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

[Deadmeat]

Funny I can walk into bestbuy and pick out SDRAM from a range of different brands..

So do you get two SDRAM DIMMS for the price of one DDR DIMM?? Or do you pay about the same price??

PC133 Standard Modules
Memory Module Part Number Retail Price You Save Your Price Qty Order
128MB, PC133, 168 Pin DIMM, 3.3v (High Density) 80933 $41.10 $4.11 $36.99
128MB, PC133, 168 Pin DIMM, 3.3v (Low Density) 81065 $54.43 $5.44 $48.99
256MB, PC133, 168 Pin DIMM, 3.3v (High Density) 80385 $66.66 $6.67 $59.99
256MB, PC133, 168 Pin DIMM, 3.3v (Low Density) 80384 $103.32 $10.33 $92.99
512MB, PC133, 168 Pin DIMM, 3.3v 80631 $127.77 $12.78 $114.99

PC266 / PC2100 DDR
128MB DDR266, PC2100 80092 $38.88 $3.89 $34.99
128MB DDR266, PC2100, ECC 80120 $44.43 $4.44 $39.99
256MB DDR266, PC2100 80393 $55.54 $5.55 $49.99
128MB DDR266, PC2100, ECC, Registered 80121 $66.66 $6.67 $59.99
256MB DDR266, PC2100, ECC 80415 $69.99 $7.00 $62.99
256MB DDR266, PC2100, ECC, Registered 80416 $77.77 $7.78 $69.99
512MB DDR266, PC2100 80636 $94.43 $9.44 $84.99
512MB DDR266, PC2100, ECC 80655 $105.54 $10.55 $94.99

Screw price, more memory = better performance, remember? Bandwidth is irrelevant.

The reason you fell flat on your face on the last arguement is that you didn't understand the memory access pattern of different hosts; CPU types typically read 32~64 bytes at a time to fill a cache line miss, so the cycle saving from DDR transfer mode is minimial because CAS latency overhead isn't reduced. GPU types tend to read longer stream of bytes so that DDR transfer mode does show its benefit there. I actually wrote a paper on this subject before and my calculation showed a performance improvement of around 5% for typical application. On the other hand, doubling of SDRAM memory would drastically improve the paging performance and wipe out whatever the advantage DDR would have, so there is no contest as to which one would perform better on desktops, 1 GB SDRAM or 512 MB DDR.

Sony can shove more amounts of SDRAM into PS3 than they could if they put XDR in, if they had the same price limits on PS3 memory(IE: They will spend 100 on memory for each PS3). So this is the best thing to do no? OR does that new Shiny XDR memory the same price as SDRAM

UMAs require GPU-type memory access pattern because of unified streaming graphics and audio data access. So both PSX3 and Xbox Next UMA systems would benefit from DDR. But not your desktop computer or server.

The Radeon has 2X the memory, but why can it not touch a Geforce 4 ti4200 using half?(64mb)

Why did you link the benchmark of videocard? We were talking about motherboard RAMs???? I am still waiting.

 

[Deadmeat]

...

OK, stop right there, You don't understand a bit about how memory type characteristics affect desktop performance.

Go and read something somewhere. You lost the arguement.

 

[mondoterrifico]

Re: ...

You seem to lack even the most basic understanding of Von Neumann architecture.

??? Which processor has Von Neumann architecture nowadays??? Almost all new CPUs implement Harvard split cache architecture...

Um you were talking about main memory. Now you want to evade your topic. This was not about cache architecture, but about your inane comments that somehow there is a linear relationship between the amount of main system memory and system performance.
Nice try though. I somehow expect you will be able to google some more buzzwords for me to cover your lack of understanding.

 

[Vince]

I actually wrote a paper on this subject before and my calculation showed a performance improvement of around 5% for typical application.

Most impressive, I didn't expect that. Where was it published? I'd like to read it. Thanks in advance.

 

[notAFanB]

Go and read something somewhere. You lost the arguement.

How on earth can he have lost anything when you have failed to show just how the memeory performence charateristics hold int he real world. hell you don't even have figures nor have you EVEN constructed an credible hypotheseis. All you've done is spew more Google garbage which conviniently side track the point.

 

[mr]

Because there really is a linear relationship between RAM size and performance.

Nope. You can never have too much RAM. The simplest and most effective upgrade you can do to your PC is installing more RAM.

Of course a PC with 1024 MB SDRAM will run faster than another PC with 512 MB DDR SDRAM. Don't tell me you didn't know that.(But then again, you probably didn't know that)

More RAM you have, less paging the system needs to perform and the faster the system runs. Bandwidth doesn't matter, only access latency and amount of available RAM capacity do.

I am looking forward to your benchmark that shows 512 MB DDR system blowing away 1024 MB SDR system on benchmarks. I challenged you to come up with one because I know it doesn't exist.

As you obviously know so much about the relationship of RAM types and size I think you need to specify under which circumstances you obtained your findings?

Which OS did you use? (I think you know what can happen on Win9x with 1GB of RAM.)
Which applications did you benchmark? (I think you know that e.g. UT2003 "can" benefit from 1GB of RAM of any type, whereas LAME sits comfortably in 128MB and likes the additional bandwith of DDR)

To make a long story short:
I think you may know one thing or the other about the topics you write about, but your "knowledge" gets lost in broad sweeping statements that anybody can read anything into. Give your statements a little context so people can work with you, otherwise this discussion is a waste of bandwith.

 

[nondescript]

Exactly. More RAM you have, less paging the system needs to perform and the faster the system runs. Bandwidth doesn't matter, only access latency and amount of available RAM capacity do.

Funny you should say that...

DeadmeatGA, a few months back[/url]]
Anyhow, Here is my response to Fox5's question.

Xbox P3 : 3 GFLOPS / 1.06 GBs = 3 FLOPS/byte
Gecko : 0.97 GFLOPS / 1.3 GBs = 0.74 FLOPS/byte

The rule of thumb is that anything over 1 FLOP/byte is a waste of FPU, since FPU will never get enough data to run at full speed.

DeadmeatGA, a few months back[/url]]
FLOPS/memory bandwidth ratio(Lower the better)

SH-4 : 1.6 GFLOPS/0.8 GBs = 2 FLOPS/byte
EE : 4.8 GFLOPS/2.4 GBs = 2 FLOPS/byte
PSP : 2.6 GFLOPS/2.6 GBs = 1 FLOPS/byte
Earth Simulator : 8 GFLOPS/32 GBs = 0.25 FLOPS/byte
BGL : 5.6 GFLOPS/22 GBs = 0.25 FLOPS/byte

CELL(DM version) : 32 GFLOPS/25 GBs = 1.3 FLOPS/byte
CELL(Sony fan version) : 1000 GFLOPS /25 GBs = 40 FLOPS/byte

So bandwidth doesn't matter anymore, eh? No more "Flops per byte" law?

I actually wrote a paper on this subject before and my calculation showed a performance improvement of around 5% for typical application.

I look forward to seeing this paper.

 

[jvd]

Well just to add some fuel to this . Star wars galaxies will see an increase in fps by having over a gig of system ram. Than having less ram at a higher speed.

a 2.5 ghz axp with a 200mhz fsb with 512 megs of ram will get 15% lower frame rates than a 2.5 ghz axp with a 166 mhz fsb and 1 gig of ram .


I remember beta testing that game haha .


Anyway take from it what you will .

 

[Paul]

On the other hand, doubling of SDRAM memory would drastically improve the paging performance and wipe out whatever the advantage DDR would have, so there is no contest as to which one would perform better on desktops, 1 GB SDRAM or 512 MB DDR.

Even if by using SD on a P4 1.7 your not fully using the potential of the P4's FSB???

And don't you love how dual channel brings up total system performance over a system that is not using the configuration? With the same memory amount I might add.

 

[Deadmeat]

...

So bandwidth doesn't matter anymore, eh? No more "Flops per byte" law?

"One FLOP per Byte" law still holds, because Pentium4s don't sustain over 1 GFLOPS in real world code. If P4 actually sustained higher FLOPS then memory would become the bottleneck and moving to a wider bus would improve the performance.

I look forward to seeing this paper.

No you won't. It was my term paper and only two people have seen it, me and the professor. The point I was trying to make was that I was well-aware of the CAS latency impact on the performance of a modern CPU architecture because I actually looked into the subject before.

If Paul was making a memory type performance comparison for GPU applications, I wouldn't have said a thing. But he was talking about the system RAM application, which is affected more by paging frequency and CAS latency than ultimate bandwidth.

 

[Deadmeat]

...

Even if by using SD on a P4 1.7 your not fully using the potential of the P4's FSB???

P4's FSB is not bandwidth restricted. The FSB is almost never utilized to 100% of peak bandwidth. Higher FSB clockspeed improves the performance only because it reduces access latency, not because it offers more bandwidth.

A 32-bit FSB clocking at 1 Ghz would trash a 64-bit FSB clocking at 500 Mhz any day.

And don't you love how dual channel brings up total system performance over a system that is not using the configuration? With the same memory amount I might add.

Same memory amount, yes. But the difference would be marginal(5~10% at best). While doubling the memory capacity would almost certainly boost the performance by at least 30% for heavy duty jobs and this why PC upgrade guides always recommends memory upgrade and HD upgrade before recommending the CPU upgrade, because memory upgrade offers the most bang for your buck.

 

[I.S.T.]

Re: ...

Even if by using SD on a P4 1.7 your not fully using the potential of the P4's FSB???

P4's FSB is not bandwidth restricted. The FSB is almost never utilized to 100% of peak bandwidth. Higher FSB clockspeed improves the performance only because it reduces access latency, not because it offers more bandwidth.

A 32-bit FSB clocking at 1 Ghz would trash a 64-bit FSB clocking at 500 Mhz any day.

And don't you love how dual channel brings up total system performance over a system that is not using the configuration? With the same memory amount I might add.

Same memory amount, yes. But the difference would be marginal(5~10% at best). While doubling the memory capacity would almost certainly boost the performance by at least 30% for heavy duty jobs and this why PC upgrade guides always recommends memory upgrade and HD upgrade before recommending the CPU upgrade, because memory upgrade offers the most bang for your buck.

You're statement about the P4 not being dependant on the bandwidth it has is incorrect. So is the statement that bandwidth does not at all matter. Read this review of the i845 to see why. http://www.techreport.com/reviews/2001q3/p4-sdram/index.x?pg=1

On a side note, I just had to register when I saw that statement by Deadmeat.

 

[Deadmeat]

...

From your review...

Conclusions
So this is what the Pentium 4 looks like with SDRAM. For all the brouhaha over DDR versus Rambus, the real-world performance difference between the two technologies—at least with these chipsets—is statistically insignificant.

With PC133 SDRAM, however, the Pentium 4's greatest performance strengths are much more muted. It isn't true that the Pentium 4 is a terrible performer when you take away a big chunk of the memory bandwidth that it has enjoyed while coupled exclusively to the 850 chipset. The 845's performance isn't stellar, but it's only about 5 to 15% slower than the 850, most of the time. Still, in some of those places where the P4 has excelled, like Quake III, the 845 exacts a sizeable speed penalty. PC enthusiasts and consumers will want to avoid the 845/PC133 combo. Both the 850/RDRAM and P4X266/DDR combinations offer much better overall performance, and AMD's Athlon with DDR memory remains the best value.

 

[I.S.T.]

The performence DID go down though, because of the fact that the bandwidth just wasn't there. Also, you forgot to hilight a part of that quote.

Still, in some of those places where the P4 has excelled, like Quake III, the 845 exacts a sizeable speed penalty.

I wonder why is that? The review states that the i850 and the i845 were extremely simular other than the memory controller, after all.

 

[Paul]

Same memory amount, yes. But the difference would be marginal(5~10% at best).

Yup thanks for proving my right DM, so you now admit that it's not only memory amounts.

Now as for the only 5-10% I want you to find benchmarks.

 

[DemoCoder]

I can MPEG-4 Compress a stream on a P4-3Ghz with 128mb of RAM far faster than you can do it on a P3 1Ghz with 1.5Gb of RAM and that's despite the fact that you could "preload" the entire dataset to be compressed off the harddrive.

Deadmeat, you're friggen clueless. Huge amounts of RAM only matter if you have to make multiple passes over a huge datastructure. Not all algorithms require large persistent datastructures and not all are bound by datarate.

You need to go back to school and take a basic course in algorithms and hardware architecture.

 

[Deadmeat]

...

IST

The performence DID go down though, because of the fact that the bandwidth just wasn't there. Also, you forgot to hilight a part of that quote.

My point is that you see a far greater performance boost from doubling SDRAM capacity than to switch to DDR of same capacity.

Paul

Yup thanks for proving my right DM, so you now admit that it's not only memory amounts.

Actually I should thank you, since you supplied us with a benchmark that proves that you gain little from switching to DDR of same capacity, while you gain BIG from doubling the RAM capacity.

Now as for the only 5-10% I want you to find benchmarks.

You showed us yourself. Go back and look.

To DemoCoder

I can MPEG-4 Compress a stream on a P4-3Ghz with 128mb of RAM far faster than you can do it on a P3 1Ghz with 1.5Gb of RAM and that's despite the fact that you could "preload" the entire dataset to be compressed off the harddrive.

That's not what was being asked here.

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

The answer is 1024 MB PC133 for most applications. Benchmarks prove this time after time.

Huge amounts of RAM only matter if you have to make multiple passes over a huge datastructure. Not all algorithms require large persistent datastructures and not all are bound by datarate.

I just checked my system's memory allocation and the XP allocated 1.2 GB, but I have only 256 BM onboard. This system would keep gaining speed until 1.2 GB of memory is filled. Same with yours too.

 

[Paul]

Actually I should thank you, since you supplied us with a benchmark that proves that you gain little from switching to DDR of same capacity, while you gain BIG from doubling the RAM capacity.

You proved me right and you wrong. Thanks again.

The answer is 1024 MB PC133 for most applications. Benchmarks prove this time after time.

Now it's for *most*? I thought bandwidth was irrelevant? Your not very consistant.

 

[Deadmeat]

..

You proved me right and you wrong. Thanks again.

You don't have to give me credit for something I didn't do. On the other hand, you don't have to decline the credit you deserve.

Now it's for *most*? I thought bandwidth was irrelevant? Your not very consistant.

 

[Psychogenics]

DM don't be a jerk!!!! I'll keep it simple......Sony is going out for new processors that NEED new Ram not old ones!!! Old ram is limited no matter how you put it.Why isn't MS using Sdram on X2?Because it's Old thats why.