What's the most powerful system that can be made from 2014 mobile parts (Tegra, arm)

Proelite

Veteran
Supporter
How many arm core cores? How many discrete cpus? One for gaming and one for OS? Frequency?

How many gigs of ram? LPDDR3, DDR3 or GDDR5? Separate or single ram pool for gaming or OS?

What gpu? K1? Series 6? What frequency? How many cores?

Limitation:

Smaller than 360 slim, and cheaper than $500
 
Last edited by a moderator:
There's really not an answer to this, you understand why, right? Let me give you some examples:


"What's the most powerful car you can make from 2014 parts?"
How many horsepower? How much torque? How many gears? How many seats? How many winglets? How much fuel? How much cargo space? What's the maximum speed? What's the braking distance?


"What's the most powerful plane you can make from 2014 parts?"
What's maximum service altitude? What's maximum speed? How many engines? How many landing gear? How many passengers? How many support staff?


These are equally absurd questions, for obvious reasons. You could build a nearly build a LEO craft with specific parts from specific planes. You could probably build a 400mph "car" using very specific parts. Would either of these be a reflection of what is actually usable?

You could build a 500-socket ARM-based system with terabytes of ram on a custom Linux OS, using any number of underlying IO / storage systems. But that doesn't really make sense in the same way as a low-earth-orbit plane or 400mph car. You, as a consumer, can't "make" those things. And while some other company could, they simply will not, because you couldn't afford it anyway.

What will be built is what can be realistically sold. Perhaps that's not the question that you wanted answered, but your existing question has no specific answer. I would instead suggest you try again.
 
Last edited by a moderator:
how many arm cores(Cortex-A57)? tens of thousands
keep adding them until you run out of money
each chip (8core) supports 128gb ram
clockspeed 2ghz
 
Smaller than 360 slim, and cheaper than $500
And I assume for a console rather than a mobile, the intention being to evaluate whether/how a mobile-based console could be competitive with the current machines?

If so, it'd be worth adding basic specs (eg. BW) to the responses. Although we still have the old chestnut of 'most powerful' trying to compare different architectures. Which mobile GPU is the 'most powerful'?
 
Tegra K1 seems to be the most powerful that I know of. But I dont keep up much and maybe that's just Nvidia's hype machine. It is 192 Cuda Cores. Should be something in league with but clearly above PS3/360. But even that depends on if it can be fed enough bandwidth in a mobile configuration.

I'm even less knowledgeable about the memory layouts of mobile stuff...

$500 would seem like an awfully hefty budget. That's equal or more than PS4/XO with hefty desktop parts!

And you could presumably of course clock things quite high compared to actual mobile in a console form factor.

I dont know, given the high budget, could you just chain multiple K1's, as many as the budget allows, if you're looking for maximum punch? No idea how/if bandwidth could keep up.
 
Do the parts need to currently be available for customers?

Imagination Technologies and Vivante at least have very modular graphics designs which in their words can scale to the TFLOPS range with enough cores.
If you went to them with enough money they'd likely to be happy to make such a semi-custom design.

Then if WideIO2 is not yet available, see if Toshiba can supply a faster version of the Stacked Chip SoC memory interface technology used in the PSVita.
 
Did you (Proelite) mean mobile parts as existing SoC or as available IPs (so custom SoC /chips)?

EDIT
IF we speak of IPs I would think either INtel or Nvidia could do great tiny things.
 
Last edited by a moderator:
I was wondering what Sony/Microsoft could have done with their machines if they'd chosen a similar budget and not gone for the cameras and casual markets.

I'm daydreaming here, but if they'd managed to produce two chips at the same size as their current setup, they could have easily gone for 32/34 CUs on the GPU and about the same number of Jaguars (i.e., 28-32) on the CPU, with a couple of Bulldozers thrown in. One Bulldozer for the job allocation and the other for single-threaded work.

I wonder whether that would have worked out at a similar cost to the Xbox One with Kinect. If they had done something like that, I would have an Xbox One now and not a PS4.
 
Some details from the PowerVR website:

Based on a scalable number of compute clusters the PowerVR Rogue architecture is designed to target the requirements of a growing range of demanding markets from mobile to the highest performance embedded graphics including smartphones, tablets, PC, console, automotive, DTV and more. Compute clusters are arrays of programmable computing elements that are designed to offer high performance and efficiency while minimising power and bandwidth requirements.

With a growing range of cores optimised for either area (maximising GFLOPS/mm2) or performance (maximising GFLOPS/mW), PowerVR Series6 GPUs can deliver 20x or more of the performance of previous GPU cores targeting comparable markets. This is enabled by an architecture that is around 5x more efficient than previous generations. PowerVR Series6 GPU cores are designed to offer computing performance exceeding 100GFLOPS (gigaFLOPS) and reaching the TFLOPS (teraFLOPS) range enabling high-level graphics performance from mobile through to high-end compute and graphics solutions.

Apparently reaching the TFLOPS range. I hope Amazon try something crazy like this.
 
the question remains, how much would it cost them and why would they bother? sony and microsoft lost billions doing that for a long time. Its no longer in to be a loss leader if your the one footing the bill
 
the question remains, how much would it cost them and why would they bother? sony and microsoft lost billions doing that for a long time. Its no longer in to be a loss leader if your the one footing the bill

If I think back to when Sony and Microsoft created their first consoles, they both went for the most powerful machine in order to actually gain a market share. What’s to say Amazon/Google/Apple don’t try something similar? Some of them certainly have the funds to do it. I’d guess that mobile parts would be an awful lot cheaper than those offered by either Nvidia/AMD

What’d make this task easier is having a resource that describes the size of mobile processors/GPUs at different fabrication sizes. Then you could estimate a chip size and a 10ish% redundancy (for something as large as a Xbox One / PS4).
 
http://en.wikipedia.org/wiki/ARM_Cortex-A15
4 cores per cluster, up to 2 clusters per chip with CoreLink 400 (CCI-400, an AMBA-4 coherent interconnect) and 4 clusters per chip with CCN-504[12]. ARM provides specifications but the licencees individually design ARM chips, and AMBA-4 scales beyond 2 clusters. The theoretical limit is 16 clusters, 4 bits are used to code the CLUSTERID number in the CP15 register (bits 8 to 11) [13].

64 Cortex-A15s sounds like it'd be fun. God knows how big the chip would be though.
 
I’ll go with the Cortex-A15, 4 cores / cluster @ 16 clusters /w 64 discrete processors. Ideally with a Bulldozer or two in there for process allocation. The PS4’s BOM suggests $88 for the 8GB GDDR5, so I think I’ll stick with exactly the same RAM setup as that. Maybe something like a AMD 280X as the GPU, as I can’t see that any of the mobile graphics providers have anything that’s much greater than 1tf.

Using the PS4 material costs ($100 for the SOC, $88 for the RAM), I’ll suggest the following costs:

CPU = $100 (or 112 if Bulldozers are possible – no idea how big something like that would be)
GPU = $100
RAM = $88
Motherboard = $100 (number guesswork)
HDD 500GB = $50
So extra ancillary chips, pad, Wi-Fi, connections, fan, etc. = $50

I might be conservative with the CPU cost, so you could probably shift some of that value over to the ancillary stuff if needed.

I’d drop the optical drive, as the device could have some form of Android and probably won’t need discs. Preload with a game. Likely much smaller than a 360 Slim as it doesn’t have as many optical drives. Costs hopefully around the $500 budget.

Anyway, I’m not all that technically minded (I work in clinical research), so I’m fairly sure I could be wrong on all of the above. ;)
 
You can make a pretty powerful machine with mobile parts with a budget of $500.
The biggest mistake I think can be made by some of these companies is making it an Android platform.
There is no real reason to make a console powerful if all you are doing is running games designed for mobile phones and tablets. If any other company wants to really compete in the home console market they need to have it run on a closed platform. Im sure they could use an android based OS, but the games would need to be made only for that platform to ever compete with the quality of the software on the big 3's systems. Just my opinion.
 
I heard Tegra K1 is only used by heavy-performing tablets/devices, like Windows tablets. Maybe it's due to the fact that Tegra K1 will likely have overheating problem, especially if it's used on casual Android tablets. But i might be wrong. It would be interesting to see Android tablets with Tegra K1. In fact, i'm actually planning to buy HP SLate 7 Extreme just because it has Tegra 4 inside. But Android tablets with Tegra K1 will need at least 3 to 4GB of RAM given that this processor really takes a lot of space in the memory, especially when the graphics intensive stuff are activated on the tablet.
 
Back
Top