Personal Revolution hardware vision.

1 PPE+ beefed up VMX + 1MB L2 + mini PPU @ 3.2GHz @ 90nm.
512MB of RAM
64MB ARAM
6MB eDRAM
512MB FLASH RAM
Modifed X1600 @ 600MHz @ 90nm.
Realtime DD 5.1/6.1 encoding
 
Some interesting ideas and some crazy ones. I can't believe the "Revolution is just a GC" idea is still rearing its insane head. There isn't a single bit of evidence to support it, in fact absolutely everything we know completely contradicts it. Yet some people persist with the idea, its a funny old world..... wide web :)
 
Last edited by a moderator:
I know with a shrink from 130nm to 90nm the TDP of the Athlon 64 dropped from 89W to 67W.

As for graphics, I doubt Nintendo went to ATi 4 years ago asking for a GPU that's an off-the-shelf part somewhere between an X1300 and X1600 with a memory controller/northbridge slapped on. I imagine they went with an out-of-the-box idea again. Akin to Flipper: a TEV like a pixel shader, then offload any sort of vertex shader onto the CPU.
 
Why would Nintendo use a general purpose CPU for something that can be performed much more efficiently by specialised hardware? Even DS has a vertex shader.. :)

Also I don't think there's any chance of Nintendo going with a proprietary pixel shader again. That happened with GC because its GPU was designed by ArtX before DX8. With ATI developing the GPU for 2006 there's no reason to alienate developers with anything other then DX shaders.

I agree that Hollywood won't be a off the shelf part though, no doubt ATI have developed a custom part in the 4 years they've spent on the GPU.
 
Last edited by a moderator:
Teasy said:
Also I don't think there's any chance of Nintendo going with a proprietary pixel shader again. That happened with GC because its GPU was designed by ArtX before DX8. With ATI developing the GPU for 2006 there's no reason to alienate developers with anything other then DX shaders.
Sorry, I didn't mean that Hollywood would have the same features as Flipper, I was just pointing out how Flipper's design was different from others at the time. What that different thinking might be this time, who knows.
 
OtakingGX said:
I know with a shrink from 130nm to 90nm the TDP of the Athlon 64 dropped from 89W to 67W.

As for graphics, I doubt Nintendo went to ATi 4 years ago asking for a GPU that's an off-the-shelf part somewhere between an X1300 and X1600 with a memory controller/northbridge slapped on. I imagine they went with an out-of-the-box idea again. Akin to Flipper: a TEV like a pixel shader, then offload any sort of vertex shader onto the CPU.

Meybe it is here the use of Fast 14, it would be great for the ratio of power, price and performance, I hope so.
 
pc999 said:
Dont know about MP but the FX uses 17W at 1,6Ghz, double that and you get my version of the MP.

You said 970MP @ 2GHz. And 2 GHz FX average power use is 40W.
 
That on 90nm, in 65nm I expect/guess it would be able to speed it up to 2Ghz as I already wrote (BTW in Intel the managed to get very nice speed improvements with 65nm at the ~same power, anyway it could not relate directely to this).
 
pc999 said:
That on 90nm, in 65nm I expect/guess it would be able to speed it up to 2Ghz as I already wrote (BTW in Intel the managed to get very nice speed improvements with 65nm at the ~same power, anyway it could not relate directely to this).

At 65nm you would expect 2 GHz MP to be somewhere around 40W. At that point you would expect multiple Gekko base CPU would reach the same theoritical peak performance for half the power.
 
V3 said:
At 65nm you would expect 2 GHz MP to be somewhere around 40W. At that point you would expect multiple Gekko base CPU would reach the same theoritical peak performance for half the power.

More like 34 or less, but anyway it should be much, much, much harder to get the theoritical peak performance from lots of Gekkos than from a MP and I think that easy of use/low cost is/should be one of the most important thinghs in Nintendo strategy.
 
pc999 said:
More like 34 or less, but anyway it should be much, much, much harder to get the theoritical peak performance from lots of Gekkos than from a MP and I think that easy of use/low cost is/should be one of the most important thinghs in Nintendo strategy.

Or you can match the power and get more peak performance. Its like a comparison between Xenos to 2 GHz 970MP.
 
V3 said:
Or you can match the power and get more peak performance. Its like a comparison between Xenos to 2 GHz 970MP.

It will always harder to programe but if it is cheaper and lower power it is possible (but IMO no prabably).

My main problem in speculation about nintendo HW is that their agreements are longer than MS, much longer, that could mean that they are creating a brand new CPU/GPU (like MS), but they are the one how need less of brand new HW to make such a investment and the ones how will gain less from that (MS need the specs because they dont anything more, Sony will use Cell in a lot of stuff) but for N there is no reason for not using costum HW (but in the other side since 2001/2 that ATI are making DX10 gfx so they may only want a costum next gen X2300 like), so anything is possible, althought I dont think that my sugestion as a bad one.
 
Okay here's some Rev specs (my idea)

Athlon 64 2 GHZ

Radeon X850 500-600 mhz (90nm)

384-512 RAM.

And you can do all that for 199. I'm not even sure an X850 would be that outpowered by Xenos.

With the Athlon, you get a cheaper die size chip by quite a lot over Cell/Xenon, and with all the devs complaining over in order they might like it. The 2Ghz would probably make for a very cheap chip without being far off the top of the line.

Hell, I'd see about pushing that X850 clock even higher on 90nm..

For maybe 260m transistors, a very capable machine.
 
that's X86 CPU and PC GPU architecture - too ineffecient and too costly for Nintendo.

expect hardware with alot more 'bang for the buck'
 
Last edited by a moderator:
I believe that we will have a Custom CPU and a Custom GPU.

Custom GPU could be:

-ATI X1600 without AVIVO and with eDRAM.
-Custom GPU based on Flipper with 8 TMU, 2 FP-TEV per TMU and support and complete support for OpenGL 2.0 ES.

Custom CPU could be:

-Solitary PPE running at 3.2Ghz with 512K L2 Cache.
-Redesigned Gekko with more L2 Cache, running at 1.5-2Ghz and with VMX on it.
-Modified for low power consumption PowerPC 970FX
 
In an attempt to reduce the number of Revolution specifications speculations threads, this one is now sticky.
 
pc999 said:
It will always harder to programe but if it is cheaper and lower power it is possible (but IMO no prabably).

My main problem in speculation about nintendo HW is that their agreements are longer than MS, much longer, that could mean that they are creating a brand new CPU/GPU (like MS), but they are the one how need less of brand new HW to make such a investment and the ones how will gain less from that (MS need the specs because they dont anything more, Sony will use Cell in a lot of stuff) but for N there is no reason for not using costum HW (but in the other side since 2001/2 that ATI are making DX10 gfx so they may only want a costum next gen X2300 like), so anything is possible, althought I dont think that my sugestion as a bad one.

The problem is that a lot of people (included me) are speculating with the PPE and the PowerPC 970FX and now I believe that we are wrong.

The fact is that Nintendo never has made a non-portable system around a CPU.
 
Thanks Ingenu it was a good decision IMO, I hope people keep this on topic and update Rev news here.

Urian said:
The problem is that a lot of people (included me) are speculating with the PPE and the PowerPC 970FX and now I believe that we are wrong.

The fact is that Nintendo never has made a non-portable system around a CPU.

I hope they care more with the CPU (or GPGPU) because I think it will be much more importante this time around, specialy once they will not go to HD, AI, animation, physics... will play a much bigger role IMO.

Anyway I think we should not consider a brand new architeture for Rev CPU or GPU simple because they wouldnt make return from such a huge R&D budget (NV said one new architeture cost around 350M$ for the GPU if I remember well) unlike Sony, so a PPE a FX/MP or even multi Gekkos (lets say a updated one + a VMX unit would be what around 15M transistores put 8 or 10 at 1,5Ghz (it should be possible given 65nm I guess) and you would got 24-30X the performance with a 120-150M transistores which is not big and probably not cool in 65nm, at least in theory), but Nintendo dont want 1) spend lots of money in devolopment in games 2) exclusives games or games that also use is special controls so it needs to be very easy and cheap to devolop; so I think that we also should take out those CPU which will take a lot of work/money to get good performance (that is way I think if they want it to be great performer it will be based in something that it is only optional not inerentely hard like Cell and Xenon), ence the many Gekkos is out of question for me.

I dont mean they can only use those components but they will use parts also based in other products from the companys it can even be a next gen PPC (PPC1070?) or the next gen low end DX10 from ATI (X2300?) as long they hadnt been the only ones spending the money in the R&D and it is cheap to devolop games, and they are only wathcing abd moddifing one of those parts for the console (ence the long time agreements).

So yes we may be wrong because we dint know what those companys are devoloping in paralel (besides Rev HW), but I belve that if we know we would be able the make some very acurrate guess about the HW in the Rev. But of curse this is only IMO I could be dead wrong too, but I think it is what make more sense).

But please (no sarcasm here) tell me why you think that we are indeed wrong, besides N never did a console based on the CPU (not even I sugested that, plus Sony/Cell/PPU/GPGPU (specialy from ATI) and myself with my opening sentence starting to see the things in others way, making CPU like/programable power starting to be much more important IMO), or at least CPU like.
 
The facts are:

-Jon "Hannibal" Stokes from Arstechnica said that IBM showned to Apple the PPC processor for the Revolution and Apple refused it.

-Where is the PowerPC 750VX alias Mojave? Shogmaster talked about in the forums and it was a PowerPC 750 with VMX. I believe that the PowerPC Broadway will be a Gekko+VMX running at 2Ghz and with 10-12W of power consumption only.
 
Urian said:
The facts are:

-Jon "Hannibal" Stokes from Arstechnica said that IBM showned to Apple the PPC processor for the Revolution and Apple refused it.

-Where is the PowerPC 750VX alias Mojave? Shogmaster talked about in the forums and it was a PowerPC 750 with VMX. I believe that the PowerPC Broadway will be a Gekko+VMX running at 2Ghz and with 10-12W of power consumption only.

IF he is right it may be from a lot of reaons, bad performance to a PC (just like Cell a brand new and very powerfull), a better deal with Intel, bad software/compability (eg OS) to a PC version, only ready to late (as long as we know it only had been complet a few weeks ago and in nest year they will use Intel so it does not make much sense put new CPUs out there), they probably would refuse any of the next gen CPUs and just like they had refuse PPC at all by 2006, so for me Apple refusing it does not show that Broadway is not based in a more conventional PPC architeture as they refused every future PPC.

However you may have a point in: why would they show them B. CPU if it is only a costum PPC, my first thought would be that in that case B will be a costum next gen PPC (as I exemplificated with, it would be inside my theory and IMO it would be great if that is the case).

A (no many cores, and the rumors is a dual core) gekko seems like very low performance (specially if they want ports), probably a 970FX would be better across the board, on the other side a Gekko+VMX would be ~15M transsistores against 58M from the FX (2x the cache too), anyway it is not a brand new architeture and also fits in my theory of no big R&D, cheap and easy to devolop, but still a low performer.

Anything is possible as long it is cheap, cool, easy, but still a nice performer (for the flops I sugest a vector coprocessor like, that can be cheap to once they are using 1T-sram, I guess).

Anyways tomorow I will see if this is right, I am already tired.
 
Last edited by a moderator:
Back
Top