Replacing the CPU?


Back in the days when I had a steady hand I stripped down 2 broken VIC 20's and built one working one. At the time I was already experiencing mild hand tremors, so I'd have a pint or 2 before soldering - I had betablockers but they gave me nightmares so booze was better as it also gave me the balls for soldering more expensive stuff (and boobs for something extra to play with).


That was a difficult enough job, and we're talking about a full size computer with massive 1970/80's components to make life easier. To attempt what you are considering you'd first have to appropriate the miniaturisation device from "Fantastic Voyage" and shrink yourself and your tools to an appropriately microscopic size (even if you already have a microscopic tool - it's probably useless). Unfortunately that 1960's classic film was not based on factual events, and the machinery doesn't exist. Therefore your fist hurdle will be inventing the miniaturisation device.


Let me know how you get on with that. I'd be interested in trying it out sometime.
 
More or less for those who don't get it, ARM SoCs do not usually have similar pinouts like some x86 CPUs. The OMAP3 used in Pandora has nowhere near the same number or configuration of pins as an OMAP4 or Snapdragon or any other non-CortexA8 SoC (the DM3730 is supposedly a drop-in replacement for the OMAP3530, and ups the clockrate and maybe memory).


Therefore, in order to use a different SoC, one would need to make an adapter board for it, plus add hardware to replace any functionality missing from the new chip that was present in the old, plus completely rewrite all the device drivers and kernel to handle the new configuration.

I don't really know first hand, but I've observed that between the 3430 and 3630, at least, even though it's supposed to be drop-in replacement, you end up having to tweak your device drivers to some extent because some of the devices included in the SoC have been updated.
 
Last edited by a moderator:
There's no need to vilify the ignorant.

Has anyone messed around with replacing the Pandoras cpu?
As already alluded to, the required precision demands equipment you're not liable to possess, with a 3430 Rev. F or 3730 Rev. C being the furthest you could dream of going.

With OPP5 unlocked the existing CPU should OC to 1150MHz and has been tested stable to 1000MHz. 45nm based parts would allow for a slightly higher ceiling, but will run into instability issues before moving meaningfully ahead of the existing 65nm 3430. That slight edge should be unnecessary for anything having to do with actual practicality, although it may help with battery life. Not that the Pandora has a battery sized where battery life is really a meaningful issue.

A quick google search reveals that the ARM® Cortex™-A8 600Mhz+ CPU is getting old. The Cortex-A9 Processor and the Cortex-A15 Processor are newer and faster.
I'm not aware of any plans for the Cortex A15 to be fabbed at 45-32nm, so it isn't relevant until near the end of the year when the 28nm node is expected to mature into mass, instead of risk, production.

It's worth pointing out if for some reason you do decide you need more power, you can setup the Pandora as a Bluetooth controller to talk with another device. Really though it has a ridiculous amount of native power being analogous to a solid Gigahertz race era machine, while being a pocket computer without the volatility or input limitation issues of the old PDAs.


If it helps, the Pandora should have more power under the hood then the original XBox ever had.
 
Last edited by a moderator:
I'm pretty sure I'll regret replying to you, but maybe we can keep this civil.


1GHz at OPP5 is far from a lock and I've never heard of anyone doing 1150 on a Pandora is uncommon:





What's important to consider is that the clocking headroom can be a lot lower depending on what the program does. For instance, any usage of NEON can lower maximum overclock a lot, as was seen when PCSX-reARMed started using it. If NEON is introduced more into standard libraries or the OS you'll see this reduction spread across the board - it only really takes one incorrectly executed NEON instruction to blow things.


So at 1GHz you might have more CPU power than XBox at 733MHz (GPU is a very different story), but at something like 800MHz it's a much different comparison. Clock for clock Cortex-A8 will usually not come near Pentium 3, even one with only 128KB of L2 cache. I really don't think we're look at "GHz race" level power, we're looking at something closer clock by clock to Intel Atom. That is before you factor in memory performance, where it's a lot worse than netbook level Atoms.
 
Come on guys, lets have a serious conversation about replacing the CPU.
There is none. You just let us know that you really don't know what you are talking about.

Do you really think the developers are fools and could not do this replacement if it is as simple as you are dreaming about?


If you think so, just do it.

Beagleboards are not specifically designed to be used as a base for mobile gaming, but it could turn into a mobile gaming device.
Yes, i heard of a group trying to add mobility to the CPU family used on the beagleboard. They added a LCD, a keyboard, an additional SD slot and some kind of gaming controls to it. Very nice work, you should get your hands on it, it could be near to your thoughts what a mobile beagleboard could be.

To help you with your wish of replacement, you will need to find the reference manual for your new CPU and compare it to the one of the beagleboards for matching pinout. If it fits it's fine, else you need to solder some wires between to match the pinout. Use at your own risk, i don't take any responsibility for that.

You wouldn't happen to have a link to the project site of those guys who turned a beagleboard into a gaming system? I've googled but I can't seem to find anyone who turned a beagleboard into a portable gaming rig.


Basically to sum up this thread, it's impractical to replace the cpu because you'll need special equipment. There are very very few drop in replacements with the same pin set up as the original cpu. You'll have to take out the ram and the cpu and solder both back in place if you did choose to replace the cpu. You'll break the volume wheel in the process of replacing the cpu and you'll need to buy another one.


Good to know.
 
I'm pretty sure I'll regret replying to you, but maybe we can keep this civil.
Your own title on the other board admits as much that, that is up to you.

1GHz at OPP5 is far from a lock and I've never heard of anyone doing 1150 on a Pandora is uncommon:
Thought I remembered someone emphasizing they could do it on their N900 and expected once things were refined down it should be possible on the Pandora. I'll concede the point on particulars although it doesn't modify the gist of the statement.

So at 1GHz you might have more CPU power than XBox at 733MHz (GPU is a very different story), but at something like 800MHz it's a much different comparison. Clock for clock Cortex-A8 will usually not come near Pentium 3, even one with only 128KB of L2 cache. I really don't think we're look at "GHz race" level power, we're looking at something closer clock by clock to Intel Atom. That is before you factor in memory performance, where it's a lot worse than netbook level Atoms.
If Intel insists you must go there it's 2DMIPS for the A8 verse 2.5DMIPS per clock for the A9 with a scheduler of Athlon quality, and a 500MHz dual core Cortex A9 dev board beat a 1.6GHz Atom in what the entire netbook concept revolves around despite the blatant processing throughput disadvantage. Hence you're not exactly setting yourself up for success by dragging that out, particularly under the paradigm of "netbook." I know what you're trying to get at, but it's splitting hairs so you can make a scene over an attempt to console a customer that his super duper handheld is plenty powerful.

The XBox 1 much like it's successor used one set of shared system RAM particularly 64MB of DDR-200MHz verse the Pandora's 256MB of DDR-333, which was a serious bottleneck to system performance as the same approach is to it's modern cousin. I'd really be surprised if the latencies aren't lower, on top of the clock difference, on the Pandora given it's been 10 years and the state of the art was a 180nm process at that point. Yes the Pandora's GPU is 10 million polygons/sec verse 125 at stock frequencies but it should beat the GameCube's 12 under any meaningful overclock and unless all the talk about being able to run Doom 3, including by yourself, when people were shouting down the 3730 back in 2008 was completely without basis it's _close enough_.
 
Last edited by a moderator:
Your own title on the other board admits as much that, that is up to you.

Guess that's a no then, right? Would somebody else please speak up on this one, I really don't want to deal with this alone this time.

Thought I remembered someone emphasizing they could do it on their N900 and expected one things were refined down it should be possible on the Pandora. I'll concede the point on particulars although it doesn't modify the gist of the statement.

How much it changes things depends on just how much any given person can overclock.

What does Atom, particularly in netbooks, have to do with anything related to Gigahertz race era computing?

I was using it as a much more appropriate comparison against Cortex-A8 (clock for clock, ignoring SMT). The reason why I mentioned "in netbook configurations" is because Atom in netbooks has MUCH more memory bandwidth than any Cortex-A8 deployed, including Pandora's.

Other then you must wave the Intel flag at every opportunity, and hence must try to pick a fight any time you think their name may even theoretically be sullied? Hence you not only try to pick a fight over an attempt to console a Pandora customer/user but pick what borders on the absolutely best case for you to wave the Intel flag from, because waving the Intel flag is apparently all that matters to you.

This has absolutely nothing to do with supporting Intel. The only reason I mentioned Pentium 3 is because you were making a comparison against XBox. Replace the original K7 Athlon with Pentium 3 and it too destroys Cortex-A8. And you are the one making this personal. YET AGAIN. All that matters to me is having an objective technical discussion about CPUs, something I'm very interested in. Unfortunately whenever I talk to you about CPUs no one else wants to join in the discussion, although I can imagine why.

Even if you want to go there it's 2DMIPS for the A8 verse 2.5DMIPS per clock for the A9 with a scheduler of Athlon quality, and a 500MHz dual core Cortex A9 dev board beat a 1.6GHz Atom in what the entire netbook concept revolves around despite the blatant processing throughput disadvantage. Hence you're not exactly setting yourself up for success by dragging that out, particularly under the paradigm of "netbook."

DMIPS is an absolutely terrible benchmark, something that everyone in the industry admits, but companies still use it anyway as a fluff figure. Especially when comparing Cortex-A8 in configurations that are very limited by their memory hierarchy (something Dhrystone ignores). Ask Ari64 what he thinks about Cortex-A8 vs Atom, or Pentium 3 for that matter. A9 absolutely does not have a scheduler of K7 quality. I think ARM would be the first to admit this - you simply don't get this level of OoO in this power envelope. A15 is more around that level. A9 gives you fairly rudimentary OoO with a small window and no centralized reservation stations, and only a small amount of renaming. This is of course with a dual-issue pipe vs triple issue P3 or K7. Furthermore NEON/VFP is in-order,


No 500MHz dual core Cortex-A9 beat a 1.6GHz Atom in anything, why don't you check your references. The video you're referring to showed them both loading websites, with the A9 only somewhat behind - not exactly surprising, given that loading web pages is prone to all sorts of bottlenecks that have nothing to do with CPU performance. The very thought that a 500MHz Cortex-A9, even with two cores, can beat a 1.6GHz Atom, is astounding. I think clock for clock Cortex-A9, with an appropriately matched memory subsystem, will perform better than an Atom, and two cores helps (but not linear scaling, and the advantage is a bit diminished vs SMT), but it's certainly not that much better.

The XBox 1 much like it's successor used one set of shared system RAM particularly 64MB of DDR-200MHz verse the Pandora's 256MB of DDR-333, which was a serious bottleneck to system performance as the same approach is to it's modern cousin. I'd really be surprised if the latencies aren't lower on the Pandora given it's been 10 years and the state of the art was a 180nm process at that point.

OMAP's memory is 32-bit, vs XBox's 64-bit, therefore XBox has more bandwidth. They both have shared memory - Pandora's GPU uses less bandwidth since it's TBDR, but since I was only talking about CPU performance it's not relevant to the comparison (if you want to compare GPU or CPU + GPU let's keep that separate). I don't know what XBox's main RAM latency was like, but OMAP3 is known to be very poor in this regard (again, ask Ari64, or Laurent).. which isn't surprising because it has another L3 bus to go through on top of the CPU's AXI bus, and the RAM is low power which means more latency. Process doesn't help latency quite as much when we're talking about things going off the chip.

Yes the Pandora's GPU is 10 million polygons/sec verse 125 at stock frequencies but it should beat the GameCube's 12 under any meaningful overclock and unless all the talk about being able to run Doom 3, including by yourself, when people were shouting down the 3730 back in 2008 was completely without basis it's _close enough_.

I don't know the extents of the GPU capability, except that SGX530 is more versatile and probably all three of those polygon counts are divorced from reality. I also doubt I said it could run Doom 3, because I wouldn't be the least bit qualified to make that determination (I'd go with whatever darkblu has to say about it instead). What I do know is that most of what people are using Pandora for right now stresses the CPU and not the GPU, so for many people the CPU comparison will be more relevant.


.. and you did mean 3530 there, not 3730, right?
 
Guess that's a no then, right? Would somebody else please speak up on this one, I really don't want to deal with this alone this time.
Sorry mate, I agree with you but you know very well I lack the articulation and specific background to back it up.
 
Well now children - sounds to me like you're trying to compare apples with oranges. What a couple of fruits. :lol:
 
Last edited by a moderator:
You are right titanuranus - and sometimes they didn't even now how an orange look like.

You wouldn't happen to have a link to the project site of those guys who turned a beagleboard into a gaming system? I've googled but I can't seem to find anyone who turned a beagleboard into a portable gaming rig.

Sure. Today i am a litte bit more up to date, the group started the production and is using a beagleboard like configuration for their device: ARM A8 CPU, TMS320C64x+ DSP Core, PowerVR SGX OpenGL 2.0 ES compatible 3D hardware, 256MB RAM, 800x480 4.3" 16.7 million color touchscreen LCD, Wifi 802.11b/g, dpad and analogue padsticks for games, 43 button keyboard.

And here it is!
Sorry, couldn't resist.^^
 
Last edited by a moderator:
Back
Top