Merrill Lynch's Next-Gen console prediction

[avaya]

Right... but why would RSX on the same process with roughly ~70 million more transistors than Cell actually be smaller? And going further, with their yield estimates they're completely ignoring the fact that die defects to a certain extent can be tolerated on Cell, whereas their numbers seem to reflect perfect dies.

I mean - don't those reasons make sense to you?

I'm agreeing with you, just pointing out their reasons dude.

 

[Carl B]

I'm agreeing with you, just pointing out their reasons dude.

Sorry sorry. I just misunderstood the context of the evidence you presented I guess. My bad on the misunderstanding!

Fixed in post.

 

[Sis]

The article seems to shamelessly only focus on MS's advantages completely ignoring Sony's advantages. Whether he was bought out or not, the article is shamelessly one-sided, whether you agree or not with it.
Unless you think that Sony has no advantages over MS (which in itself is an outrageous statement), the article is over the top with its Sony iz teh doomed scenarios.

The article focused on the cost advantages. I take it you disagree with the cost analysis and expect the PS3 to have a the advantage?

.Sis

Edit:

In the end MS supporters will always agree with such articles regardless of the validity of them.

Before you implicate everyone who disagrees with your point as an "f-boy", I would suggest taking a more respectful read of the thread. The pros and cons of the report have been thoroughly laid out to which you've addressed none.

 

[Shifty Geezer]

The back-end process comes next, and we forecast yields of around 75% during packaging and testing processes. We assume the number of chips derived will thus further decline from the 73 good chips yielded from front-end processes, ultimately amounting to 55 good chips.

30% yields? Isn't that amazingly crap? Does this include the redundancy consideration? Makes sense that Sony would take the redundancy hit if yields are as bad as 30% for error free, but that should mean the actual useable yields are far higher. 70% perhaps? I've no idea what defect rates ar so I won't even hazard a guess (which I've just done. Umm...ignore that 70% figure everyone!)

Still looking at >$100 per chip. That's terrible. Still hasn't DaveB said price are like that for large GPUs? Maybe that actually is the cost

 

[Mintmaster]

Right... but why would RSX on the same process with roughly ~70 million more transistors than Cell actually be smaller?

Cell runs at 6 times the clockspeed, buddy. You have to worry about the EMF produced and crosstalk. The transistors are designed to switch at a higher speed in Cell and the wires need more room. Another factor is probably that CPU's can't have any error show up, but graphics are a little more tolerant. If every billionth pixel has something wrong, no big deal. If every billionth instruction ran incorrectly, the CPU would be unusable.

e.g:
FX-51 is 105M transistors and 193 mm2 on 130 nm
NV35 is 130M transistors and ~195 mm2 on 130 nm ==> 24% more transistors.
R300 is 107M transistors and ~205 mm2 on 150 nm ==> same density as FX, but with 30% larger transistors/wires.

And the FX-51 isn't designed for sky high clock speeds either.

I think it's quite easy for RSX and Cell to have similar die sizes.

 

[Shifty Geezer]

Well, the other question is - why didn't NEC get the logic portion?

The current solution is probably just representative of where they felt th fabs were with each of the solutions given the timescales they were shooting for. If I were to take a guess then I'd say if/when it does happen TSMC will get the final design; if you take a look around the TSMC site you'll see that they have solutions for eDRAM on each of their major nodes, with 65nm planned as well.

I assumed TSMC didn't have eDRAM capability, hence the need for NEC. If TSMC does have eDRAM, as they say they do, why the need for NEC? Unless they're buckets cheaper for eDRAM. In which case why aren't they cheaper for logic? Sounds to me like two printing presses, one that's cheaper to print pictures of animals and another cheaper to print pictures of people, and a brochure needing both being sent to both companies to print different parts of the same image.

 

[Brimstone]

Right... but why would RSX on the same process with roughly ~70 million more transistors than Cell actually be smaller? And going further, with their yield estimates they're completely ignoring the fact that die defects to a certain extent can be tolerated on Cell, whereas their yield estimate numbers seem to reflect the percentage of perfect dies, as would traditionally be the case.

Well CELL has a lot of SRAM on it which will take up a lot of area I presume. SRAM isn't as dense as eDRAM like on the Gamecube Flipper GPU or Xenos GPU module.

 

[Carl B]

Cell runs at 6 times the clockspeed, buddy. You have to worry about the EMF produced and crosstalk. The transistors are designed to switch at a higher speed in Cell and the wires need more room. Another factor is probably that CPU's can't have any error show up, but graphics are a little more tolerant. If every billionth pixel has something wrong, no big deal. If every billionth instruction ran incorrectly, the CPU would be unusable.

e.g:
FX-51 is 105M transistors and 193 mm2 on 130 nm
NV35 is 130M transistors and ~195 mm2 on 130 nm ==> 24% more transistors.
R300 is 107M transistors and ~205 mm2 on 150 nm ==> same density as FX, but with 30% larger transistors/wires.

And the FX-51 isn't designed for sky high clock speeds either.

I think it's quite easy for RSX and Cell to have similar die sizes.

I think 3.2 GHz is going to be a walk inthe park for these chips - have you seen the schmoo for them? Certainly given a similar architecture on the PPE cores as XeCPU, there's no reason that one PPE and a bunch of SPE's shouldn't reach 3.2 GHz more readily than 3 PPE cores, each one with a chance of not being able to 'make it' at a certain voltage.

Also it's not a good comparison to use NV vs R vs AMD series chips; the architectures and the processes of different fab houses all play a role. NV suffering with a bad IBM process back in NV30 days is just one example of freak occurence and deviation from the norm.

But even with all of that aside - and I grant that the die sizes might be similar in size in the end due to different densities, certainly you understand where Cell with larger economies of scale and the ability to withstand defects should come in naturally lower in price to RSX, right? And to say nothing of the fact that on the same process, with ~70 million more transistors, and using the EE and GS as precedent with Sony, I fully expect RSX to be larger as well.

(Brimstone I agree with your points, but this post applies to you as well )

I mean c'mon, this most recent BOM has the Cell at over three times the cost of RSX in certain places. WTF?

 

[mckmas8808]

How much is such a financial analysis worth, if they fail to look at the bigger picture (technology, the fabbing process, CELL compromising of redundant structures etc) that would change the result of the financial analysis greatly?

In other words, based on the fact that CELL, although larger in die-size, has the advantage of it being used even with defects to a certain degree and the point that defective CELL chips can be used for other products, I for one would be carefull to weigh CELL to be more expensive than Xenon. Considering that, why even take claims of $$$ for chips that aren't taking these factors into account even serious? Certainly doesn't sound "well qualified" if the average B3d member can come up with better estimates based on these few (well known) established facts...

Hence my statement in the first post which I said...

I know everyday posters that can predict better than this.

Yep I got beat up for making that comment though. Well looks like I was right all along. With a few others too of course.

 

[Phil]

What makes you so certain that they didn't?

Because with data missing that only Sony has an inside on, no cost you ever attach to a unit will ever be taking the larger picture into consideration - and that's ultimately that CELL is designed to be used in multiple products, blu-ray's advantage extends well into Sony Group as a whole entity and beyond SCE and PS3 (which this analysis is solely looking at), own fabbing and manufacturing processes and their cost disadvantage/advantage that goes with it... in short, there are simply too many factors that are relevant that make it *impossible* for an outsider company to be calculating with a few $$$ here and a few $$$ there to form any kind of analysis on financial advantages.

In other words, what makes you certain that these analysts have an inside on all the costs & spending of Sony to calculate accurate prices of the multiple units within the PS3?

These numbers have absolutely no substance to them. The prices of these units could be in the first shipment at point x, at the next at point y which all extend into Sony's cost strategy as a whole.

 

[Brimstone]

I think 3.2 GHz is going to be a walk inthe park for these chips - have you seen the schmoo for them? Certainly given a similar architecture on the PPE cores as XeCPU, there's no reason that one PPE and a bunch of SPE's shouldn't reach 3.2 GHz more readily than 3 PPE cores, each one with a chance of not being able to 'make it' at a certain voltage.

Also it's not a good comparison to use NV vs R vs AMD series chips; the architectures and the processes of different fab houses all play a role. NV suffering with a bad IBM process back in NV30 days is just one example of freak occurence and deviation from the norm.

But even with all of that aside - and I grant that the die sizes might be similar in size in the end due to different densities, certainly you understand where Cell with larger economies of scale and the ability to withstand defects should come in naturally lower in price to RSX, right? And to say nothing of the fact that on the same process, with ~70 million more transistors, and using the EE and GS as precedent with Sony, I fully expect RSX to be larger as well.

(Brimstone I agree with your points, but this post applies to you as well )

I mean c'mon, this most recent BOM has the Cell at over three times the cost of RSX in certain places. WTF?

Also the Rambus XDR controller on the CPU is designed to handle a 2 PPE 16 SPE CELL CPU. This may take up more area than the RSX memory controller.

 

[jvd]

Because with data missing that only Sony has an inside on, no cost you ever attach to a unit will ever be taking the larger picture into consideration - and that's ultimately that CELL is designed to be used in multiple products, blu-ray's advantage extends well into Sony Group as a whole entity and beyond SCE and PS3 (which this analysis is solely looking at), own fabbing and manufacturing processes and their cost disadvantage/advantage that goes with it... in short, there are simply too many factors that are relevant that make it *impossible* for an outsider company to be calculating with a few $$$ here and a few $$$ there to form any kind of analysis on financial advantages.

In other words, what makes you certain that these analysts have an inside on all the costs & spending of Sony to calculate accurate prices of the multiple units within the PS3?

These numbers have absolutely no substance to them. The prices of these units could be in the first shipment at point x, at the next at point y which all extend into Sony's cost strategy as a whole.

This report was talking about the ps3 and its costs . For what it is talking about its extremely accurate . We also have another report from another company where prices mainly line up with each other and the end result is the xbox 360 being cheaper to make

 

[Powderkeg]

Because with data missing that only Sony has an inside on, no cost you ever attach to a unit will ever be taking the larger picture into consideration .

Why not?

While you might not have accurate figures certainly you can come up with reasonable estimates for them. It's not like Sony is the only company in the world that has these kinds of expenses, right? It is possible for them to take the larger picture into consideration by comparing that larger picture to similar pictures made by other companies.

Or can you think of some reason why a company that specializes in these types of analysis would be absolutely clueless to the big picture, and wouldn't have any comperable data from any other company on earth?

 

[Phil]

This report was talking about the ps3 and its costs . For what it is talking about its extremely accurate .

On what grounds do you find these numbers extremely accurate? How can you even say something like this based on absolutely nothing?

 

[Sis]

Because with data missing that only Sony has an inside on, no cost you ever attach to a unit will ever be taking the larger picture into consideration - and that's ultimately that CELL is designed to be used in multiple products, blu-ray's advantage extends well into Sony Group as a whole entity and beyond SCE and PS3 (which this analysis is solely looking at), own fabbing and manufacturing processes and their cost disadvantage/advantage that goes with it... in short, there are simply too many factors that are relevant that make it *impossible* for an outsider company to be calculating with a few $$$ here and a few $$$ there to form any kind of analysis on financial advantages.

In other words, what makes you certain that these analysts have an inside on all the costs & spending of Sony to calculate accurate prices of the multiple units within the PS3?

These numbers have absolutely no substance to them. The prices of these units could be in the first shipment at point x, at the next at point y which all extend into Sony's cost strategy as a whole.

While I disagree that the analysts ignored something--they addressed the Cell and Blu-ray and fabbing in the document--I do think that its really unlikely that the analysts have any inside information, especially with regards to dollar figures. This would be a serious compromise to both MS and Sony's competetive advantage, and neither one seems likely to show their cards. The best any outsider can do is estimate (ie, guess), and these analysts are included.

.Sis

 

[mckmas8808]

Because with data missing that only Sony has an inside on, no cost you ever attach to a unit will ever be taking the larger picture into consideration - and that's ultimately that CELL is designed to be used in multiple products, blu-ray's advantage extends well into Sony Group as a whole entity and beyond SCE and PS3 (which this analysis is solely looking at), own fabbing and manufacturing processes and their cost disadvantage/advantage that goes with it... in short, there are simply too many factors that are relevant that make it *impossible* for an outsider company to be calculating with a few $$$ here and a few $$$ there to form any kind of analysis on financial advantages.

In other words, what makes you certain that these analysts have an inside on all the costs & spending of Sony to calculate accurate prices of the multiple units within the PS3?

These numbers have absolutely no substance to them. The prices of these units could be in the first shipment at point x, at the next at point y which all extend into Sony's cost strategy as a whole.

Phil you are 100% correct. Ken Kutaragi said this himself in an interview in May.

Question:
Though it must be difficult for you to tell the very price of PS3, will PS3 take the same depreciation model as the one of PS2? For instance PSP was cheaper than most expected.

Kutaragi: PSP was evaluated by many people as inexpensive, but still 24,900 Yen. Its cost model can have a feasible cash flow by the higher in-house development ratio and other factors.

The same thing can be said for PS3, but a far more number of PS3 will be sold if you compare it with PSP

Link http://www.beyond3d.com/forum/showthread.php?t=20062&highlight=Kutaragi+interview

Ken has already been through this talk. The man has said that the PS3 will be like the PSP model because Sony will have more feasible cash flow due to higher in-house development and other factors. And this is exactly what Phil and some others are saying in this thread right now. I wonder how ML is looking at this exact same issue? It doesn't look like they are putting in weight behind it.

 

[Carl B]

Also the Rambus XDR controller on the CPU is designed to handle a 2 PPE 16 SPE CELL CPU. This may take up more area than the RSX memory controller.

How much room though? Look - ok first do you think that Cell is going to be 3x the cost of RSX? Three times? No.

Secondly, here's the die size of Cell at 90nm: 221mm^2.

Here's the G70 at 110nm: 334mm^2.

Will the RSX, if really a modified G70, be so much smaller than 221mm^2 on 90nm going from 334mm^2 at 110nm? No. And I posit it will likely be larger, but that's neither here nor there. Let's just assume for the moment they're the same size. With the Cell able to take a hit to an SPE and still be useful in PS3 - indeed to take multiple SPE hits and still find use in other possible Sony devices - how in the world are they coming to a three times the cost of RSX estimate?

No, Cell should cost less, or be at least guesstimated at the same cost. Cell will be fabbed at OTSS, Nagasaki, and East Fishkil. RSX only at the former two.

The estimate given in that BOM is madness.

 

[Sis]

This report was talking about the ps3 and its costs . For what it is talking about its extremely accurate . We also have another report from another company where prices mainly line up with each other and the end result is the xbox 360 being cheaper to make

This to me is the key take away and I've yet to see anyone disagree with it. Instead we attack the credibility of the analysts, the posters agreeing with the analysts, or the exact dollar figure of the analysts.

.Sis

 

[Bobbler]

This to me is the key take away and I've yet to see anyone disagree with it. Instead we attack the credibility of the analysts, the posters agreeing with the analysts, or the exact dollar figure of the analysts.

.Sis

What else are people supposed to do? I don't think anyone disagrees with the overall sentiment of the article, but the fine details are dubious at best. You know as well as I do it's the details that B3D likes to pick at. It seems much of this thread has been pretty valid conversation talking about what the actual costs could be and where they could go wrong -- most of the thread has been well manored and doesn't smell of damage control as much as it does trying to get the truth of the matter (because it happens to be an interesting topic).

I don't see the problem in agreeing with the sentiment but questioning the details.

 

[Phil]

Why not?

While you might not have accurate figures certainly you can come up with reasonable estimates for them. It's not like Sony is the only company in the world that has these kinds of expenses, right? It is possible for them to take the larger picture into consideration by comparing that larger picture to similar pictures made by other companies.

Or can you think of some reason why a company that specializes in these types of analysis would be absolutely clueless to the big picture, and wouldn't have any comperable data from any other company on earth?

Alright powderkeg, then please tell me which other company has a PS2 on the market, a PSP, a own fab it's devoting to these two products and using them at the same time to gain in experience and infrastructure advantage to shift significant resources onto a PlayStation successor, which they are again, utilizing different factors to their advantage as redundancy within a chip to improve yields. Heck, name me a company that is bringing out a mass-market-capable device that are striving for synergy among their products and with that will improve yields AGAIN since they can still use CELLs with a larger defection rate for those products?

And if you do happen to find a company that does have very similar resources and targets, where would these analysts get this kind of inside information from? Or are you going to argue that Merrill Lynch has some fabs as well and thus has enough "inside" to magically produce *realistic* cost estimates like they have next?

Merrill Lynch is obviously looking at the situation from a very narrow point of view - as if Microsofts that is manufacturing their console entirely external while Sony is doing a lot inhouse are comparable. Let me ask you, how do you factor in this difference? What about the royalities or the expenses that are Microsoft is facing solely on the fact of producing their product with the help of partners like IBM, ATi & Co while Sony has their own fab? Did they also take into account that Sony invested massively into their own fabs & R&D costs? That these fabs are not only for PS3 but for other products as well? And that because they have their own fabs they also have direct control over the fabbing proccess which can result in strategic cost savings over time as seen with the PS2? And if they've considered all of these in their analysis, what numbers did they use? $$$ x or $$$ y - and why? Based on what? Based on their humble experience? Yeah right, give me break.