Powervr Underclocked?


1magus

Member
Joined
Jun 29, 2007
Messages
462
I hate quoting Wikipedia but it says the SGX530 runs at 200mhz (14 MPolys/s, 500Mpx/s@200MHz). On the fact sheet for Pandora it says the GPU runs @ 110. Is it dangerous or even possible to overclock the GPU inside the Pandora? I could easily be wrong about the GPU Clock Rate since the Pandora Wiki does not list it only Wikipedia :p

Also I just noticed the Dev units only contain 128MB of RAM does that mean it is impossible to design a program for Pandora that would require more?
 
Seriously, are you having some kind of long-lasting stroke where you are forced to create as many threads as possible?
GPU clock is 110. it's locked with other things like the ram rate so it can't be oc'd, IIRC. I could google for you but perhaps you could do that yourself? Maybe it'll take your mind off making more threads.

Dev units are just units that were already made. You can develop on any unit, you don't need a special one (that's one of the reasons Pandora is so awesome.) They all have 256 mb of ram.
 
rabidpoobear said:
Seriously, are you having some kind of long-lasting stroke where you are forced to create as many threads as possible?
GPU clock is 110. it's locked with other things like the ram rate so it can't be oc'd, IIRC. I could google for you but perhaps you could do that yourself? Maybe it'll take your mind off making more threads.

Dev units are just units that were already made. You can develop on any unit, you don't need a special one (that's one of the reasons Pandora is so awesome.) They all have 256 mb of ram.
Someone needs to chill. Is there something wrong with asking questions?
 
Last edited by a moderator:
Exophase said:
(unfortunately, the SGX clock is the same and still limited to 110MHz, and if you were somehow able to overclock it you'd be overclocking the RAM too which is probably not good)
 
Last edited by a moderator:
fischju2000 said:
Exophase said:
(unfortunately, the SGX clock is the same and still limited to 110MHz, and if you were somehow able to overclock it you'd be overclocking the RAM too which is probably not good)
OIC Yeah I remember trying to overclock my GPU on my computer I mean it worked but I had to overclock the Video RAM and Shader units and I'm no so sure it was stable. Overclocking RAM usually does not work too well AFAIK.
 
Last edited by a moderator:
TylerAW said:
Someone needs to chill. Is there something wrong with asking questions?

Not if you've already made a decent effort to search the internets and these boards and still can't find the answer.
 
Last edited by a moderator:
TylerAW said:
Someone needs to chill. Is there something wrong with asking questions?

I don't think you've made too many threads. But, in fairness, this is probably something you could have searched (evidenced by the post above). But, since it was posted well over 1/2 a year ago, it's not sooooo bad.

BTW: By overclocking RAM, it's meant that you would be overclocking the actual RAM for the Pandora, not VRAM. Overclocking VRAM is not a big deal. When raising your GPU's clock speed, shader clocks are *supposed* to be linked if I'm not mistaken, but you're free to leave VRAM clocks as they are. If you do OC VRAM, you're not going to OC it a whole lot though.
 
Last edited by a moderator:
quadomatic said:
TylerAW said:
Someone needs to chill. Is there something wrong with asking questions?

I don't think you've made too many threads. But, in fairness, this is probably something you could have searched (evidenced by the post above). But, since it was posted well over 1/2 a year ago, it's not sooooo bad.
I'm trying to get the most up to date info on the Pandora. I can't tell you how many threads I have made on various forums where someone said search google calling the poster lazy when they themselves probably know the solution to the question :). I do search google before I post to forums always.
 
Last edited by a moderator:
I'll try to explain this.

IMGTech, makers of SGX, is like ARM - they don't sell mirochips, they sell designs. This means they license you the rights to take their design (distributed in some low-level abstract hardware description, sort of like an executable file - pure data, but a black box you can't study or modify) and incorporate it into your design of some chip you're making.

When they say that SGX can clock at 200MHz it means that it's possible, using some readily available process technology, to implement the chip such that it'll reliably operate at 200MHz. But if you choose another technology you could be limited to something slower (likewise, if you reimplement it in a technology 5 years down the road you could possibly clock it much higher). Also, the other components of the greater design can be limiting the clock speed, especially if the clock is shared with other domains. Or perhaps the memory subsystem won't work at higher clocks.

In TI's case, OMAP35x is using a 65nm process and probably one optimized for low power consumption instead of speed. Process decisions are going to be determined by price, stability, availability, power consumption, and how much the rest of your design can be made to work in it. Since the original OMAP35s were released TI has come out with TI35x on a 45nm process, which is allegedly capable of clocking the SGX at 166MHz.

Incidentally, you may be wondering why the best commodity processes can only get SGX 530 to 200MHz while desktop GPUs can reach much higher clocks. If you want I can explain that, but I won't bother unless someone really is wondering.
 
TylerAW said:
quadomatic said:
TylerAW said:
Someone needs to chill. Is there something wrong with asking questions?

I don't think you've made too many threads. But, in fairness, this is probably something you could have searched (evidenced by the post above). But, since it was posted well over 1/2 a year ago, it's not sooooo bad.
I can't tell you how many threads I have made on various forums where someone said search google calling the poster lazy when they themselves probably know the solution to the question :) . I do search google before I post to forums always.

google query

5th result is where that Exophase quote came from.
Do you think the guy that replied with the link REMEMBERED where that post was or do you think he searched for it so he could post the link?

It's not a question of us knowing the answers, it's a question of you showing enough respect to spend 15 seconds submitting a Google query so you don't waste 15 seconds of EVERYONE'S time reading your question. That's a lot of seconds to be wasting. It then angers people because they know that you could have answered the question yourself and saved everyone a bunch of time, and in their bitterness they no longer feel like giving you an answer.

Sure, the way you ask questions will eventually get you an answer from the nicer guys on the forums. But it will also lose you a whole lot of respect.

Anyway, since you don't seem to appreciate my replies I'll try not to bother you with them anymore.
 
Last edited by a moderator:
Exophase said:
I'll try to explain this.

IMGTech, makers of SGX, is like ARM - they don't sell mirochips, they sell designs. This means they license you the rights to take their design (distributed in some low-level abstract hardware description, sort of like an executable file - pure data, but a black box you can't study or modify) and incorporate it into your design of some chip you're making.

When they say that SGX can clock at 200MHz it means that it's possible, using some readily available process technology, to implement the chip such that it'll reliably operate at 200MHz. But if you choose another technology you could be limited to something slower (likewise, if you reimplement it in a technology 5 years down the road you could possibly clock it much higher). Also, the other components of the greater design can be limiting the clock speed, especially if the clock is shared with other domains. Or perhaps the memory subsystem won't work at higher clocks.

In TI's case, OMAP35x is using a 65nm process and probably one optimized for low power consumption instead of speed. Process decisions are going to be determined by price, stability, availability, power consumption, and how much the rest of your design can be made to work in it. Since the original OMAP35s were released TI has come out with TI35x on a 45nm process, which is allegedly capable of clocking the SGX at 166MHz.

Incidentally, you may be wondering why the best commodity processes can only get SGX 530 to 200MHz while desktop GPUs can reach much higher clocks. If you want I can explain that, but I won't bother unless someone really is wondering.
Well I thought it was due to size, heat and power consumption.

Anyway thanks for the explination.

Since someone on this board who shall remain nameless seems to have a heart attack when I make a new thread I figure I will ask a few other questions on this thread :)

1. Is it possible to run Blender 3D on Pandora? I mean would one of the devs need to write drivers for PowerVR to work in Blender?

2. Can the PowerVR be underclocked to save power when it is not being used much?
 
Last edited by a moderator:
quadomatic said:
BTW: By overclocking RAM, it's meant that you would be overclocking the actual RAM for the Pandora, not VRAM. Overclocking VRAM is not a big deal. When raising your GPU's clock speed, shader clocks are *supposed* to be linked if I'm not mistaken, but you're free to leave VRAM clocks as they are. If you do OC VRAM, you're not going to OC it a whole lot though.

Shenanigans! GDDR3 was meant to be pushed to 2.7ghz and beyond! ;)

Exophase said:
Incidentally, you may be wondering why the best commodity processes can only get SGX 530 to 200MHz while desktop GPUs can reach much higher clocks. If you want I can explain that, but I won't bother unless someone really is wondering.
Not really, no. It's all tradeoffs, I gather. AVR Microcontrollers do a lot of stuff in 1 clock cycle, but can't scale beyond ~50mhz. If you wanted one to run at 100mhz, you'd have to redesign the internals so some things take 2 clock cycles.

TylerAW said:
1. Is it possible to run Blender 3D on Pandora? I mean would one of the devs need to write drivers for PowerVR to work in Blender?

2. Can the PowerVR be underclocked to save power when it is not being used much?

1) Not right now. Needs OGL ES 1.1 or ES 2.0 backend.

2) No. It can be ran at 55mhz or 110mhz, but that's it. However, the SoC may automatically cut power when it isn't needed or in heavy use. Many chips have such features - usually they happen beneath the OS, and cannot be controlled.
 
Last edited by a moderator:
TylerAW said:
1. Is it possible to run Blender 3D on Pandora? I mean would one of the devs need to write drivers for PowerVR to work in Blender?


if you searched the boards for "Blender" you would have gotten a few pages of results. In there somewhere is your answer. I could link to it but, like you, I can't be bothered with the effort.
 
Last edited by a moderator:
PhillipJ said:
TylerAW said:
1. Is it possible to run Blender 3D on Pandora? I mean would one of the devs need to write drivers for PowerVR to work in Blender?


if you searched the boards for "Blender" you would have gotten a few pages of results. In there somewhere is your answer. I could link to it but, like you, I can't be bothered with the effort.
I did search Blender I don't know why everybody always assumes someone has not searched the boards or google.
 
Last edited by a moderator:
TylerAW said:
I did search Blender I don't know why everybody always assumes someone has not searched the boards or google.

If I can find it fairly quickly using search and you claim that you searched and didn't find it then am I supposed to think that you're somehow bad at searching or that you didn't actually search?
 
Last edited by a moderator:
Kramy said:
Not really, no. It's all tradeoffs, I gather. AVR Microcontrollers do a lot of stuff in 1 clock cycle, but can't scale beyond ~50mhz. If you wanted one to run at 100mhz, you'd have to redesign the internals so some things take 2 clock cycles.

Sorry, but I'd say that's not the primary thing that happens - as CPU clock speed has increased they've also trended towards pushing more instructions per cycle on average. CPU designers generally aren't going to want to trade efficiency for clock speed because the extra clock speed increases heat, power consumption, and yield failure rate (therefore cost) and back in the day would raise complexity of the rest of the system. 6502, for instance, traded clock speed for increased efficiency, although it had to adopt an annoying ISA to do it - but it was still very competitive. Of course there was the whole Pentium 4 debacle, but that's generally regarded as a mistake.

What you're describing does show up a bit in increased latencies, but issue rate doesn't usually go down much.

Atmel could probably make at least an 800+MHz AVR w/o compromising timings at all, but it'd require a significantly different design that'd make no real sense in any market.
 
Last edited by a moderator:
TylerAW said:
Also I just noticed the Dev units only contain 128MB of RAM does that mean it is impossible to design a program for Pandora that would require more?

Some of the *very* early dev boards (sans case) only had 128Mb RAM. Still I don't think any DEVs are trying to use up all the RAM anyway!
 
Last edited by a moderator:
Whenever I see an Exophase comment, I immediately imagine this:
05.jpg
 

Attachments

  • 05.jpg
    05.jpg
    388.1 KB · Views: 82
Back
Top