Microsoft moves into chip world with Xbox Next
- Thread starter Megadrive1988
- Start date
Not just choice of words, your sentence structure degenerates as well. By taking the "shortcut" approach you force your audience to do more translation and interpretation. Making it easier for yourself, you make it harder for them - a discourtesy to say the least and potentially a blatant sign of disrespect on your part since, as I've pointed out, there are clearly times when you *voluntarily* switch from the little kid talk to a distinctly better grasp of the English language. And its not as if you're struggling with it, chap, because when you choose to use the English language properly, you're able to use various terms, grammar and so forth correctly on multiple occasions.chaphack said:kaching,
oh, you be meaning my choice of words? well it be easier and quick to type/reply informally(as in this topic). long structured posts tired me.
Incidentally, if you had used better grammar in your response above, you would have saved yourself from typing a few characters. For someone who makes it sound like they're replying on the verge of exhaustion, surely that means something
Paul said:
How many times does it have to be said before you unerstand it - MS haven't bought "built" parts, they have bought IP. At present it is up to MS to choose the process in which to build them, not the respective suppliers of the IP.
Is that a difficult concept to understand?
DaveBaumann said:
Well, quite honestly I have trouble comprehending this concept of a highly "plastic" IP that's a structured architecture yet retains the ability to be designed to N process at the whim of the buyer.
Never before have I see such a design, shit... lets end the semantics, it's just not an intelligent decision to forgo ATI's expertise. Why not just have a patent cross-licensing deal or other alternative?
But, what Paul and I have seen (and ATI has compared MS's agreement to) is when ArtX/ATI sold Nintendo "IP" and it then took this completed design (which inheriently must be targeted at a process) and manufactured it elsewhere.
Yet, you keep going back to this fallicious argument of "It's just IP" and then inevitably falls back to "we just don't know at this point" line when questioned. Well if we 'just don't know' then how the heck can you question Paul? Especially where, in the same console market with the same supplier, your wrong based on precedence.
This concept of "IP" is really funny. It's like your saying that Microsoft has just taken a license to ATI's engineers thoughts and ideas and will fully design the part themselves with their own front-end work done internally. Which is... highly doubtful, highly improbably, and ATI has already stated that it's akin to the Nintendo deal - which didn't see this happening.
Because anything they buy that's written down is going to targeted to a specific process at a specific time with a specific implimentation. Unless you can tell me that a libraries for, say, 130nm are equivalent to those for, say, 65nm. Yet, I distress... because I reckone, "we just don't know...." except that Paul's wrong.
Brimstone said:
(a) Why do we assume Microsoft could use MRAM, and STI not? It would appear the level of cooperation favors the inverse.
(b) MRAM? This was brought up why?
(c) Memory Bandwith is not as significant a factor as it once was and will only continue to decline in the future.
(d) MRAM?!? Why.... Where...
cmon guys, transistors != performance
for instance, if intel created a 486 chip with 512mb of embedded ram, it would have a boatload of transistors but still not much performance. or a 486 with 256 cores on one chip, alot of transistors but not much performance. everything doesn't boil down to transistor count.
for instance, if intel created a 486 chip with 512mb of embedded ram, it would have a boatload of transistors but still not much performance. or a 486 with 256 cores on one chip, alot of transistors but not much performance. everything doesn't boil down to transistor count.
randycat99
Veteran
You might be surprised what those 2 examples could bring, especially at clockrates commensurate with current die processes.
nobie said:
Plot a logarithmic graph of preformance against Moore's Law and then tell me that. Especially in the 3D preformance market, which relies on extreme concurrency to achieve preformance advancement.
Just think about it. At most there has been a 10X increase in graphics IC clock since the early 90s (50 -> 500MHz), yet the increase in relative preformance is orders of magnitude when all is said and done. Can you guess where it's coming from?
Take a look at PowerVR and other IP licensing companies Vince - they do it all the time. MBX isn't inhernatly targetted at any single process.
IP does not equal patents Vince.
How do you know they took a design that was targetted at a particular process?
Quite easily. He's set in the opinion that what we've seen from the vendor up until that point is whats going to occur here, but the fact is we just do not know that. Sorry if that offends your sensibilities Vince (which I know is very easy to do), but thats the fact of the matter - IBM themselves have said that what MS does with the IP is entirely up to them at this point, and we still don't know how that part of the equation will play out. AT present the only thing we know is that MS have licensied IP from IBM,ATI and SIS - how that IP is implemented is not known.
No, they have licensed a bunch of adders and ALU's and texture samplers and the knowledge about the circuits required to tie these together to form pipelines, etc., etc.
Once again Vince, I'm not taking the point of view that these things are set in stone - it may well be the case they they will be 130nm or sothing else, but I'm not going around stating these as facts. Why not wait and see what else transpires before these stating these things as fact? Afterall, it was only a few months back when you and other were suggesting there would be not possibaility that MS would ditch NVIDIA for the XBox, when I was saying "wait and see"; I'm sure it threw a number of you for a loop that MS chose IBM instead of a standard x86 vendor. MS have thrown us all a few surprises so far, and I don't necessarily see that they are going to stop now.
Vince said:
I have no idea what Microsoft is going to do. My post wasn't an assumtion, but a possible scenario that would illustrate the "lithography is always king" mantra wrong.
MRAM ending up in the PS3 would be suprising to me. Toshiba has made claims on having a small edram cell size, but no breakthroughs regarding MRAM. If Toshiba has somethng cooking with MRAM, I've missed it. As far as getting MRAM know how from IBM sure they could, but as far as I know, that wasn't apart of the "Cell" deal.
MRAM isn't just about bandwidth. It's very fast and low power. Low power means less heat. MRAM is an ideal memory to be integrated with logic.
I'd guess hardware engineers could to some intresting tricks with a MRAM console. Maybe they could have it as an option when you turn off the console, the last place in the game you were playing would be saved to memory. Then that place would automatically boot up when you turn on the console again. Myself I tend to play the same game a lot in a row, so I'd find this usefull. I hate loading times like I think most people do. I think a lot of average users would like this type of convinence. This type of instant boot up would make the console seem extremely faster than a volatile ram based console to the typical consumer in my opinion.
There is no evidence that Microsoft will use MRAM. I do think Microsoft will at least consider it, and has a shot to be included. MRAM has some really strong advantages.
Because in all likelyhood this is probably what they are doing.
Tell me, which sounds more belivable and logical.
1. That MS has indeed just taken IP and will go about building the IC themselves. From IP to cutting edge IC in two years.
2. IBM and ATI will design and engineer IC based on a specific process, be it 90 130 or whatever. IBM and ATI will thus give the product to MS and MS then goes and mass produces the IC themselves in order to control their own production and cut costs.
We just don't know. So maybe I'm right, maybe I'm wrong and Jesus will come down to Earth tomorrow. Who knows.
It is of my belief(a not totally baseless one) that both the XCPU and XGPU will not be mass produced on the 65 nm node.
Paul said:
I didn't ask what was likely, I asked how Vince knows what he said was the case?
And could there not have been the option that the person that designs the IP works in close collaboration with the people that implemtents that IP to target it at the process the customer requires?
Megadrive1988
Veteran
Playstation4 memory might be MRAM, perhaps? if we are talking about the 2010-2011 timeframe....
Disclaimer: I only know a little bit of chip manufacturing from a friend of mine (who did MPEG chips), so I may not be right in the following paragraphs.
Simplified chip production (I may have missed some major steps)
1. Have a set of algorithm to solve/produce a specific set of results/problems
2. convert that to VHDL (or any other specific chip language, I may not be update on this as I know only a little)
3. Simulation run of the VHDL in software to see if the program works as expected
4. Convert to data for creating a photo mask with the specific fab lithology library
5. Send the data to the fab for pilot run
6. Verify the condition of the chip
7. Refine the program/data for creating mask if needed for improving yield or for solving issues and send it to the fab
8. Repeat until the chip produced is acceptable for release
I think the current licensing may be for step 1 or 2. MS can outsource the other steps to IBM as IBM should be very experienced in the process.
MS can select whatever process that is feasible in terms of budget and deadline.
For sure I might as well be completely wrong with the above, as I am not at all interested during the time my friend was a chip designer. Those are from my faint memory on the chit chat with him.
Simplified chip production (I may have missed some major steps)
1. Have a set of algorithm to solve/produce a specific set of results/problems
2. convert that to VHDL (or any other specific chip language, I may not be update on this as I know only a little)
3. Simulation run of the VHDL in software to see if the program works as expected
4. Convert to data for creating a photo mask with the specific fab lithology library
5. Send the data to the fab for pilot run
6. Verify the condition of the chip
7. Refine the program/data for creating mask if needed for improving yield or for solving issues and send it to the fab
8. Repeat until the chip produced is acceptable for release
I think the current licensing may be for step 1 or 2. MS can outsource the other steps to IBM as IBM should be very experienced in the process.
MS can select whatever process that is feasible in terms of budget and deadline.
For sure I might as well be completely wrong with the above, as I am not at all interested during the time my friend was a chip designer. Those are from my faint memory on the chit chat with him.
WRT to what occured with Flipper, I decided to ask Eric Demers (sireric) who is around this forum quite a lot to see if he knew anything about Flipper, to which he replied:
"Yes. I Architected & designed a big chunk."
Feeling he may be fairly qualified to answer some questions on how it was implemented I asked if they sold something targetted at a particulr process, to which his responce was
"It's a pretty secret project, so I can't really tell you much. However, ATI and NEC worked closely on this project."
That didn't go very far, but it certianly seems to say that people aren't going to know exactly what happened with the IP and how it was sold - it also doesn't suggest that something was just handed over and that was implemented. So, trying a different tack I asked if IP has to be targetted at a different process should some be sold to a prospective console vendor.
"As a generalized comment, if one were to buy IP in the form of RTL code, then one could port it from technology to technology. If one were to buy netlists, then if one is porting from one tech to another, the tech have to be similar, otherwise the design will not meet timing."
I assume not many of you have more expericence with this sort of stuff than an ARTX/ATI engineer that designed part of Flipper and is probably working on either one of the next gen console parts?
"Yes. I Architected & designed a big chunk."
Feeling he may be fairly qualified to answer some questions on how it was implemented I asked if they sold something targetted at a particulr process, to which his responce was
"It's a pretty secret project, so I can't really tell you much. However, ATI and NEC worked closely on this project."
That didn't go very far, but it certianly seems to say that people aren't going to know exactly what happened with the IP and how it was sold - it also doesn't suggest that something was just handed over and that was implemented. So, trying a different tack I asked if IP has to be targetted at a different process should some be sold to a prospective console vendor.
"As a generalized comment, if one were to buy IP in the form of RTL code, then one could port it from technology to technology. If one were to buy netlists, then if one is porting from one tech to another, the tech have to be similar, otherwise the design will not meet timing."
I assume not many of you have more expericence with this sort of stuff than an ARTX/ATI engineer that designed part of Flipper and is probably working on either one of the next gen console parts?