PPU Overclocking

[gaara]

No doubt this will be a popular topic and this will be possible eventually, therefore any information you would like to share in regards to this topic please put it HERE

I've been searching around for a few weeks now and have suprisingly found absolutely no information on PPU overclocking or how it will be done, however I can pretty much reach the following conclusions

-PPU will most likely have a dedicated BIOS similar to GPU which stores specific GPU related information such as clock frequencies and voltage adjustment. This will most likely be the old school primitive way of getting anything more out of these cards and you will need a BIOS writing/editing utility and a BIOS flashing utility
-PPU will most likely not be built with a PCB with adjustable voltages therefore it will overclock similarily to a GPU with no vcore/vram adjustments, however no doubt there will be existing voltage mods and in some cases a BIOS softmod may allow a higher voltage if the PCB will allow it (this seems unlikely as there will only be one type of PPU, unlike multiple types of GPUs eg. 7600, 7800, 7900)

Current PPU information availible is as follows:
125 million transistors(PPU) vs. 384 million transistors(r580) vs. 233 million transistors (FX-60)
182mm2 190nm die (PPU) vs. 315mm2 90nm die (r580) vs. 199mm2 90nm die (FX-60)

Basing my opinions on this information, the PPU appears to be much less powerful than any CPU or GPU availible that would fall under high end. I am unaware of how exactly the PPU architecture will operate and couldn't find any information, however I assume it will be much much much less complex than a GPU and very similar to a CPU with a single scaling pipeline design, but don't quote me on that. Based on the fact that it's only slated for 128bit GGDR3 at a capacity of 128MB, I'm guessing RAMDACs will be around 600MHz with 1200MHz effective speeds, and a core speed somewhere around 500-600MHz. The card itself will probably be very similar in power to a 6600GT IMO. Interest points to raise:

-This card still uses 128 bit memory interface. This means you're gonna have to start bumping up RAM at least 100MHz over stock speeds to notice any change. This may have been to keep production costs down, but 256bit would have been a better option as it yields much higher memory bandwidth. 265bit would give the card much more potential for much more intensive processing IMO
-Card is still based on a larger fabrication. It'll run hot but if you can get more voltage through it, it'll be an advantage. I have no idea why they haven't moved to 90nm fabrication however it leads me to believe that they may have been worried about yields with a 90nm fabrication which means that the PPUs themselves may not be optimized or stable

 

[Static_11]

cant wait for one of these things once this thread gets popular a nd info gets dumped hear i vote sticky

 

[TheOtis]

Ya, as always, good post gaara.

I myself am looking forward to these and seeing how well they perform. And how much of an actual advantage it'll bring in real world games, not benchmarks.

My only consern is that you said,

"This card still uses 128 bit memory interface. This means you're gonna have to start bumping up RAM at least 100MHz over stock speeds to notice any change"

That's is quite a statement. That sounds like one heck of a drawback.

EDIT: If I understood it right.

 

[gaara]

Well what I mean is 128bit memory has to operate at higher speeds to compete with the wider 256bitrate memory and even then there isn't even any competetion between the memory bandwidth

So really if you were to get 100MHz on both 128bit and 256bit memory interfaces, you wouldn't see much of a difference with the 128bitrate but the 256bitrate would be significantly faster. It's not a huge deal but it just seems to ridiculous to me

 

[wtk pro l]

is this the ageia physx cards, if so someone on x-s got one. had it for about a week. tested 3dmark05 and 3dmark06 with it and saw no improvement so he gave it to someone else (probably kingpin) to see if he can utilize it in the tests.

 

[gaara]

Well obviously it isn't going to make any difference in 3Dmark as it isn't written with the code to use it so the PPU would be laying idle the whole time...these things are useless right now but I'm still a tad disappointed at the lack of information availible about them. Even the actual operation of them remains a mystery

I think the real thing to be wary of is the process size. Generally you see a larger process size being used when you can't get high enough yields on a lower process size, and considering now even more complex GPUs are using 90nm fabrication processes whereas this seemingly less complex PPU is still using 130nm, it leads to one to believe there are many many bugs not worked out in the production runs. In other words, you're looking at crappy overclocking potential

 

[Ridlyn]

Well, that's because there isnt any application or game(atleast that i know of) that supports a PPU, so right now you wont see any performance gain with or with out.

 

[locki433]

Ghost Recon advanced warfighter is suppose to use the ppu

 

[wtk pro l]

ok who says you can even overclock it? no apps support overclocking and there is no info on overclocking it.

 

[gaara]

Because everything that runs at a specific clock frequency must have a BIOS that stores the core frequency and voltage settings...no doubt the PPU will have its own BIOS with its own settings and the only particularily hard thing to do would be to manual adjust the BIOS settings