Revolution specs - the good side

[ERP]

I did read it, but it was just a speculative scenario. What kind of quality software wouldn't check that a DMA has completed? I can't see it being common with Gamecube games.

It's common in PS2 titles.....
You pad out DMA cahins with NOPs to match known timing.

I can't think of a general case on Gamecube though, but it's not massively uncommon for products to be timing sensitive, the question is what the result of the timing variation is. I'd still expect High 90's percent compatability though.

 

[Shifty Geezer]

What else would it be? Seems like the only conclusion that can be drawn in this case.

Why could it not be an underclocked 970, or any other normally faster processor underclocked? I can't see how a low clockspeed limits the choice of more modern and usually faster processor.

 

[pc999]

Why could it not be an underclocked 970, or any other normally faster processor underclocked? I can't see how a low clockspeed limits the choice of more modern and usually faster processor.

To be fair Fafalada actually give anouther reason (external Ram) to say that it may be a 750 CPU and he refered the rumors, so he do have reasons to say that, so all of this is moot point, as he didnt said - See less than 800Mhz ergo 750 like.

But yes speed alone is useless to say the CPU speed, althought a 970FX would be very well in the power department up to 1,6Ghz (17W) so it really would be strage if they do have that kind of CPU at such low speed, IMO.

 

[Guden Oden]

970FX is probably a much too big and expensive a chip compared to what they're using, even though its power consumption might not be cripplingly high...

 

[Bohdy]

Seriously, Nintendo invested a decent bit of money into the Broadway chip. I find it really hard to believe that it is simply a higher-clocked 750CXe...

 

[Guden Oden]

I think I'm going to assume from now on that Rev is simply a cost-reduced and upclocked GC with roughly twice the RAM, and then I can be pleasantly surprised come E3 if that turns out to not be the case.

 

[Shifty Geezer]

To be fair Fafalada actually give anouther reason (external Ram) to say that it may be a 750 CPU and he refered the rumors, so he do have reasons to say that, so all of this is moot point, as he didnt said - See less than 800Mhz ergo 750 like.

Ignoring his other points, which he explained as moot because of the clock speed...

Anyway I think it's all a moot point. The very fact that Rev CPU is rumoured at sub 800Mhz clocks pretty much leaves no doubt that it's from the same 750x family as Gekko.

...the reads very much to me as 'less than 800 MHz ergo 750 like'. If Faf had said 'along with such and such features, sub 800 MHz leaves no doubt...' it'd be a different story.

 

[ban25]

Why could it not be an underclocked 970, or any other normally faster processor underclocked? I can't see how a low clockspeed limits the choice of more modern and usually faster processor.

It's just not logical to go through the hassle and potential compatibility quirks of a new microarchitecture only to clock it at a third its native frequency. Who's to say a 970 would even be much faster than a 750 at such a speed?

 

[Shifty Geezer]

It's just not logical to go through the hassle and potential compatibility quirks of a new microarchitecture only to clock it at a third its native frequency. Who's to say a 970 would even be much faster than a 750 at such a speed?

I dunno that it would for sure, but a later architecture should have architectural advantages. And you woudn't hve to redesign a 970 to run at a lower clock, just use an off the shelf part. As for using a new processor instead of a faster Gekko, you're right about the compatibility, but that's not part of the 'low clock-speed == older CPU design' argument!

 

[Entropy]

What else would it be? Seems like the only conclusion that can be drawn in this case. In fact, I wouldn't be surprised to find that it is a 750CXe, since that processor tops out around 700 MHz anyway.

The 750CXe has a die area of 42.7 mm2 at 0.18 um lithography. At 0.09 um this would correspond to just over 10 mm2, and if we assume that Nintendo will scale it down with time, it corresponds to roughly 3 mm2 (!!) at 0.045 um.

I'll submit that this seems like a strange way to go unless they intend to make a system on a chip. (Small detail - the 750 CXe consumes 6W at 600MHz and 0.18um lithography. Using modern processes and the low frequencies that have been mentioned, we would be looking at around a Watt or so in power draw. No need for a fan, or even a heatsink.)

 

[ban25]

The 750CXe has a die area of 42.7 mm2 at 0.18 um lithography. At 0.09 um this would correspond to just over 10 mm2, and if we assume that Nintendo will scale it down with time, it corresponds to roughly 3 mm2 (!!) at 0.045 um.

I assume the chip will be pad-limited before then. However, a SoC approach is not out of the question. I believe Fafalada's interpretation was that the 64 MB "external" memory reference could mean the original 24 MB was integrated on-die. The way I look at it, the memory is probably segmented into two pools (if the original report is accurate), a 24 MB pool and a 64 MB pool on a separate interface. One could be integrated or both could be external, but functionally they would be the same.

I'll submit that this seems like a strange way to go unless they intend to make a system on a chip. (Small detail - the 750 CXe consumes 6W at 600MHz and 0.18um lithography. Using modern processes and the low frequencies that have been mentioned, we would be looking at around a Watt or so in power draw. No need for a fan, or even a heatsink.)

 

[pc999]

Ignoring his other points, which he explained as moot because of the clock speed...
...the reads very much to me as 'less than 800 MHz ergo 750 like'. If Faf had said 'along with such and such features, sub 800 MHz leaves no doubt...' it'd be a different story.

Ok, it is a interpretation issue, personally I think that this is in a context so it is all involved being more a conclusion than anything else.

BTW althought I had quote you, I as trying to adress all the "problem" not just you.

I must confess that lately I am expecting also some sort of a 750, yet extention is doing a lot of damage here again (like in every Rev specs thread).

 

[Teasy]

What else would it be? Seems like the only conclusion that can be drawn in this case. In fact, I wouldn't be surprised to find that it is a 750CXe, since that processor tops out around 700 MHz anyway.

What can it be other then an overclocked but otherwise unmodified Gekko? An upgraded 750 based CPU, say a 1MB 4 way cache and VMX unit added just for example. I can see the logic in thinking that its 750 based. But there's no logic in coming to the conclusion that the CPU is basically an unmodified overclocked Gekko CPU based soley on clock speed..

 

[Shifty Geezer]

To be radical, how's about a dual-core CPU, a Gekko and a more modern something-else, with switchable clockspeeds? For BC use the one core and drop the clock, while for Revolution enable both cores and rev up to full throttle. If they're similar cores, just one with a better cache and registers or somesuch, it shouldn't make the devs lives too hard.

 

[OtakingGX]

What can it be other then an overclocked but otherwise unmodified Gekko? An upgraded 750 based CPU, say a 1MB 4 way cache and WMX unit added just for example. I can see the logic in thinking that its 750 based. But there's no logic in coming to the conclusion that the CPU is basically an unmodified overclocked Gekko CPU based soley on clock speed..

Well, the 750GX does have 1 MB L2. I'm not sure on the VMX unit though, I don't think it has one. I'm sure if that's what they based the processor off of it would likely look something like a 750GXe with Gekko's VMX derivative on it.

 

[pc999]

There is rumors as "extreme" as the 750VX, that is supossed to be up to 2Ghz, VMX, 1mg L2, and dual core ready, with only 256Kb and DualCore it should be a great CPU for a low cost console (as each upgraded CPU shouldnt be 50% bigger than the original Gekko, so with Moores law it would end a very small CPU, less than Gekko already a very small chip.

Althought I think that Fafalada (?) said that this never existed.

BTW the Gekko memory is already great (even ERP said that it would prefer if Nintendo hadnt so worried with memory so they didnt cut corners in the rest), why wast so much transistores in L2.

What about some hard wired or special HW for the most intensive task? Like the animation low level AI, physics... etc, would that be possible and economically feasible (from a R&D POV at least)?

 

[Squeak]

I know this is the third time I mention this (just so you know I know I’m repeating myself), but IF it is a SoC with 24Mb on die, wouldn’t the potential extra bandwidth CPU<--->VPU and memory speed allowed by the integration, to a large degree outweigh the disadvantages of the low clockspeed?

 

[Entropy]

I know this is the third time I mention this (just so you know I know I’m repeating myself), but IF it is a SoC with 24Mb on die, wouldn’t the potential extra bandwidth CPU<--->VPU and memory speed allowed by the integration, to a large degree outweigh the disadvantages of the low clockspeed?

Only to the extent that it would alleviate those situations where the system would otherwise be constrained by bandwidth or latency.

Just how much performance could be gained by having 24 MB integrated into a SoC feels like a pretty academic question however. Whereas integrating memory controller, peripheral control, CPU and possibly GPU might make sense from an economic standpoint, putting 24 MB of memory onto the same die looks like an expensive proposition motivated by performance alone (at this point in time). Doesn't seem to fit Nintendos design ethos for the Revolution.

I've never been able to wrap my head around the idea of Nintendo simply offering the same chips as before but on newer processes. It would result in tiny (particularly if they take advantage of future process shrinks!) chips with no improvements in either packaging costs or interconnect/PCB requirements. Some kind of SoC would make more sense, but has its own problems. Just what would it include? Who would design it? Who would fab it? Too many pieces of the puzzle are missing.

 

[Rolf N]

Just how much performance could be gained by having 24 MB integrated into a SoC feels like a pretty academic question however.

It does, but in some ways I find it to be the most interesting path of speculation wrt Revolution for what ... the past three months?

There's plenty of precedent. Nintendo has used SoCs with embedded memory in all their handhelds and I think it served them pretty well, both in terms of manufacturing costs and performance.
What immediately sprung to mind is that Nintendo might just be an as of yet unannounced user of Z-Ram. I think this might actually enable embedded memory on that scale, if they are (I don't know).

Does Gecko really have 1MB of cache? That's a whloe lot. It seems such a waste of space if you consider the very low latency of the main memory, and now they'll have to carry it forward for BC.
(I know it is still under debate whether or not cycle-exactness matters for BC, but it's my prerogative to personally believe that yes, it does)

 

[pc999]

Does Gecko really have 1MB of cache?

Gekko only have 256K of L2.

BTW that Z-Ram is interesting, althought it seems to new (ence costly, althought in the long term it may be cheaper if I understand well)) for being used in Rev, I guess.