Congratulations to Vince

[cthellis42]

I think the main thing that changes about the first figures and bandying about is that the 1TFLOP figure, fanciful or not, was aimed for the longest time as split between the "CPU" and "GPU" engines. Considering the GPU is both definitive and outside a "pure CELL" structure, how does that figure in relation?

Can the still rack up that contentuous FLOPS total between all the chipwear in the PS3 and between ALL manners of measurement? (And here we bring "NVFLOPS" into play again.) Perhaps. Is it worth it, and does it tell us anything...? Not really.

I do think people still looking to see as much CELL as was contemplated from the beginning are fooling themselves, though. Considering the talk the past few years was that half of it would be applied to the Visualization end (Or at least be the part with Visualization-geared alterations.) and we know that to be coming from another source instead.

 

[Acert93]

By the way, PS3 RAM will have 100 GB bandwidth, not 50

Are you sure?

David Wang wrote:

6. As previously announced, the off chip I/O interface is Rambus Redwood and the memory interface is XDR. Similar clocking/deskewing schemes. Looks to be about ~50 GB/s BW to memory, and 50~100 GB/s to I/O.

Memory bandwidth is the same as the projected (same as the patents?) 50GB/s.

 

[Mythos]

The issue is not pure power but priceerformance. Remember, these machines need to come in at ~$300 and hit ~$200 in 2-3 years.

I never said not to increase the power at all, but consider: Top of the line P4 is in the high 20's gof GFLOPs (that was off a website, IGN I believe) and XeCPU is rumored to be ~80ish GFLOPs (no on who knows is saying, so pure speculation). If 1 CELL is 256GFLOPs you are looking at 8x the performance of a top of the line desktop and 3x the performance of the XeCPU (although comparisons at this point are sketchy).

Sony performance figures for the PS3 are independent of what Microsoft wants to accomplished with their system. Sony reaching their goals have little (I'm sure there is a little eye out for competition) to do with what power is under the Xbox 360/Xenon. With that in mind lets not forget how much work ST/I has already put in to leading edge manufactoring Tech. @ 65NM to reach their performance goals out of Cell/GPU.

 

[Li Mu Bai]

Quote:
The issue is not pure power but priceerformance. Remember, these machines need to come in at ~$300 and hit ~$200 in 2-3 years.

I never said not to increase the power at all, but consider: Top of the line P4 is in the high 20's gof GFLOPs (that was off a website, IGN I believe) and XeCPU is rumored to be ~80ish GFLOPs (no on who knows is saying, so pure speculation). If 1 CELL is 256GFLOPs you are looking at 8x the performance of a top of the line desktop and 3x the performance of the XeCPU (although comparisons at this point are sketchy).

I completely agree here. Ummmmm......are we completely forgetting that Sony is also competing in the handheld sector as well now? Losing considerable money on each PSP sold? Check. Is Sony already operating in debt? Check. Does CELL have much farther reaching impending tech. applications other than the PS3 that have yet to be implemented by Sony? Check. Will the PS3 be sold at a loss initially? Check. Why add more power at a cost, when you are already substantially in the lead? Sony is not MS, & even MS couldn't/wouldn't pull another XBX this time around.

 

[DemoCoder]

They should fit in the maximum amount of power they can to hit their price point. If they plan to sell the thing for $300, and lose $150 per unit as a loss leader, than they should fit the technology that hits that point.

Yes, 1 TFLOPs is arbitrary, but so is the line you are drawing. The fact is, we don't know what their economic calculations are, or what the business plan is. Perhaps they are willing to burn a few billion on CELL, subsidizing its costs, so that they can own more of the marketplace in the future. If CELL seriously hurts x86 marketshare, Sony will go along way towards staving off Microsoft and other cheap commodity hardware competitors (those running home/media devices on x86/strongarm with Linux/WINCE)

Perhaps Sony calculates huge gains from other markets, and PS3 is merely a loss-leader to entrench them. What is it worth to them to ensure that no tweaks to Intel/AMD, no crappy dual-core x86s, will provide any competition? If CELL really can deliver 10x the performance of x86s, then MS is is *serious* trouble in the server market, and even may be in trouble in the desktop market, if cheap, high performance Linux and Mac OS X based clients deliver unbeatable performance and experience.

The STI alliance may actually be able to finally to free us from the Wintel duopoly. But to get people to upgrade, you've got to offer something so compelling, that the switching cost is worth it, and that means earth shattering performance.

If you think they should put in less power, you'll have to show how much the PS3 is going to cost, how much they plan to spend to agressively market it, and how much ROI they expect on future software sales, and sales of other home media products, and that the costs do not justify it.


Ask yourself why IBM wanted to sell their PC division? Because they know CELL can destroy the PC. Sony knows this too, and the STI is betting they can make a play for the whole Wintel monopoly.

 

[Li Mu Bai]

Perhaps Sony calculates huge gains from other markets, and PS3 is merely a loss-leader to entrench them. What is it worth to them to ensure that no tweaks to Intel/AMD, no crappy dual-core x86s, will provide any competition? If CELL really can deliver 10x the performance of x86s, then MS is is *serious* trouble in the server market, and even may be in trouble in the desktop market, if cheap, high performance Linux and Mac OS X based clients deliver unbeatable performance and experience.

The STI alliance may actually be able to finally to free us from the Wintel duopoly. But to get people to upgrade, you've got to offer something so compelling, that the switching cost is worth it, and that means earth shattering performance.

While I didn't want to go into depth, these were exactly the future tech. implementations I was referring to DC that should be taking precedence. A calculted risk to be subsidized currently by what? Their sub-par & loss-incurring TV/DVD/Electronics division? To gamble on a server or PC platform base that does not yet currently exist? And will indeed take time to cultivate a mass adoption of? With no forthcoming technical competition by MS? If launching with a Blu-ray player (as stated) an $150 loss estimate is not nearly high enough. Earth shattering performance will not be best realized upon a gaming platform, when your scope is so much larger.

 

[DemoCoder]

The server side is not really a gamble. Rack mounted cell servers running Linux certainly would have a market. The risky part is beating x86 on the desktop. Apple could help along way there.


Did you ever stop to think that maybe, just maybe, the PS3 is intended to be more than just a gaming box, and is fact, a Trojan horse in disguise?

 

[Geeforcer]

Wasn't the exact same thing said about PS2 back in 1999, when everyone was touting the convergence around the “digital hubâ€￾ that was "PS2, the living room supercomputer"? For all the hype and conjecture, PS2 never really moved beyond a simple gaming console.

 

[Megadrive1988]

It's Broadband Engine that everyone has been predicting to be a TFLOPS, this "prototype" shown today is just a PE.

I find it funny how the patent(and it's a few years old by now) was still on point about the structure and even about many of the performance figures of the prototype revealed today....

yes.

 

[Megadrive1988]

Acert,
The goal shouldn't be to just build a console better than the X-Box2. If Sony was the only company (monopoly on video game market) would you suggest that they never increase power at all?

The purpose of having such a huge increase in power is to meet the needs of future games, which need massive amounts of processing to realistically simulate physics, AI, sound, etc -- to increase realism and believability.

It takes a man of vision, crazy, or creative genius to take a big financial risk, and go all out on something revolutionary, but that is how we make the big leaps.

I disagree. The issue is not pure power but priceerformance. Remember, these machines need to come in at ~$300 and hit ~$200 in 2-3 years.

I never said not to increase the power at all, but consider: Top of the line P4 is in the high 20's gof GFLOPs (that was off a website, IGN I believe) and XeCPU is rumored to be ~80ish GFLOPs (no on who knows is saying, so pure speculation). If 1 CELL is 256GFLOPs you are looking at 8x the performance of a top of the line desktop and 3x the performance of the XeCPU (although comparisons at this point are sketchy).

So I ask: Why would they need to make it even more powerful? Would not the cost for more CELLs be better spent on other areas? It is like the body builder with big arms and chicken legs. He might impress a few people, but when it comes to the competition or sports the well rounded athlete kicks his butt.

I agree we need more power for AI, physics and general believability but 1 CELL is a good step toward that. 1TFLOPs is a totally arbitrary number. Wny not 2TFLOPs? or 16TFLOPs? 1 CELL is substantially faster than what is going to be available, so why 4? It will cost more AND most cross platformed games wont take advantage of it.

If the concern was "massive amounts of processing to realistically simulate physics, AI, sound, etc" on a $300 machine I would look to spend some more resources on memory. This is just me, but a 256GFLOPs CELL and 512MB of XDR is more appealing to me than a 1TFLOP machine with 256MB of XDR although we do not know enough to compare pricing. A 1TFLOPs/256MB PS3 may cost a lot more than a 256GFLOPs/512MB CELL so the comparison would not be fair.

Anyhow, good design is more important than putting all your eggs in one basket. 1TFLOPs is just a number. The rest of the system needs to be able to feed that CPUs. And at a $300 price point, I would think if you were beating your competition by 3x in CPU performance you would look to put more money in other areas. You have to remember if they spend too much on CELL and neglect other areas that leaves room for competitors to move in. A great CPU with a weak GPU would be worse than a solid CPU and GPU. Balance is more important than pure power.

this is all true. besides, it will be the Nvidia GPU and memory architecture that will determine what we will be seeing on screen, in terms of a sheer visual leap from current generation.

Sega's Model 3 board in 1996 was not a huge leap because of the CPU it had (a 66 MHz 603e) it was a huge leap because of the performance and quality of its GPUs.

therefore, I am much more interested in the Nvidia GPU for PS3. what visualizing feature it will provide, that Sony would not have had if it built its own GPU or partnered with Toshiba. I am ready to see Playstation games with tremendously good image quality, texture, pixel shaders, compared to what the PS2 could provide.

 

[Megadrive1988]

PS3 will reach the teraflops figure. Just wait the March presentation. Cell chips in PS3 will be fabbed at 65 nm. Congrats to Vince and Panajev.

1. Does it need to reach 1TFLOPs? If XeCPU is in the 80GFLOPs range, why would it need 1TFLOPs? I am not going to compare them yet (we still do not know enough about both designs... XeCPU seems to be a good at general processing with multiple PPC cores and if the CELL is doing vertex shading and XeCPU is not the CPUs are then really playing different designs goals), but lets say CELL is 2-4x more powerful. Why would it then need to be 8-16x as powerful? That power costs money and with Rambus RAM, BR, licensed GPU tech, and a lot of investment to make a return on plus a lot of work to be done on making development easier/faster/more affordable for developers that extra money may be best invested elsewhere. Also, smart design is better than spamming power everywhere. Will the PS3 have enough memory and memory bandwidth to feed 1TFLOP? If the memory bandwidth is shared you have a 1TFLOPs CELL and a GPU fighting for ~50GB/s. Considering current top of the line GPUs use ~32GB/s I am not sure there will be fast enough RAM, or enough, to feed 1TFLOPs. Maybe there is, but smart effecient design is important and there is no point having power sitting idly by.

Power is nice, but smart design is better. If the PS3 ends up with 256MB of RAM then where is the PS3 going to get the memory to feed that much power? CELL + nVidia GPU + BR + Rambus is a killer combo. I would not be totally surprised if the PSE BE has 2 CELLS (although 1 looks like a safe bet), but I would be shocked at 4.

Prepare to be shocked. That's what Kutaragi is aiming for since 2000. He is so serious about his next console it's scary.

By the way, PS3 RAM will have 100 GB bandwidth, not 50

1 TFLOP would not shock me. it would simply be 'really nice'. what would shock me would be a 4-8 TFLOPs Playstation3. something that is in the ballpark of 1000x the PS2 performance, what Sony has been boasting about since 1999.

 

[Acert93]

They should fit in the maximum amount of power they can to hit their price point. If they plan to sell the thing for $300, and lose $150 per unit as a loss leader, than they should fit the technology that hits that point.

No one argued against this. It is a given that they coming in at a high price, and even then are losing money. PS3 will probably come in at $300-$400 and lose another $100-$200 per unit. Of course Consoles are designed on economics of scale and process shrinks that save money by the time casual consumers start purchasing the product. Initially it is the early adopters who buy at launch and suffer through small game libraries with generally poor quality. I know, I have been there for 3 going on 4 generations of this.

Yes, 1 TFLOPs is arbitrary, but so is the line you are drawing.

But you did not dry a line. You stated:

PS3 will reach the teraflops figure. Just wait the March presentation.

All we ever did was give reasons WHY that may not be the best move for Sony. And while the 1TFLOPs number is totally arbitrary ("One just sounds so good!" and "990GFLOPs is not enough!") the reasons I gave were not: Cost of the console, heat, power consumption, yield issues, etc... Li Mu Bai also gave a lot of reasons why puting MORE Cells in "just because" could harm them financially. No one knows the budget for the PS3, the fact is you are the one making claims it WILL have 1TFLOPs. We countered that comment with information released today asking why it will, or even should. The only arbitrary part of all this was your claim. Where is the proof?

SiBoy also has STI on record calling the presented CELL (1:8, 256GFLOPs) as the Broadband Engine.

The fact is, we don't know what their economic calculations are, or what the business plan is. Perhaps they are willing to burn a few billion on CELL, subsidizing its costs, so that they can own more of the marketplace in the future. If CELL seriously hurts x86 marketshare, Sony will go along way towards staving off Microsoft and other cheap commodity hardware competitors (those running home/media devices on x86/strongarm with Linux/WINCE)

Does Sony have money to take such gambles?

Btw, I may get a PS3 but there is no way I am going to a CELL platform. I have thousands of dollards in software and hardware/periphrials and as a business owner there is no way I am tossing that out to be a lab rat. You can talk about "crappy x86" dual cores, but until we see how CELL handles desktop aps and the entire spectrum of typical user tasks AND supports the MASSIVE x86 library of software (and does so well) CELL will have a hard time doing what you want it to do. Which is good because I am not aware of Sony (or IBM) being ready to actually server that type of marketbase.

If you think they should put in less power, you'll have to show how much the PS3 is going to cost, how much they plan to spend to agressively market it, and how much ROI they expect on future software sales, and sales of other home media products, and that the costs do not justify it.

No one ever said they should put in less power!! Sheesh! You made the claim of the PS3 CELL being 1TFLOPs. All we did was point out some reasons why that may not be as easy as, "1TFLOPs sounds great, so be it!" And if SiBoy is right then BE is 256GFLOPs.

This all comes down to you wanting some majical number--not us trying to put less power in the PS3. Do you realize that in FLOPs the PS3 is 8x more powerful than a top of the line P4 (and will be about 4x as powerful by launch) and that depending on how many CPUs are put into the Xenon it could be up to 3x faster than the XeCPU??

We are not talking 10%, or 50%, or even 80%. We are talking magnitudes better, as in 3-8x faster in FLOPs. The fact is the system needs the memory to feed the CPU and GPU and the GPU needs enough power to output what the CPU is inputing. If the projection of 256MB is correct I have no clue why you would want them to go with 4 CELLs (which will cost more no matter how you cut it) and stick with such a small amount of RAM that both the GPU and CPU must use. From a design perspective it makes no sense to have limited RAM (which will impact performance and the quality of software) and have 4 CELLS when you could have 1 CELL with 3x the performance of your competitor and spend the saved cash on more RAM.

Since the 256GFLOPs CELL is being called the BE, it is up to you to prove why "PS3 will reach the teraflops figure. Just wait the March presentation." And no, adding in the nvflops does not count

 

[Megadrive1988]

oh yeah, as someone mentioned in this thread, if Sony adopts Nvidia's 'Nvflops' Sony could easily have its '6.2+ TFLOPs' PS3 that they could claim is 1000x the PS2


btw, I am predicting PS3 CPU will be 256 or 512 GFLOPs. not 1 TFLOP.

even 256 GFLOPs is 2 to 3 times more than what Xenon CPU is reported to be.

 

[DemoCoder]

Since the 256GFLOPs CELL is being called the BE, it is up to you to prove why "PS3 will reach the teraflops figure. Just wait the March presentation." And no, adding in the nvflops does not count

Just one problem: I never said the quote you are attributing me. (watch march presentation. What presentation?)

 

[Acert93]

therefore, I am much more interested in the Nvidia GPU for PS3. what visualizing feature it will provide, that Sony would not have had if it built its own GPU or partnered with Toshiba. I am ready to see Playstation games with tremendously good image quality, texture, pixel shaders, compared to what the PS2 could provide.

I agree. The 6year PS2 cycle is kind of long and I am ready for HD video gaming! With 256GFLOPs for AI, Physics, logic, etc... the games should just come alive. If CELL does Vertex Shading that means the nVidia GPU may only do Pixel Shading, and if that is the case that is a lot of transistor realestate to push out some insane effects. Throw in the XDR to prevent the system from bottlenecks and then top it off with BR for HD Movies and you got a dream system. Yeah, 500GFLOPs or 1TFLOPs would be nice, but the way the PS3 is shaping up who cares?

I still think Nintendo and MS will be competitive (even in hardware) because their focus will be different (Nintendo on some innovation and clean design; MS with a clean/effecient design with a lot of flexibility) but there is no denying the PS3 is going to be a good console for a HW perspective. And love them or hate them (personally I prefer Nintendo 1st party games and am a PC gamers addicted to BF1942 type games), Sony has software that makes a lot of gamers happy. If Sony can deliver on the software end--meaning helping developers make games cheaper, quicker, and easier--then this next gen is gonna be hard for MS or Nintendo to gain market share, let alone significant marketshare.

 

[Megadrive1988]

btw, I'm sure if STI wanted to, they could design a Cell-2.0 or Cell-1.5 with a dual-core, quad threaded POWER5+ based PU/PPE and 16 APU2s/SPE2s
that provided a good 512+ Gflops, all on 65 nm or 45 nm

 

[one]

btw, I am predicting PS3 CPU will be 256 or 512 GFLOPs. not 1 TFLOP.

Super-optimistic prediction:

3 PPEs or 2 PPEs, with 24 SPEs @ 5GHz = 960GFLOPS

 

[Phil]

Acert93:

(1)
Broadband Engine was nothing more but the name of 4 PEs described in the corresponding patent. It was merely popular belief that the broadband engine would be PS3's configuration. A PE being called Broadband Engine, I doubt, proves anything about what's going to be in PS3 nor do I think will Sony stick to a specific configuration just because they changed the description in mid term development.

(2)
Just because you won't buy into a lab rat doesn't mean a company that wants to broaden its horizon won't check out possibilites OR take risks. Calculated risks are often a big factor in success - as such, it should be clear that CELL is a major undertaking and has cost Sony as a company millions in building what's necessary to make a viable product out of it, that is planned to be integrated into many things. If PS3 is among this strategy is left to be seen and surely has a big influence if we'll see a single CELL configuration or an unbelievable 4 CELL / 1TFLOP monstor. It's all relative. At this point, WE don't know what Sony's definite plans are with CELL and how far they'll take things (including risks). Just because YOU don't see it because you choose to shut out anything non-gaming related doesn't mean Sony does as well (in fact, I see things pointing in the opposite direction).

 

[Acert93]

Since the 256GFLOPs CELL is being called the BE, it is up to you to prove why "PS3 will reach the teraflops figure. Just wait the March presentation." And no, adding in the nvflops does not count

Just one problem: I never said the quote you are attributing me. (watch march presentation. What presentation?)

Correct, that was Devourer. Sorry for the confusion. I apologize and will take out the quote. It is late (but that is not an excuse). Please forgive me.

Btw, I think the part you are misreading in what I said was the comparison to the Xenon and PC. You are looking at it as "More powerful is better". I am looking at it as "The base design is 3x better than our competition. There is no reason to burn money making it 6x or 12x more powerful just to claim we are more powerful when we have to consider a ton of factors like heat, power consumption, die size and yield, and also have an expensive optical drive that is brand new, a licensed cutting edge GPU that runs $400 at retail, and cutting edge RAM that required a premium--and we got to cram that all in for $300 and try to break even in 3 years at $200".

I am not saying Sony should make the PS3 less powerful. I am suggesting that statements that they should spend even more money (how many Billions did they spend to develop CELL?) just to get to 1TFLOPs is bad business. The current 90nm CELL is like 200nm^2. Even with a process shrink you are looking at a HUGE die with 4 CELLs. That means bad yields, a lot of heat and probably more power, a bigger system case, more expensive die, and so on. This is a $300 consumer GAMING device. What I am looking for is a good reason to go for 1TFLOPs and how they expect to manage all these issues.

Obviously you have some Wintel resentments, but put those aside and look at it from Sony's viewpoint. Why should they go for 1TFLOPs (besides the Wintel deal)? And if they should, how do you suggest they overcome all the hurdles? Heat, power consumption, case size, etc... are not going to go away just because you throw money at it. 1TFLOPs would be great, and I would not be too shocked if 500GFLOPs (2 CELL) BE was announced. But as it stands right now it looks like 256GFLOPs is what it will be--and you know what? That is mindboggling!

Only people who are fixated with 1TFLOPs are dissappointed. For them, all I can say is go play with your Emotion Engine!

 

[PC-Engine]

IIRC Vince predicted the single chip "peferred embodiment" (4 PE 1 TFLOPS BE) would be in PS3 with 32MB-64MB of eDRAM. He also predicted an in-house Visualizer and that PS3 would not follow the CPU/GPU PC paradigm.

He was absolutely sure that it would have a lot of eDRAM siting it's high bandwidth benefits used to help deliver high realworld FLOPS numbers. The reason why I remember was due to the fact I predicted CELL would be very inefficient and realworld performance would suffer.

He also predicted PSP would use a cutdown CELL and that HD DVD had already lost even before the format war began and that no studios would support it.