpu)

http://techreport.com/reviews/2001q1/p4-vs-athlon/index7.x
But here ddr can outperform sdr by a large margin, even when the computer shouldn't be limited by bandwidth that much yet.(the athlon 1.1 gig I believe has sdr, versus the athlon 1.2ghz with ddr) True, it isn't twice the sdr versus half the ddr, but I don't think that would matter. I happen to have a second computer set up with 640MB of ddr, and it has pretty much nothing on bootup except windows xp and whatever drivers it loads(so no memory hogging microsoft office booting up or something), and I can safely disable the page file and will run any game basically forever with no problems, yet if the computer had to make use of more than that 640MB of ram even once, the computer or program would crash. Thus I think I can safely say 512 MB ddr would maintain its large performance advantage over a gig of sdr, unless you somehow managed to use over 512MB of ram for a significant portion of time, which would probably require multiple applications running.
However, I would say if you're running a sub ghz computer, sdr ram is probably a safe choice, and even ok for athlons, but not for the bandwidth starved p4's or motherboards with lots of integrated stuff.

http://www.hardwareanalysis.com/action/printarticle/1499
If you look here, you'll see a p4 with rdram(hey, it still has bandwidth) greatly outperforming a p4 with sdram in games, and most other things as well.
"DRV-07 in particular performed no less than 50% better when paired with RDRAM! We double and triple checked these results, and they kept coming up the same. Suffice it to say that users interested in professional graphics should definitely look toward a high-bandwidth memory solution."

Come on, memory bandwidth has been one of the biggest limitations for comptuers for quite some time, or they wouldn't have caches on cpus, and wouldn't have to invent all sorts of bandwidth saving techniques.(I think that's what sse and hyperthreading pretty much are)

Well, the early P4s were horrible, to say the least. I'm not surprised to hear that a 1GHZ P3 outperforms a 1.5GHZ P4. Other than that, I agree with your post. Edit: Typo fixes.

 

[Megadrive1988]

Thought ps2 was only 32MB, not 40. And where do you get 274 MB for ps3 from?

40 MB is PS2's total amount of RAM

(32 MB main, 4 MB embedded on GS, 2 MB audio, 2 MB for PS1 CPU which is PS2's I/O)

a straight comparison would be:

2 MB main memory in PS1 - 32 MB main memory in PS2.

or

3.5 MB total memory in PS1 - 40 MB total memory in PS2.

 

[Lazy8s]

This whole goal of 'X times more powerful' is just PR, so they will of course achieve it as such - with the same spinning of ambiguities and after-the-fact comparisons.

All of this makes me wonder how much more powerful Sony would say the Xbox is compared to their PS2, considering they rate PS2 at 300x the "performance" of PS and capable of handling "nearly 50 times more 3-D image data" than DC. I'd have to imagine Microsoft's own figuring of a three fold "performance" improvement for Xbox over PS2 would be quite conservative going by Sony's metric.

 

[Guden Oden]

Lazy,

Your post sounds like (very) sour grapes to me. Knock it off, okay?

Ok, so Sony marketing is full of BS, in what way is that news to anyone? The same's true of all marketing departments, Eidos spindoctors keep telling us what kickass game this new Lara Croft title is, rest of the world goes *yawn* and fires up CS instead, etc etc.

Why get so worked up about it? PS3 will be more powerful than PS2. Wether it will be 18.000x PS or not won't change it will be a totally kickass machine. Let the marketing liars do their stuff, we just wanna play great games!