Book: The Race for a New Game Machine
- Thread starter AzBat
- Start date
Search some of the older threads as there are a number of examples noted here and there (surprisngly large latencies being a common theme).
I'm on page 84 so far in this book. I have to say its pretty bad and the guy seems to think that home consoles died in the late 70s and it wasn't until the 90s when sony brought the market back with the playstation brand. Reading stuff like that makes me fearfull of the rest of the book.
Thanks for being the guinea pig & the update. That doesn't sound so good, but hopefully the rest is much better.
Tommy McClain
I should finish the book tommorow
page 81
"After we hung up, I leaned back in my chair, rested my head, and closed my eyes, trying to divine the underlying meaning behind Ha'ls words. Something was fishy. I slowly rolled my chair back and forth. His pronouncement wasn't news to me. I had seen the studies. Howeer , right before exiting the high-level design phase was no time to make such a major change. hardware multithreading was a technique that gave the illusion to a programmer that there were actually two powerpc cores instead of just one. Dual-issue allowed two instructions to be issued and executed in parellel. Both were excellent ways to bump up performance of a microprocessor, but our design was already locked down."
This is apparently late 2001 after getting a phone call to take out dual issue and putting in this instead. They said dual issue was 5 to 10% performance gain but multithreading was twenty to thirty percent. Apparently they extended the time frame 2 months and made the swap. now i'm reading about him going to a 24 hour fitness club and eating fast food mexician.
page 81
"After we hung up, I leaned back in my chair, rested my head, and closed my eyes, trying to divine the underlying meaning behind Ha'ls words. Something was fishy. I slowly rolled my chair back and forth. His pronouncement wasn't news to me. I had seen the studies. Howeer , right before exiting the high-level design phase was no time to make such a major change. hardware multithreading was a technique that gave the illusion to a programmer that there were actually two powerpc cores instead of just one. Dual-issue allowed two instructions to be issued and executed in parellel. Both were excellent ways to bump up performance of a microprocessor, but our design was already locked down."
This is apparently late 2001 after getting a phone call to take out dual issue and putting in this instead. They said dual issue was 5 to 10% performance gain but multithreading was twenty to thirty percent. Apparently they extended the time frame 2 months and made the swap. now i'm reading about him going to a 24 hour fitness club and eating fast food mexician.
For some other context with regards to AMD,IBM and Intel at the time:
In my hazy memory of the time period, we should note that AMD was in a different spot than it is now, but that IBM Micro was in a suspiciously similar situation to where it is now: an obviously well-funded production division with little commercial success stemming from its hardware desperate to have something for its engineering and fab resources to do, if only to justify its existence to the parts of IBM that actually made money.
The signs of IBM's woes were already present at the time. The PPC 970 wasn't setting the world on fire. In SpecInt, it lagged Opteron per-clock, and it was for the most part relegated to a niche market supplying Apple. Its power numbers and clock speeds could not match the higher volume x86 competition, and were failing to impress for some time prior.
The high-end POWER chips often pulled impressive numbers, but much of it could be attributed to the stupendously massive system infrastructure put around them. A PPC core without this massive investment and massive service revenue was what Apple got prior to the switch to Intel.
The spectacle of IBM doing a PR dance with Apple and in effect being jerked around by Steve Jobs (who, for all his influence in other fields, was a bit player in the hardware field most relevant to this discussion), was a sign that IBM Microelectronics was not negotiating from a position of strength.
AMD at the time was making some serious cash and would for a time be capacity constrained trying to supply high-performing and high-margin processors.
Confusing legalities concerning x86 IP aside, any deal with MS would have been with a more dominant company.
IBM's rather generous terms and initial contribution of fab work for its console partners wouldn't have happened if it were in a position like AMD or Intel's.
IBM's struggle to maintain its relevance in semiconductor manufacturing made a lot of this possible.
Will it happen again next gen?
IBM Micro is probably more of a money sink now than it was then, and it has no real presence in the client field taken up by AMD, Intel, and Via.
It's expanded its foundry services, and will be in competition with AMD's foundry company.
IBM needs the service revenue for its products, but the speculation of just how long it will be expected to foot the bill for the semiconductor treadmill persists.
(Neat rumor is that IBM benefited a bit from the circuit design techniques it used both MS and Sony's console parts as guinea pigs for.)
From that standpoint, IBM Micro looks as desperate or worse for some kind of console-related "makework" and now has a history.
AMD also appears desperate, but its lack of resources is obvious, and survival is an open question.
Perhaps Nvidia might make a nice resurgence in consoles, if Intel doesn't get more flexible.
Nvidia is in a rougher patch and might be willing to bargain. Thanks to AMD, ATI's remnants might not exist at that point either.
In my hazy memory of the time period, we should note that AMD was in a different spot than it is now, but that IBM Micro was in a suspiciously similar situation to where it is now: an obviously well-funded production division with little commercial success stemming from its hardware desperate to have something for its engineering and fab resources to do, if only to justify its existence to the parts of IBM that actually made money.
The signs of IBM's woes were already present at the time. The PPC 970 wasn't setting the world on fire. In SpecInt, it lagged Opteron per-clock, and it was for the most part relegated to a niche market supplying Apple. Its power numbers and clock speeds could not match the higher volume x86 competition, and were failing to impress for some time prior.
The high-end POWER chips often pulled impressive numbers, but much of it could be attributed to the stupendously massive system infrastructure put around them. A PPC core without this massive investment and massive service revenue was what Apple got prior to the switch to Intel.
The spectacle of IBM doing a PR dance with Apple and in effect being jerked around by Steve Jobs (who, for all his influence in other fields, was a bit player in the hardware field most relevant to this discussion), was a sign that IBM Microelectronics was not negotiating from a position of strength.
AMD at the time was making some serious cash and would for a time be capacity constrained trying to supply high-performing and high-margin processors.
Confusing legalities concerning x86 IP aside, any deal with MS would have been with a more dominant company.
IBM's rather generous terms and initial contribution of fab work for its console partners wouldn't have happened if it were in a position like AMD or Intel's.
IBM's struggle to maintain its relevance in semiconductor manufacturing made a lot of this possible.
Will it happen again next gen?
IBM Micro is probably more of a money sink now than it was then, and it has no real presence in the client field taken up by AMD, Intel, and Via.
It's expanded its foundry services, and will be in competition with AMD's foundry company.
IBM needs the service revenue for its products, but the speculation of just how long it will be expected to foot the bill for the semiconductor treadmill persists.
(Neat rumor is that IBM benefited a bit from the circuit design techniques it used both MS and Sony's console parts as guinea pigs for.)
From that standpoint, IBM Micro looks as desperate or worse for some kind of console-related "makework" and now has a history.
AMD also appears desperate, but its lack of resources is obvious, and survival is an open question.
Perhaps Nvidia might make a nice resurgence in consoles, if Intel doesn't get more flexible.
Nvidia is in a rougher patch and might be willing to bargain. Thanks to AMD, ATI's remnants might not exist at that point either.
tongue_of_colicab
Veteran
A bit off topic, but why is everything so doom at AMD anyway? ATI isnt doing that bad right? and I know AMD wont be your first choice if you want pure power but for a little over the price of Intels cheap c2q you can get a low power x2, mainboard with everything on it including hdmi and fast enough to render 1080p and 2gb ram. Ofcourse not as fast but does 90% of the pc using people at home or at the office really needs anything faster? I doubt you'll notice the difference in performance on normal apps. Ofcourse, I always wonder why all the big contractors wont sell AMD, certainly not in large quantities. Something fisy going at at Intel for sure I think, but if they can prove it?
I certainly think amd could be, and should be in alot better condition than they are.
I certainly think amd could be, and should be in alot better condition than they are.
There's an AMD doom and gloom thread that goes over the myriad of woes AMD is stuck with.
There is a mix of historical reasons, Intel competition and its pressure on OEMs and the channel, and long-standing trends in semiconductor design that have only become more unforgiving with each process node.
This is one example of the somewhat unique state IBM's Microelectronics division was at the time of the PS3 and 360 design phases.
Console chips are a low-margin business, one that neither AMD nor Intel at the time would have been willing to devote too much in the way of resources to design for while also cutting prices to fit a console maker's desires.
Intel and AMD had far more valuable high-volume products to devote their engineers and fab capacity towards.
IBM Micro's POWER chips are comparatively niche, but semiconductor manufacturing and R&D isn't cheap, and idling fabs and engineers are expensive.
At least in part, IBM Micro's nicer deals with the PS3 and Xenon were helped along by the fact that the initial fab space and engineering resources were better spent on low-margin deals instead of nothing.
AMD and Intel did not have that problem at the time.
IBM the software service company nowadays still needs to justify the expense of hardware manufacturing. AMD is actually in a similar situation with regards to the process treadmill, but it is in a far more precarious state.
The question is whether AMD might implode before IBM decides it doesn't want to bother with hardware that does not make it money.
There is a mix of historical reasons, Intel competition and its pressure on OEMs and the channel, and long-standing trends in semiconductor design that have only become more unforgiving with each process node.
This is one example of the somewhat unique state IBM's Microelectronics division was at the time of the PS3 and 360 design phases.
Console chips are a low-margin business, one that neither AMD nor Intel at the time would have been willing to devote too much in the way of resources to design for while also cutting prices to fit a console maker's desires.
Intel and AMD had far more valuable high-volume products to devote their engineers and fab capacity towards.
IBM Micro's POWER chips are comparatively niche, but semiconductor manufacturing and R&D isn't cheap, and idling fabs and engineers are expensive.
At least in part, IBM Micro's nicer deals with the PS3 and Xenon were helped along by the fact that the initial fab space and engineering resources were better spent on low-margin deals instead of nothing.
AMD and Intel did not have that problem at the time.
IBM the software service company nowadays still needs to justify the expense of hardware manufacturing. AMD is actually in a similar situation with regards to the process treadmill, but it is in a far more precarious state.
The question is whether AMD might implode before IBM decides it doesn't want to bother with hardware that does not make it money.
Worth mentioning though that Sony's relationship with IBM was different than MS' or Nintendo's as it relates to the above; remember that Sony actually expended directly part of the capex required for East Fishkill modernization, in return purchasing output rights. Beyond the ~$400 million for Cell R&D that the three split, Sony put hundreds of millions into East Fishkill, and billions total into fab capacity when their own Nagasaki 2 line and Toshiba JV are included. SCE/Sony Semi under Kutaragi was a much more active/proactive animal than it was after the fact.
************************************
Note concerning last several posts: This thread isn't a sales thread, so... no more sales talk, comparative or otherwise.
Last edited by a moderator:
Interesting thing of note, now that we've covered IBM, Toshiba, Sony, AMD is that this foursome and Chartered were the primary members of the process alliance at 90nm.
Their fortunes since that point are rather sobering.
ITSA (minus Chartered) posted process numbers for 65nm in 2005, and we've only recently started to track down consoles in the channel that have 65nm chips.
Of the group, perhaps we can count Sony's selling facilities to Toshiba as a positive for Toshiba.
Outside of that, the good news has been in short supply for the rest.
I think for the 65nm Cell, volume production began early last year, but there was of course a lot of 90nm inventory to work through. 65nm RSX on CMOS seemed like it took a little longer to ramp, but that's entirely the realm of Toshiba now at this point essentially.
I agree that the IBM alliance had issues with 65nm SOI in general, compared to their initial target dates; hopefully 45nm will be kinder.
Well I honestly (unfortunately) think it was a poorly timed deal for Toshiba, because I think they were thinking of capacity utilization beyond just the Sony/Toshiba in-house projects, and now the entire semi space seems awash in capacity. I think they went from too much memory and flash capacity to adding too much complex IC capacity to the list as well.
I finished the book. Took an hour in total.
Don't buy it , its a waste of $23 bucks
The most interesting things is that MS wanted 4 cores but later reduced it to 3 when they found out that weren't going to be able ot make it profitable.
IBM decided to go with MS because sony added another 2 spu units up from the original 1/6 design to 1/8 design and that would cut into the profit margins for ibm who was supposed to produce the chip at their fab.
THe original dual issue that IBM moved out for multithreading was put back in by ms and since they used a common core for everything but vmx units (vector units )
A sony engineer who worked on the original vmx units fixed a bug on the ms verison of the unit which was enhanced over the sony one. The bug didn't affect the sony one. he still fixed it.
OOE was removed because IBM couldn't get all 3 chips (apple , ms , sony) done in the time frame and keep OOE. That pissed off Apple and seems to have set them on the road to Intel
Thats really all the technical stuff. Now if you want to hear about this guys few stressfull years making these chips its a good read. But not so much if you want to learn about the chips .
Don't buy it , its a waste of $23 bucks
The most interesting things is that MS wanted 4 cores but later reduced it to 3 when they found out that weren't going to be able ot make it profitable.
IBM decided to go with MS because sony added another 2 spu units up from the original 1/6 design to 1/8 design and that would cut into the profit margins for ibm who was supposed to produce the chip at their fab.
THe original dual issue that IBM moved out for multithreading was put back in by ms and since they used a common core for everything but vmx units (vector units )
A sony engineer who worked on the original vmx units fixed a bug on the ms verison of the unit which was enhanced over the sony one. The bug didn't affect the sony one. he still fixed it.
OOE was removed because IBM couldn't get all 3 chips (apple , ms , sony) done in the time frame and keep OOE. That pissed off Apple and seems to have set them on the road to Intel
Thats really all the technical stuff. Now if you want to hear about this guys few stressfull years making these chips its a good read. But not so much if you want to learn about the chips .
Thanks for the quick review. Looks like it wasn't a total loss. Interesting tidbits, but wished there was more to it. Oh well.
Tommy McClain
you want me to send you out the book ?
Thanks for the synopsis Eastmen, seems as expected. Still any new information towards the knowledge base is always welcome. That 6/8 SPE thing reminds me of when Kutaragi found out about that change and had them redesign around 8 due to the aesthetics (awesomeness). 
Anyway for any of you interested in the origins of these chips who have never read through One's "Engineers who created Cell" translations, it's a great thread IMO, and gives a lot of history/color that you'd by all rights never expect to see from the Western press.
http://forum.beyond3d.com/showthread.php?t=20563
Anyway for any of you interested in the origins of these chips who have never read through One's "Engineers who created Cell" translations, it's a great thread IMO, and gives a lot of history/color that you'd by all rights never expect to see from the Western press.
http://forum.beyond3d.com/showthread.php?t=20563
Thanks for the offer, but I'm sure there are others here that would want it more than myself. Plus, I can't remember the last time I read a book. LOL
Tommy McClain
That is true. When I wrote that, I was thinking at the time that it could at least be spun that Toshiba's manufacturing business was arguably growing, and that perhaps it got a decent deal on the purchase since Sony needed to offload the facilities. Given the current market downturn and the rapid depreciation of fabs, it probably isn't so much of a bargain.
