Unlocking Pipelines

Has anybody tried to unlock pipes on their Go 7900 GS or Mobility X1800. Both these cards are slightly modified versions of their more powerful siblings. Now, I'm not too sure about this, but are these cards not the same core, but with the pipes locked.

The 7900 GS has 20 shader pipes, whereas the 7900 GTX has 24. I would imagine it would be expensive for nVidia to develop a new core, which physically only has 20 pipes. It just wouldn't be feasible. I'm wondering if unlocking a few other pipelines can be done.

Its a similar story with the Mobility X1800. The XT version has the full 16 pipelines enable, but the X1800 (non XT), has 12. Ihaven't had much time to investigate the cores and specifications, but it seems that there is very little information on the Internet.

What do you think people?

 

Unless you got one of the very first "non xt 1800's" or 7900GS's than the other piplines are just gonna probably damaged/mest up/not working.

 

I heard that the 4 pipelines on the GS were physically "locked" during manufacturing. If they could be unlocked with software, I am sure the GTX bios would have done the trick already.

 

This is true about the go 7600 as well. It does not use all the pipelines that are physically there, but they are permanently disabled.

 

It would require an unmodified bios and then flashing your bios with the different one. However more recently they have wised up to this and lock them physically inside the chip.

 

yup 7900 GS is laser-locked to only get a max of 20 pipes.

too bad tho, thats the main bottleneck that max its slower than the 7800 GTX (24 pipes)

 

Thanks for the information everyone. I was also thinking that they would do somthing to stop the extra pipes from being accessed. What do you think about the Mobility X1800. There seems to be very little infromation about the X1800, but thats almost certainly because there aren't that many notebooks with an ATI high end card in them.

 

Usually, when pipelines are locked, it's because they don't work.
It generally makes sense to sell the card as the most expensive model possible (which means enable all pipelines that work).
(Note I said *generally*. Sometimes they do just lock a few pipelines to keep up with demand on cheaper cards, without having to lower the price on all their high end cards. But normally, the reason some cards have disabled pipelines is because they can't run with them enabled)

So even if you do manage to enable the pipelines, it's quite possible that you'll either:
1: End up with a GPU that just doesn't work (And good luck reverting your changes then)
or 2: Get a GPU which doesn't crash, but shows graphical artifacts

 

I heard it was the same with cpus, all are made to higher standards than they are and some are clockable at higher frequencies than others which dictates where they end up for sale i.e. 1.66 1.8 whatever. That may no longer be the case though.

 

No, I don't want to unlock any of the pipes on my card. I was just wondering exactly how the pipes are disabled, and if they could actually be unlocked. I understand entirely what you are saying, that they are locked because they didn't function, or didn't work correctly. Intel and AMD also pull the same trick, by disabling the cache on their processors, as I'm sure you're aware of.

 

From what I gather, they're still doing this. Most notably with the Core Solos--when one core doesn't function properly, they disable it. It helps them improve their yields.

 

Ah right, guess I misunderstood you then. Sorry

Yeah, it's true. All CPU's of a given stepping are the same during manufacturing. Once they're manufactured, they're tested to see:
1: Which ones don't work at all
2: Which ones work if we lower the clock speed
3: Which ones work if we disable half the cache
4: Which ones work if we both disable half the cache, *and* lower the clock speed
5: Which ones work just fine

(Of course, different variations on this theme exist. As said above, Core Solo's simply have one core disabled. And on the other hand, AMD's CPU's with 1MB vs 512K cache are currently different dies. The 512K versions aren't 1MB with half of the cache disabled. I'd guess this is because they've got excellent yields already, so they simply didn't have many with defective caches. So more cost efficient to simply develop a smaller die with only 512K cache in the first place)

But that's really all that differentiates, say, the E6300 and E6800. Or the Athlon 64 3800+ and 4800+.

(Again, they sometimes just downgrade CPU's which could work at full speed, just to fit market demands, which is why some low-end CPU's can still overclock like mad)

 

Like that Celeron that got to 4500MHz? Celeron-Ms are the new overclockers baby, seen 420s (1.6GHz stock) at 2500MHz getting 21 seconds for 1 mill digits of pi! Stock cooling too I think!

 

Have you got a link to that, Jess? Thats impressive!

 

I can imagine in a few years, or even now, somebody building their own laser from a broken cd player so that they can unlock the pipelines...

Why you'd wanna go to that trouble is up to you...

 

I have done overclocks like the Celeron that you speak...
Dual Core Opteron 165 (1.8ghz and 1mb+1mb= 2mb cache total)

with stock cooler and basic MSI Neo4-F motherboard.
it went up to 2.9 ghz stable (no voltage mod).... thats pretty much $150 for a CPU thats faster than FX-60.

Freakin awesome, I about to do the same with a Core 2 Duo E6300 when I get my hands on this year's Intel Retail Edge combo (E6300, Intel board, and Vista Ultimate for $200)

 

Well, Opterons are awesome overclockers.