A Summary of the Huge Wii Thread

[The_legend_of_drtre]

Wasn't the 970 a much wider and deeper (pipeline) design than the 750?

Yeah that's why I said basically lol. It looks likes it's a dramatically improved 750 cl because the die size is bigger than a standard 750cl, yet not quite as powerful as a 970.

In between a 750CL and 970 in performance is what I meant to say.

 

[Teasy]

How would it be transparently compatible with Gamecube then, especially at such a low clockspeed? I think the CPU is almost certainly nearly identical to the Cube CPU. They shrunk it, and then went as high in clock as such an old architecture could comfortably go.

If by nearly identical you mean only around 5-10 million transistors added to the chip when compared to Gekko (which was about 21 million transistors) then I'd agree. Not on clock speed though as the 750CL can run at a clock speed of 1Ghz.

it's natural that PAL regeons would have PAL releases of PAL games (50hz and all), anyway. do you expect all of those people to have to buy a new TV just to play snes games on the Wii VC? and yes, i'm aware that some TVs sold in pal territories accept 60hz or NTSC signals, but it doesn't mean they all do

Its not some to be honest, its about 95% (and that's conservative IMO), its been 10 years since they produced PAL50 only TV's, in the UK at least. When I went to the isle of mull for a week a few years back there was a single old 17" CRT TV in the house and even that ran PAL60

 

[DuckThor Evil]

Its not some to be honest, its about 95% (and that's convervative IMO), its been 10 years since they produced PAL50 only TV's, in the UK at least. When I went to the isle of mull for a week a few years back there was a single old 17" CRT TV in the house and even that ran PAL60

True and it makes the VC such a disapointment in Europe, many of those PAL conversions are just horrible, why is VC area coded for crying out loud!

 

[rounin]

plasma static image burn-in is also a problem depending on your usage. Game time with static healthbars etc would have to be carefully managed.

I agree with your wii port sentiment and we should start seeing much more impressive titles showing up on wii in a year or two. Problem will be how these ported titles look next to the originals as I'm sure ps2 gen is also holding next-gen back somewhat also.

Has anyone discussed the possibility that as a result of the wii, the PS2 will last longer in terms of games ported to and from the 2 systems ?

 

[sfried]

I've played a few Cube games: Rebel Strike, RE4, and XG3 offhand. They played absolutely the same as on Cube. Rebel Strike still had all its slowdowns, too. Some have said the Wii will run the games better, but I've definitely not seen that at all. If it does, it's in a way that's almost not worth mentioning.

Have you tried timed based games like Ikaruga?

 

[TheChefO]

Has anyone discussed the possibility that as a result of the wii, the PS2 will last longer in terms of games ported to and from the 2 systems ?

Interesting theory. As long as people are buying then people will be selling.

Personally I think sftware sales for ps2 will drop like a rock this year.

We'll see though - good theory.

 

[Fox5]

If by nearly identical you mean only around 5-10 million transistors added to the chip when compared to Gekko (which was about 21 million transistors) then I'd agree. Not on clock speed though as the 750CL can run at a clock speed of 1Ghz.

Doesn't Boardway have more cache, which would account for the increase in transistors?

 

[Capeta]

Gekko had no SIMD unit, either. However, I believe it was capable of splitting the FP registers to operate on 2 SP numbers at the same time, sort of a "half" vector-unit.

Gekko does have a SIMD unit. It could operate in 2-way or 4-way modes.

 

[MDX]

It is based on a 750 CL. I'm the one who found the pdf file for it and posted it in a whole bunch of forums.

On IGN they said it was based off a 750 CL then 5 minutes later they took that part of the article out. So I got curious and I searched IBM.com for any info I could find on it and I found the detailed PDF file on the chip. It is the same exact numbers for the CPU as max console posted.

If you think the Max console specs are correct, then it is definately a 750 cl.

So I was reading this old article:

IBM delivers Wii chip 10:52AM, Monday 11th September 2006
IBM has delivered the first batch of processors for the forthcoming Wii games console from Nintendo. The chips are Nintendo customised versions of the PowerPC chips that powered the previous generation of Macintosh. "Broadway" will be designed on a 90-nm SOI process, shipping from IBM's East Fishkill, N.Y. facility.

This made me start to think that Broadway could still be based on the 970xx since that would be
Apple's last gen, and the article could have easily have said version of Gekko:

PowerPC G3 (IBM- 750- Gekko)
PowerPC G4 (Motorola)
PowerPC G5 (IBM- 970- Wii?)- Discontinued August 2006
Intel x86 (Intel) - Introduced January 2006

But trying to match Broadway to either a 750xx or 970xx wasnt going anywhere. So I started to wonder, what if Nintendo truly this time around didnt base Broadway on any existing IBM microprocessors like they did with Gekko (to save some cash), but built it around an IBM core like the 360? Because Broadway IMO doesnt neatly fit either 750xx nor the 970xx.

When the 750Cl and 970GX was announced,

Specifically, IBM introduced two new off-the-shelf standard PowerPC single-core processors, both available immediately. The PowerPC 750CL, a 32-bit microprocessor, consumes half the energy as its predecessor, and performs at speeds ranging from 400MHz to 1GHz. The 750CL includes a 256KB L2 cache, and is targeted at networking, storage, imaging, consumer electronic and other high-performance embedded applications.

The PowerPC 970GX, a follow-on to the PowerPC 970FX, supports both 32-bit and 64-bit operations. It features the same power capabilities as its predecessor, but incorporates twice the integrated L2 cache at 1MB. The range of frequencies for the 970GX is 1.2 to 2.5GHz, enabling the chip to support high-bandwidth data processing and algorithmic intensive computations, making it suitable for communications, storage, multimedia and graphics based devices.

So was this:

IBM also announced three new 32-bit processor cores, including:
The 460S synthesizable core, which will allow designers to select the size L1 and L2 cache sizes and processor local bus (PLB) version necessary for implementing a single or cache coherent multi-processor design, and can be manufactured in any fabricator worldwide.

The 464FP H90, which is similar to the 464 H90 hard core, but with an integrated double precision floating point unit. Both application specific integrated circuit (ASIC) cores allow customers to more easily customize a chip design and have it manufactured with IBM or at Common Platform manufacturing facilities at Chartered Semiconductor Manufacturing and Samsung Electronics Co, Ltd.

Both PowerPC processors and the two ASIC hard cores are manufactured using IBM's 90 nanometer (nm) copper processing technology. The microprocessors also feature silicon-on-insulator (SOI) tech....

could Broadway be built from something like this?

IBM’s newest core offerings are based upon the PowerPC 460 embedded core technology and capable of supporting cache coherent multiprocessor designs. The PowerPC 460 core is a 32-bit high performance, low-power embedded processor core which is fully compliant with the flexible and scalable Power ISA™ 2.03. Offered in both synthesizable and hard core forms the new processor core offerings provide higher performance and greater design flexibility to meet specific SoC needs.

http://www-03.ibm.com/chips/power/powerpc/cores/ppc460.html

Your thoughts...

 

[Capeta]

Whichever processor has the best backwards compatibility is the one Nintendo went with IMO.

 

[fearsomepirate]

But trying to match Broadway to either a 750xx or 970xx wasnt going anywhere...

Actually, it was, it has, and the speculation's pretty much over here. Broadway is the right size to be based on the 750. It may have more cache, extra functions, better OoE, additional FP registers, more redundancy, or something along those lines, but it is most certainly 750-based.

Gekko supports some SIMD instructions but does not have a dedicated SIMD unit like Altivec.

 

[compres]

Then how does it work?

The relation between area and linear dimensions are like this:

area = linear squared
or
A = L^2

Transstor lenght in process changes tend to change with a .707 proportion aproximatly, which is 1/(2^.5), a common value on electrical engineering but that is not related at all with this matter we are discussing here.

This proportion has been used becouse it allows EEs to pack twice as much transistors in the same area each time the manufactorin process changes.

since
A1 = L1^2 and A2 = L2^2
L2 = .707*L1
the new area is
A2 = .5*L1^2
which is half of the old area given the same chip.

 

[Capeta]

The relation between area and linear dimensions are like this:

area = linear squared
or
A = L^2

Transstor lenght in process changes tend to change with a .707 proportion aproximatly, which is 1/(2^.5), a common value on electrical engineering but that is not related at all with this matter we are discussing here.

This proportion has been used becouse it allows EEs to pack twice as much transistors in the same area each time the manufactorin process changes.

since
A1 = L1^2 and A2 = L2^2
L2 = .707*L1
the new area is
A2 = .5*L1^2
which is half of the old area given the same chip.

Great explanation so at 90nm the same chip would be half the size of a 130nm chip and 1/4 the size of a 180nm chip?

 

[crystall]

The 750CL has no SIMD unit and does not support Alitivec/Vector operations making it a poor console CPU. It would make the Wii CPU weaker than it is already rumored to be. At the very least they would have added some sort of vector unit to it. It has to be a modified 750CL if anything.

The 750CL has a SIMD unit, does not support AltiVec (just like the Gekko) and is just a shrink of the Gekko for general purpose use. Now the chances that what's in the Wii is a 750CL are very high, here's an hint:

http://www-306.ibm.com/chips/techli...1D10065F7D5/$file/ppc750cl_ds_dd20_5oct06.pdf

Look carefully at the power consumption table on page 17, the Power column contains data for every frequency except for the 733/800 ones, an area curiously close to what we found in the Wii. Coincidence? Also look on page 18, maximum SYSCLK (i.e. the 60x bus clock) is 240 MHz, 40 more than the higher clocked 750GX maximum, why? Maybe because it's just 3 shy of Broadway's 243 MHz 60x bus? Coincidence? I think not. Anyway my guess is that Broadway is just a slightly customized 750CL.

 

[DuckThor Evil]

Great explanation so at 90nm the same chip would be half the size of a 130nm chip and 1/4 the size of a 180nm chip?

Pretty close. a 90nm chip is about 48% of a 130nm chip and exactly 25% of a 180nm chip.

 

[fearsomepirate]

The 750CL has a SIMD unit

The document you linked nowhere mentions a SIMD unit or any other kind of additional vector unit. The floating-point unit simply supports paired single-precision arithmetic, which is exactly what I've been saying.

 

[Capeta]

Pretty close. a 90nm chip is about 48% of a 130nm chip and exactly 25% of a 180nm chip.

Ok if that's the case then Hollywood must have a lot more logic than Flipper since it's more than half the size at 90nm vs 180nm.

 

[DuckThor Evil]

Ok if that's the case then Hollywood must have a lot more logic than Flipper since it's more than half the size at 90nm vs 180nm.

That is true, although I should have also said that in real life the chips won't shrink exactly like that, since there are other variables as well, but I'm guessing that the difference is fairly small.

 

[compres]

Ok if that's the case then Hollywood must have a lot more logic than Flipper since it's more than half the size at 90nm vs 180nm.

Yes, as evil said, these manufacturing processes are far too complex; that said we can be sure the margin of error is small enough to be irrelevant in this discussion.

I'll take the estimates from the first post:

106mm^2@180nm would be (106/4)mm^2@90nm which is:

26.5mm^2

and we have that the Hollywood has 72 mm^2, that is

(72mm^2)/(26.5mm^2) ~= 2.717

or in layman terms close to 3 times as big. Basically it should have at least about +150% more transistors.

Does it make sense to call it an over clocked flipper really?

PS: On Stefan's post the estimate is 110mm^2, the fact of the matter is that this chip has a lot more transistors than flipper.

 

[crystall]

The document you linked nowhere mentions a SIMD unit or any other kind of additional vector unit. The floating-point unit simply supports paired single-precision arithmetic, which is exactly what I've been saying.

Paired single-precision arithmetic = 2/4 floating point operations per instruction on couples of single-precision floats held in the same registers. In my book this is called single-instruction multiple-data: SIMD.