The plan a bit more detailed (2011-09-27)


Each core is not much different to the older Cortex A8 cores so they should be comparable by clock rate, to get their aproximate speed. So these chips should be 8 times faster then what we have on the Pandora, both systems not overclocked.
The significant addition of the out of order engine, in addition to pipeline improvements, leads to the Cortex A9 being rated at 2.5 DMIPS/MHz compared to the Cortex A8's 2.0DMIPS/Mhz. This leads to straight up 25% improvement on a per clock basis from A8 to A9, and a more significant improvement with workloads in-order does poorly with. The Cortex A15 due to going to triple issue, from dual, and pipeline improvements is believed to represent at least a 50% improvement over the A9 on a clock per clock basis.

Ergo:


A9/clock>=1.25*A8


A15/clock>=1.5*A9>=1.875*A8


While this is rather impressive the rule of thumb when emulating is a ten fold processing advantage and the PS2 is a rather quirky piece of hardware. In my current unprofessional opinion OMAP 5 with settings appropriate for a handheld device, that is to say 1-1.5Ghz dual core according to ARMH, would probably insure that Nexius and whatever people do with IdTech4 in the near future, but not allow for emulating most 6th gen consoles like the PS2. Even if we're generous and go with A9 equivalence we're talking 1.5-2.25GHz equivalent verse 486MHz with the GameCube and 733MHz with the XBox for a maximum factor of 4.6 and 3.1 respectively. The PS2 is quirky enough to ask for at least a 3.2GHz Core 2 Duo equivalent at this point, which a handheld configured OMAP 5 has nothing on.


The funny thing to me is people like to say the Pandora is obsolete because there are newer ARM processors on the market, but the developers have shown a strong ability to get 5th Generation and backwards console emulators operational. Yes more processor capability is more processor capability and customer buying decisions are often heavily influenced by a shallow reading of specifications, but power should have a purpose. So seems as how we're really not budging meaningfully from the current set of emulated systems in the foreseeable future, I am kind of curious what people would use that power for if the company managed to create a magic wand and materialize even the still unreleased OMAP 5 on 28nm process and magic up the rest of the stuff to make it all work.


Don't misunderstand me. I'm not advocating simple stagnation. I'm just genuinely curious what people are expecting.
 
^ Except for referencing the rule-of-thumb-that-creates-misunderstandings (it might be a good first measure of feasibility, but it creates baseless expectations in people, like for instance the whole DS emulation thing), I nominate this post's third paragraph ("The funny thing...") to be a great response to those calling pandora obsolete, especially for emulation. Given the amount of extra processing power (per watt, we don't want to sacrifice battery life, right?) gained from newer SOCs, what new use cases would it enable to switch?
 
Each core is not much different to the older Cortex A8 cores so they should be comparable by clock rate, to get their aproximate speed. So these chips should be 8 times faster then what we have on the Pandora, both systems not overclocked.
The significant addition of the out of order engine, in addition to pipeline improvements, leads to the Cortex A9 being rated at 2.5 DMIPS/MHz compared to the Cortex A8's 2.0DMIPS/Mhz. This leads to straight up 25% improvement on a per clock basis from A8 to A9, and a more significant improvement with workloads in-order does poorly with. The Cortex A15 due to going to triple issue, from dual, and pipeline improvements is believed to represent at least a 50% improvement over the A9 on a clock per clock basis.

Ergo:


A9/clock>=1.25*A8


A15/clock>=1.5*A9>=1.875*A8


While this is rather impressive the rule of thumb when emulating is a ten fold processing advantage and the PS2 is a rather quirky piece of hardware. In my current unprofessional opinion OMAP 5 with settings appropriate for a handheld device, that is to say 1-1.5Ghz dual core according to ARMH, would probably insure that Nexius and whatever people do with IdTech4 in the near future, but not allow for emulating most 6th gen consoles like the PS2. Even if we're generous and go with A9 equivalence we're talking 1.5-2.25GHz equivalent verse 486MHz with the GameCube and 733MHz with the XBox for a maximum factor of 4.6 and 3.1 respectively. The PS2 is quirky enough to ask for at least a 3.2GHz Core 2 Duo equivalent at this point, which a handheld configured OMAP 5 has nothing on.


The funny thing to me is people like to say the Pandora is obsolete because there are newer ARM processors on the market, but the developers have shown a strong ability to get 5th Generation and backwards console emulators operational. Yes more processor capability is more processor capability and customer buying decisions are often heavily influenced by a shallow reading of specifications, but power should have a purpose. So seems as how we're really not budging meaningfully from the current set of emulated systems in the foreseeable future, I am kind of curious what people would use that power for if the company managed to create a magic wand and materialize even the still unreleased OMAP 5 on 28nm process and magic up the rest of the stuff to make it all work.


Don't misunderstand me. I'm not advocating simple stagnation. I'm just genuinely curious what people are expecting.

+1
 
You probably won't be losing much battery life with a faster SoC, the new ones tend to use about the same amount of power.


But I agree with you, purely for games and emulations we don't need more speed, but on the desktop we could really benefit from a dual core. You said power should have a purpose, and there it is, not in emulation and running newer games but getting the Pandora to be able to do more on the desktop, at the same time, and when doing heavy loads like flash video in software.


Running Midori with 5 tabs, Pidgin, Exaile, and 2 PDF's open and the system is JUST able to handle it, though you may say the system is not meant for that, if it had another core and more ram it could easily be meant for that. Not that it is no capable with such a workload at the momen, but lag is noticable. Not that I'm complaning, my workload on the Pandora is quite heavy.


Looking at the raw numbers, it is theoredically possible to emulate the PS2 on a A15.


ARM Cortex-A15 (Quad Core) 35,000 MIPS at 2.5 GHz


AMD Athlon FX-60 (Dual Core) 18,938 MIPS at 2.6 GHz


Not saying this is the whole picture, because there is still the graphics to talk about, and the fact that thats a quad core system and the other is a dual core system.
 
You probably won't be losing much battery life with a faster SoC, the new ones tend to use about the same amount of power.
Not quite.  The A8 is designed with an _under_ 0.3W draw in mind.  A Cortex A9 on the same process as the current OMAP is rated at 0.4W when clocked at 830MHz and a dual core A9 on TSMC 40nm @ 800Mhz is rated at 0.5W.  Take that to the max at 2GHz and your processor die is now drawing 1.9W, plus whatever the other SoC components are adding and the rest of the system.  Newer Cortex processors however benefit from increased use of power gating, being able to process more at lower frequencies, and being quicker in the race to execute.


For comparison the 3DS has a 5Whr battery and hence in battery burner configuration is drawing roughly 2W, for the entire system.  Resistive heating loss in a circuit is related to square of the current, which is directly related to the clock frequency.  While ARM processors tend to be low power enough to not fry themselves thermally, that power has to come from somewhere.  This is why Texas Instruments has intentionally designed the OMAP 4 and OMAP 5 to keep the Cortex processors and GPU gated off as much as is practical with ARM M-series and an unknown 2D compositor taking up the slack.

Running Midori with 5 tabs, Pidgin, Exaile, and 2 PDF's open and the system is JUST able to handle it, though you may say the system is not meant for that, if it had another core and more ram it could easily be meant for that. Not that it is no capable with such a workload at the momen, but lag is noticable. Not that I'm complaning, my workload on the Pandora is quite heavy.
Fair enough.  As I said I'm not advocating technological stagnation.

Looking at the raw numbers, it is theoredically possible to emulate the PS2 on a A15.


ARM Cortex-A15 (Quad Core) 35,000 MIPS at 2.5 GHz


AMD Athlon FX-60 (Dual Core) 18,938 MIPS at 2.6 GHz


Not saying this is the whole picture, because there is still the graphics to talk about, and the fact that thats a quad core system and the other is a dual core system.
A) You didn't specify what type of MIPS those are.  Assessment programs vary, hence ARMH is rather specific about using DMIPS aka Dhrystone MIPS in their figures.  That said branch prediction for the A9 has been noted as being along the lines of the early Athlon series.


B) Integer Instructions are not equivalent across architectures, especially when you're talking a 32-bit RISC ISA verse x86-64 with SSE3.  This especially matters when the emulated system is 64-bit and hence your 32-bit processor is going to need to do fancy tricks that are going to tend to demand more then twice the number of equivalent operations.


C) You're proposing a maxed out _server_ configured part at the max clock rate which at 28nm is liable to eat through the Pandora's battery in a couple of hours.  The effect of that on the battery tending to be far from what one would call comfortable.  It's not something you'd put in a handheld until at least another node down the line and I remain on the sceptical side of it being practical then.


D) I see no justification for your baseline, which does fall below what I brought up.


If I was going to make this case I'd be focussing on the yet unreleased PowerVR series 6 GPU and using it as a GPGPU asset for emulation.  That however would represent quite a bit of development effort with OpenCL.  As far as I'm aware such efforts do not currently exist, which is part of why it's hard for me to gauge if that'd really help that much.  Series 6 can beat the PS2's theoretical polygon count, Mpixel rate, and GigaFLOPS with a solid margin. From what we do know of Series 6 GPU at this point it compares favourably with the XBox 360's GPU, but isn't quite at the same level. While the PowerVR SGX 544 will have a advantage over the PS2's graphic capabilities it won't simply be dominating in a single core configuration and thus I'd tend to say wouldn't be setup to go crazy with attempting GPGPU emulation.
 
Last edited by a moderator:
Back
Top