D
Deleted member 13524
Guest
Oh just forget it. We're doing circles here.
Predict: The Next Generation Console Tech
- Thread starter Acert93
- Start date
- Status
IBM has not been pushing (or they kept it secret) anything new that could be a match for console makers needs (I don't think POWER6,7 or A2 qualify). I don't see either Nintendo founding R&D for something completely new on its own. That leaves the possibility of an evolution of Px/PPU.
It would make sense as those chips clock nicely within acceptable tdp. It would make sense for Nintendo to ship something as fast as Xenon/PPU as in some case clock speed rules. If they consider portability for cheap they have to ship something clocked as fast as PPU/Xenon (so 3.2GHz). I'm not sure that Nintendo would consider faster clock speed as power doesn't grow linearly with speed and even with a meatier "power/thermal" budget they will have to do arbitrations. That leaves architectural improvements
What can they do (insights welcome) to make their CPU significantly better than Xenon (by any info we have it's their reference as far as perfs as concerned)?
Would they make the CPU wider, so 3 issue (instead of 2)?
Could they have implement a cheap form of OoO execution?
Could they move to 4 way SMT?
A bit shorter pipeline?
Could they have implement a better branch predictor?
Better SIMD?
Wider SIMD?
Better pre-fetching capabilities?
Various change to the caches?
More L2 caches?
Fix thing here and there?
etc. ?
I would not be surprised if Nintendo come with a Xenon 1.4 than a Xenon 2.0. So it could be like a Xenon done right. To me it's more Nintendo like to "make it right" than to implement complex feature as OoO or 4 way SMT. They may have get rid of obvious offender we heard often about like: cache trashing, LHS, pre-fetching capabilities (I can't find where I read it but remember complains on the matter, might be wrong tho) and do minor improvements here and there.
I would put my bet on minor improvements like a slightly shorten pipeline, slightly better branch predictor, slightly lower access latencies to L1&L2, more L2 cache( I would bet on 2MB), better implementation on the silicon (in regard to power efficiency, cache density for example).
I wonder if Nintendo will have dare the jump to 8 wide SIMD units. Again I would be inclined to say no. They may simply have come with their rendition of VMX128 (I assume it would a bit better). But looking forwarding it would be a nice (the best?) place for big N to invest its silicon as it becoming the std on PC and would consistently raise the CPU throughput.
Overall I'm looking for minor improvement in perfs per cycles ~10%, possibly more in some cases. I expect the CPU to include power management feature and if TDP allows a "turbo mode" could be a nice thing to set the CPU more appart from Xenon perf wise.
It would make sense as those chips clock nicely within acceptable tdp. It would make sense for Nintendo to ship something as fast as Xenon/PPU as in some case clock speed rules. If they consider portability for cheap they have to ship something clocked as fast as PPU/Xenon (so 3.2GHz). I'm not sure that Nintendo would consider faster clock speed as power doesn't grow linearly with speed and even with a meatier "power/thermal" budget they will have to do arbitrations. That leaves architectural improvements
What can they do (insights welcome) to make their CPU significantly better than Xenon (by any info we have it's their reference as far as perfs as concerned)?
Would they make the CPU wider, so 3 issue (instead of 2)?
Could they have implement a cheap form of OoO execution?
Could they move to 4 way SMT?
A bit shorter pipeline?
Could they have implement a better branch predictor?
Better SIMD?
Wider SIMD?
Better pre-fetching capabilities?
Various change to the caches?
More L2 caches?
Fix thing here and there?
etc. ?
I would not be surprised if Nintendo come with a Xenon 1.4 than a Xenon 2.0. So it could be like a Xenon done right. To me it's more Nintendo like to "make it right" than to implement complex feature as OoO or 4 way SMT. They may have get rid of obvious offender we heard often about like: cache trashing, LHS, pre-fetching capabilities (I can't find where I read it but remember complains on the matter, might be wrong tho) and do minor improvements here and there.
I would put my bet on minor improvements like a slightly shorten pipeline, slightly better branch predictor, slightly lower access latencies to L1&L2, more L2 cache( I would bet on 2MB), better implementation on the silicon (in regard to power efficiency, cache density for example).
I wonder if Nintendo will have dare the jump to 8 wide SIMD units. Again I would be inclined to say no. They may simply have come with their rendition of VMX128 (I assume it would a bit better). But looking forwarding it would be a nice (the best?) place for big N to invest its silicon as it becoming the std on PC and would consistently raise the CPU throughput.
Overall I'm looking for minor improvement in perfs per cycles ~10%, possibly more in some cases. I expect the CPU to include power management feature and if TDP allows a "turbo mode" could be a nice thing to set the CPU more appart from Xenon perf wise.
why in your opinion a reworked and simplified power7 is not suitable for a console?
considering the eerie ibm's ability to regularly deliver every 3 year a new architecture, can't the next sony/ms console use a parallel developed/derivated power8?
D
Deleted member 13524
Guest
Actually, if Nintendo is getting anything near "off-the-shelf", the PowerPC A2 sounds just about right.
Some people seem to consider it the successor for the PPU.
Each core is 4-threaded and uses the latest Power ISA v.2.06 spec (everything in the PPU + hardware virtualization + VSX (which might get a custom upgrade to VSX128 to keep up with VMX128)). There's already a 4-core (16-thread) version at 1.4GHz consuming a mere 20W. There's a lot of unneeded stuff in there, like 10Gbps ports and PCI-Express lanes.
I think the PowerPC A2 could very well be the basis for the Stream's CPU.
A 3-core version with some of the network hardware functions stripped down, could go down to ~280M transistors (the 16-core version has 1.4billion). Shrinking this simplified 3-core Power A2 to 28nm could keep the power consumption well within some ~25->30W while clocking it up to 3GHz, and easily beating the Xenon.
Some people seem to consider it the successor for the PPU.
Each core is 4-threaded and uses the latest Power ISA v.2.06 spec (everything in the PPU + hardware virtualization + VSX (which might get a custom upgrade to VSX128 to keep up with VMX128)). There's already a 4-core (16-thread) version at 1.4GHz consuming a mere 20W. There's a lot of unneeded stuff in there, like 10Gbps ports and PCI-Express lanes.
I think the PowerPC A2 could very well be the basis for the Stream's CPU.
A 3-core version with some of the network hardware functions stripped down, could go down to ~280M transistors (the 16-core version has 1.4billion). Shrinking this simplified 3-core Power A2 to 28nm could keep the power consumption well within some ~25->30W while clocking it up to 3GHz, and easily beating the Xenon.
That, or they'll do the cheap, easy thing and ask for something PPU derived.
2.06 does not mandate VSX, and A2 does not feature it. POWER7 does, tough.
Where? It clearly says "Prototype" (=TapeOut) in early 2011 and production start in H1/2012.
I'm not sure either options are worse it, A2 is a "non FP" design and for only three cores it looks like "over designed" for the job (even IBM claim that single thread perf will be low they went for reduced power consumption). For POWER7 I don't see the point to keep a a wide heavily multi-threaded front end to feed a number of execution units drastically reduced.
I still put my bet on something PPU based.
D
Deleted member 13524
Guest
Care to elaborate with a source?
Power7 isn't really an option, though.
liolio said:
Reduced power consumption sounds just about right for a console that will supposedly be the same size as a SNES.
Refreshment
Regular
I don't think the rumors say same size as the SNES, its the aesthetic or apperance of the console that its similar. If im wrong please correct me. I bet Nintendo would probably go with something like the first PS2 or 360 slim model in size.
I don´t mean this particular roadmap, especially since it is a fairly new one. If you read my post carefully, you wilI notice that I used term "such" not "this". They cannot foresee all problems they will encounter, these roadmaps are marketing anyway to draw customers, because once a chip is designed for a certain process it requires great affords to port it to another foundries process. Just look at these roadmaps...
Global Foundries Roadmap from November 2009, just one year older than the one you posted.
http://www.behardware.com/news/10543/globalfoundries-roadmap.html
And here one from TSMC.
http://dnenni.files.wordpress.com/2010/01/tsmc-roadmap3.jpg?w=600&h=344
Last edited by a moderator:
Didn´t the rumors say that the console is as big as the launch 360?
Refreshment
Regular
Holy Sh!t!
Bless the amzing Sh!t! AU Optronics have true geniusess working for them. The kind of displays they come up with are nothing short of extraordinary. Stuff like fingertip scaning touch displays or transparent multi touch screens There's nothing Nintendo could deliver that top that for me, i hope someone eventually put stuff like this in a pad.
I missed this replay:
Much, much volume than a SNES.
Bless the amzing Sh!t! AU Optronics have true geniusess working for them. The kind of displays they come up with are nothing short of extraordinary. Stuff like fingertip scaning touch displays or transparent multi touch screens
There's nothing Nintendo could deliver that top that for me, i hope someone eventually put stuff like this in a pad.I missed this replay:
Well even bigger then
Much, much volume than a SNES.
Last edited by a moderator:
Megadrive1988
Veteran
Do we really think Microsoft and Sony will wait until 2014 ( or later ) to release Xbox3/PS4 with the new Nintendo console coming mid-late 2012 ?
That may well be true, I think it happened in the Xbox 1 days (I seem to remember Doom 3 getting more GPU assistance on the Xbox than on the PC).
There may be other benefits too. For example, for games that fully utilised both cores on my Opteron 170, there were significant drops in performance for going from 1T to 2T memory command rate (benchmarks showed that this cut my memory b/w by about 1/3) and for dropping back to stock DDR 400 memory clocks from 500 (even with tighter memory timings at the lower speed). In other words the A64 architecture on DDR1 may have faced additional bottlenecks that Xenon doesn't have to face.
That may very well be true also, and I wouldn't argue against this. This would perhaps support John Carmack's claim that fixed platforms see more performance extracted from them than PC.
Last edited by a moderator:
That's not just a claim, it's a well known fact that you can get more juice out of fixed box when you know exactly what machine will be running it, instead of PC reality where you have to make sure it runs on "anything" out there
I know this, you know this, lots of people know this! But on this very forum were PC gamers calling bullshit on Carmack for saying this, basically calling him a liar and saying it was all Rage PR.
Got link to the statement by Carmack by any chance? If it was just saying you get more juice out of closed box, I can't understand how PC gamers (which I'm one of, too) here would have raised hell about it - however there's more ways of saying that, which of some could indeed raise a hell.
- Status
Similar threads
- Locked
