AMD: R9xx Speculation

[Broken Hope]

It... changes core clock.. dynamicaly and continously?

it's basically reverse Intel turbo boost from what I can see, the default clock speeds are the maximum and then depending on what you have the TDP slider set to, the card underclocks dynamically to maintain that TDP.

 

[Broken Hope]

it's the same friggin' card

Heh are you suggesting the only difference between the 6950 and 6970 is that the 6970 can have the TDP set higher?

 

[Alexko]

it's basically reverse Intel turbo boost from what I can see, the default clock speeds are the maximum and then depending on what you have the TDP slider set to, the card underclocks dynamically to maintain that TDP.

I think you probably won't see all that much throttling during games, from what I gather, 3DMark's Perlin noise test is pretty math-heavy.

 

[itsmydamnation]

it's basically reverse Intel turbo boost from what I can see, the default clock speeds are the maximum and then depending on what you have the TDP slider set to, the card underclocks dynamically to maintain that TDP.

then why is default power a lower frame rate?

 

[Bouncing Zabaglione Bros.]

 

[Broken Hope]

then why is default power a lower frame rate?

Because the card is barely ever at its rated 800MHz clockspeed due to throttling.

 

[NathansFortune]

So it is ATi's anti-Furmark switch...

 

[3dilettante]

it's basically reverse Intel turbo boost from what I can see, the default clock speeds are the maximum and then depending on what you have the TDP slider set to, the card underclocks dynamically to maintain that TDP.

It's like speed-step or cool'n quiet, outside of the somewhat larger performance penalty than what would be acceptable for most CPUs on the desktop.

 

[Mize]

So it is ATi's anti-Furmark switch...

LOL.

 

[R300King!]

So does the 6950 have the turbo "switch" too?

If so, how would a "switched" 6950 compare to a non-switched 6970? The same?

 

[mczak]

So it is ATi's anti-Furmark switch...

That's a LOT more than a anti-Furmark switch.
Though I also find it strange that clock doesn't change in steps, maybe it's smoothed out?
Interesting that it says changing clock for "various blocks" so not everything has to run at the same clock?
For maximum power efficiency you'd also want it to change voltage that dynamically, something it apparently can't do. Still, I think this system makes sense in theory. You could ask though why it's needed if the card is built to handle the higher TDP level anyway.

 

[Mize]

When I read the slides I got the gist that it wasn't just clock speed but voltage as well.

 

[ZerazaX]

It adjusts each clock cycle

 

[Mat3]

So wouldn't this give every card a potentially difference experience? Just like over-clocking, it all depends on whether you got one of the good ones. So how much throttling it does to stay within a certain power budget might vary from card to card.... no?

 

[mczak]

When I read the slides I got the gist that it wasn't just clock speed but voltage as well.

I think if it could do that it would be mentioned explicitly, not only in section about the "old" (indirect) method.

 

[Arty]

So the next obvious (important) question - what directives did AMD give to reviewers? Bench at default, +20% or something else.

 

[liolio]

It would interesting if AMD comes with power profile for games

Some games run ridiculously fast (+100fps) it could be interesting to run the card at really low power, just to ensure the game run @+60 fps.

 

[DarthShader]

Interesting that it says changing clock for "various blocks" so not everything has to run at the same clock?

It would be too optimistic to expect different clock domains within the chip, right? At least for this generation.

So wouldn't this give every card a potentially difference experience? Just like over-clocking, it all depends on whether you got one of the good ones. So how much throttling it does to stay within a certain power budget might vary from card to card.... no?

Hard to say at this moment, but I think it might indeed happen, that one card runs at default power throtling, like a different card at +5% power (see Neliz' graph). But at power throtling off, each card would run at stock clock giving the same experience, possibly at different power draw and different oc headroom. Like it is now.

 

[turtle]

On topic: So how long until MSI (for example) changes the max TDP allotment from AMD's spec to the max PCIe spec for the dynamic clocking? Should we expect something like that soon, or rather rely on complete disabling of the 'feature' from someone like W1z? While it makes sense for both AMD and the consumer, certainly folks that over-volt lesser cards will not be happy with such developments, and could be a potential market lost to nvidia. Overclocking, over-volting, and going over not only max tdp but PCIe spec is almost a given for a product like the GTX570. If restraints are put in place mainly for product differentiation now (6970/6950) or in the future, ala we see a '6980' with a 300W max TDP, that is worrisome and there's potential value loss against nVIDIA products when measuring absolute performance (without using perhaps headache-inducing work-arounds); an area where something like GTX460 already trumps Barts without such a barrier, and GTX570/GF114 will likely hurt Cayman if not simply because of it. I imagine situations like this will continue into the future and this tech would do nothing but hurt AMD, considering nVIDIA's methodology of using greater IPC with a lower clockspeed (greater potential, with no 'cap').

I learned a hard lesson during this GPU's silly season: Don't even bother with speculating until the card is out.

Good lord I feel like such a tool for reading the last 1500 posts of this thread over the last month.

While I too often feel like a complete tool after things become known, I would reckon the vast majority of people here are forward-thinkers (call them/us INTP or whatnot) that thrive of deciphering the unknown, and care much less when the curtain is finally pulled back and all information wrung from the architecture; moving onto the next discussion. That's the great thing about this forum; people with a vast array of applicable knowledge throwing out interesting ideas based on sometimes obscure theories or esoteric understandings, realistic or not, good and bad, and the community hashing it out. Even if we don't get the perfect product from the result of that discussion, it's not to say those ideas won't be incorporated somewhere down the line, perhaps even because of them.

To make an analogy: I feel frustration for the individual that retroactively decides they wasted 6 years of their life after LOST's final episode because they didn't successfully predict or perhaps even like the result, but rejoice for the person that realizes that the 6 years of discussion/thought and knowledge learned, related or not, about something they enjoyed was worth more. While you're entitled either way, I think you'll find much fulfillment if you look at this thread (and others like it) in the later regard.

Sorry for OT; I just thought it warranted response so future participation, from the master of mints or otherwise, is not discouraged.

 

[flippin_waffles]

They aren't seriously planning to market Cayman against the 285, are they?
Sheesh.

Naw, they're marketing to users of the 285, in this particular slide. That's a very good marketing strategy.