Revolution's Broadway CPU to be G3+VMX variant?
- Thread starter PC-Engine
- Start date
The G4 has 2 families named PowerPC 7400 and PowerPC 7450.
http://arstechnica.com/articles/paedia/cpu/ppc-2.ars/3
http://arstechnica.com/articles/paedia/cpu/ppc-2.ars/2
http://arstechnica.com/articles/paedia/cpu/ppc-2.ars/3
http://arstechnica.com/articles/paedia/cpu/ppc-2.ars/2
Altivec was licensed by IBM before the GameCube even came out. It is part of the IBM/Freescale "Book E" spec, so any Book E compliant IBM proc implements Altivec as far as i remember
The clock speeds on Flipper (and ultimately on Gecko, thanks to multiplier dependency) were changeg because NEC hit a 165MHz ceiling with their NED3 process on which Flipper was manufactured
The clock speeds on Flipper (and ultimately on Gecko, thanks to multiplier dependency) were changeg because NEC hit a 165MHz ceiling with their NED3 process on which Flipper was manufactured
Shogmaster
Regular
emerge said:
Interesting! Thanks for the info.
Oh ho! We were right, faffy!
There were only two possible explanations for the change in clock speed and both were equally plausible AFAICS (it could even have been a bit of both). It just came down to which one you wanted to believe at the time. So I wouldn't get too carried away Shogmaster
Last edited by a moderator:
Oh dear...
Nope... At least not in comparison to it's two competitors... I also don't think you'd see a 750 core hit 1.6GHz too easily to be cost effective...
By dual core 970FX you mean the 970MP? It would pretty handily slaughter a 750...
Well that, and the VX doesn't even exist...
Mythology...
I wouldn't necessarily call the SH4's implementation a SIMD implementation at all...
This is incorrect...
For starters, AltiVec and VMX are the same thing... AltiVec is simply the name trademarked by Motorola/Freescale. (that and most people who've programmed for it still call it AltiVec 'cause it's a cooler name and has been around much longer). Secondly, IBM hasn't made any "G4 compatibles" so to speak. They did for a brief time fab 7400s for Apple and Moto however.
Sigh... Now we're getting bad here folks...
AltiVec/VMX/Velocity Engine (or any other name for the given SIMD extension), is not a part of PowerPC Book E, PowerPC or PowerPC AS. AltiVec has never been licensed by IBM as IBM participated in the development of it.
You mean 7400/7410. And no the 7400 is not *just* as 750+AltiVec. While having common heritage with the 750, the 750 lacked the MPX bus support, only supported MEI (instead of MESI the 7400 supported) coherency, had fewer rename resources and smaller completion queues. The FPU of the 7400 also had a lower latency multiplier for double-precision. And of course once you get to the 745x/744x models, you're talking about a significantly changed design from the 750...
Die size would be of minimal concern really (especially with the differences between 750 and 7400 dimensionally being rather small).
And I doubt Flipper would care whether the CPU's SIMD implementation is 2-way or 4-way...
Less balanced? You can never have enough CPU..!
I doubt that as well... By the time the GCN launched Moto would've been able to provide plenty of 7410s to Nintendo (they actually for a long time had way more fab capacity than IBM)...
Anyways, the ideal choice would be a 1.6GHz (or faster) 970MP... If you absolutely *had* to use something lesser, than one of the newer e600 dual-core jobbies like Freescales MPC8641D would be WAAAAY more appealing than some hypthetical 750+AltiVec...
Nope... At least not in comparison to it's two competitors... I also don't think you'd see a 750 core hit 1.6GHz too easily to be cost effective...
By dual core 970FX you mean the 970MP? It would pretty handily slaughter a 750...
Well that, and the VX doesn't even exist...
Mythology...
I wouldn't necessarily call the SH4's implementation a SIMD implementation at all...
This is incorrect...
For starters, AltiVec and VMX are the same thing... AltiVec is simply the name trademarked by Motorola/Freescale. (that and most people who've programmed for it still call it AltiVec 'cause it's a cooler name and has been around much longer). Secondly, IBM hasn't made any "G4 compatibles" so to speak. They did for a brief time fab 7400s for Apple and Moto however.
Sigh... Now we're getting bad here folks...
AltiVec/VMX/Velocity Engine (or any other name for the given SIMD extension), is not a part of PowerPC Book E, PowerPC or PowerPC AS. AltiVec has never been licensed by IBM as IBM participated in the development of it.
You mean 7400/7410. And no the 7400 is not *just* as 750+AltiVec. While having common heritage with the 750, the 750 lacked the MPX bus support, only supported MEI (instead of MESI the 7400 supported) coherency, had fewer rename resources and smaller completion queues. The FPU of the 7400 also had a lower latency multiplier for double-precision. And of course once you get to the 745x/744x models, you're talking about a significantly changed design from the 750...
Die size would be of minimal concern really (especially with the differences between 750 and 7400 dimensionally being rather small).
And I doubt Flipper would care whether the CPU's SIMD implementation is 2-way or 4-way...
Less balanced? You can never have enough CPU..!
I doubt that as well... By the time the GCN launched Moto would've been able to provide plenty of 7410s to Nintendo (they actually for a long time had way more fab capacity than IBM)...
Anyways, the ideal choice would be a 1.6GHz (or faster) 970MP... If you absolutely *had* to use something lesser, than one of the newer e600 dual-core jobbies like Freescales MPC8641D would be WAAAAY more appealing than some hypthetical 750+AltiVec...
You are forgotten that an huge part in research is the transistor research for to take better clock speeds. I am sure that a lot of the money in the Broadway is for hitting speeds that a normal variation of the chip couldn´t use.
I am with Shogmaster when he says that we must see the heat consumption of the CPU but I cannot believe that a souped up Gekko+VMX will be the CPU of Revolution.
Why not?
Because an alternative exist.
The alternative is the 1T-SRAM, imagine a PowerPC 970FX with 1MB L2, with it you have 82 milion transistors (The actual 970FX has 512K L2 and 58 milion transistors) if you use the the 1T-SRAM for the L2 cache the size of the chip goes from the 82 milion transistors to the 36 milion transistors, an huge difference and the only that you need is an internal memory controller for it and I don´t believe that a memory controller will use 48 milion transistors and with all this we can hit a better speed and having multithreading in the CPU using less transistors than the normal versión of the Microprocessor.
I am with Shogmaster when he says that we must see the heat consumption of the CPU but I cannot believe that a souped up Gekko+VMX will be the CPU of Revolution.
Why not?
Because an alternative exist.
The alternative is the 1T-SRAM, imagine a PowerPC 970FX with 1MB L2, with it you have 82 milion transistors (The actual 970FX has 512K L2 and 58 milion transistors) if you use the the 1T-SRAM for the L2 cache the size of the chip goes from the 82 milion transistors to the 36 milion transistors, an huge difference and the only that you need is an internal memory controller for it and I don´t believe that a memory controller will use 48 milion transistors and with all this we can hit a better speed and having multithreading in the CPU using less transistors than the normal versión of the Microprocessor.
Ooh-videogames
Regular
Thats what making the discussion interesting, there's so many variables that have yet to be considered. The transistor count of 1T-SRAM is something I hadn't thought about.
Cannot be competitive?
I don´t believe all the numbers that Microsoft and Sony are giving since their CPU lacks a lot of prediction and they aren´t OoOE CPU this two things are a hit to the performance of both microprocessors.
How many people that is creating rumours in the net knows how is the Nintendo policy?
From what I have seen few people.
Nintendo never created a SuperCPU, they ever use variants of a low cost CPU, the NES uses a variant of the 6502, the SNES uses a 16 bits versión of the 6502 running at 3.58Mhz, the N64 version of the R4000 was more simpler that the R3000 version used on the Playstation and the Gekko is not more than an enhaced PowerPC 750FX.
This is why some of us are speculating with existing CPU.
PD: Replace the PowerPC 970FX logic without cache for a Dual Gekko with multithreading in my theory.
I don´t believe all the numbers that Microsoft and Sony are giving since their CPU lacks a lot of prediction and they aren´t OoOE CPU this two things are a hit to the performance of both microprocessors.
How many people that is creating rumours in the net knows how is the Nintendo policy?
From what I have seen few people.
Nintendo never created a SuperCPU, they ever use variants of a low cost CPU, the NES uses a variant of the 6502, the SNES uses a 16 bits versión of the 6502 running at 3.58Mhz, the N64 version of the R4000 was more simpler that the R3000 version used on the Playstation and the Gekko is not more than an enhaced PowerPC 750FX.
This is why some of us are speculating with existing CPU.
PD: Replace the PowerPC 970FX logic without cache for a Dual Gekko with multithreading in my theory.
I'm not sure if this is worth mentioning, but as long as we're dealing in speculation, I noted that when IBM unveiled the 970MP (in Japan - odd), they explicitly mentioned embedded use.
The only place I could envision that word being used in conjunction with the 970MP is in a console. The chip is specced at frequencies from 1.4-2.5 GHz and a die size of 154 mm2, all at IBMs current process. If it were to be used by Nintendo at that process, I'd assume that the frequency would be at the lower end, say 1.5 GHz.
I can't quite see the 970MP fit the bill until 65nm, but at that point it could probably serve its purpose quite well as a console CPU, and at a reasonable price (lowish clocks and sub 100 mm2) if they got a decent deal from IBM.
The only place I could envision that word being used in conjunction with the 970MP is in a console. The chip is specced at frequencies from 1.4-2.5 GHz and a die size of 154 mm2, all at IBMs current process. If it were to be used by Nintendo at that process, I'd assume that the frequency would be at the lower end, say 1.5 GHz.
I can't quite see the 970MP fit the bill until 65nm, but at that point it could probably serve its purpose quite well as a console CPU, and at a reasonable price (lowish clocks and sub 100 mm2) if they got a decent deal from IBM.
The 970MP is too big, expensive and hot for most embedded use and if it follows the rest of the 970 line it would require a separate microprocessor to get started. I don't see it ever fitting the bill for embedded, that seems like a niche that a certain in-order core would have been designed for.Entropy said:
It sounds like Nintendo is going for the low end of the market this time around, and with Sony and MS aiming for HD systems when most of the world has non-HD televisions they might be more successful than anyone thinks even if their system isn't as powerful as the others. If they can make it cheap enough they can move a hell of a lot of systems, not only in the industrialized countries but think about emerging markets like China or India where $400 for a console is out of the question but $100-150 might be within reach.
Shogmaster
Regular
jvd said:
970MP is not big physically, but I would think it's significantly less efficient at heat dissipation than the low power 970FX that was announced.
Shogmaster
Regular
jvd said:
LP970FX is suppose to dissipate 16w @1.6Ghz. So I'd hazard a guess that 1.6Ghz 970MP could be around double of that.
Well not publicly, but according to people who know the design was completed.
Again according to people who know, the VX starts at 1.6GHz and can go up to 2GHz. The VXe which was planned was to go from 2GHz and beyond.
Nintendo cares since 2-way is cheaper.
How easy would it be to make it backwards compatible with GCN games?
Guys, we are still speculating on that which we know nothing about. We do not know Nintendo's choice of heat dissipation methods, but the whole architecture seems to be constructed around it, so basing its projected power solely off of size may be wholly inaccurate. Even their choice of selecting MoSys 1-TSRAM-Q with its reduced power dissipation characteristics. As far as central processors are concerned, straight from the IBM horse's mouth:
You all are letting a prototype case size deceive you, both the GPU & CPU will be more than capable of next-gen visuals & feature sets. Perhaps while not including HD capable resolutions, they could easily do what Polyphony did with GT4, which wasn't a "true" HD resolution either. We simply do not know enough yet, Su's aforementioned comments surely do not sound like they are going with antiquated or "low end" CPU technology imo.
You all are letting a prototype case size deceive you, both the GPU & CPU will be more than capable of next-gen visuals & feature sets. Perhaps while not including HD capable resolutions, they could easily do what Polyphony did with GT4, which wasn't a "true" HD resolution either. We simply do not know enough yet, Su's aforementioned comments surely do not sound like they are going with antiquated or "low end" CPU technology imo.
Last edited by a moderator:
