So does the 40nm Cayman have significant amounts of hardware disabled? Obviously there is some redundancy in there of course, but reading how it was a 32nm design makes me think that it could be like Fermi and have some extra disabled hardware to make it a more manufacturable giant chip. Especially since the 32nm design was said to have been planned with more shaders and redesign is awfully time intensive.
AMD: R9xx Speculation
- Thread starter Lukfi
- Start date
I don't believe we can ignore it simply because it's the only example so far. Do you have any data?
The simple reason is the lower stock voltage the HD6950 uses. Undervolt a HD6970 to 1.10V (instead of the default 1.175V) and you will also see lower power consumption. The question is if all Caymans can handle the HD6970 clocks at HD6950 voltages, and that reliably. I guess we agree that it is quite doubtful.
Edit:
Looks like GPU-Z is claiming the unlocked HD6950s are also running at 1.175V. So it will come down to binning and the different memory.
Last edited by a moderator:
He had a sample of [strike]15[/strike] 24 cards
At least for the possibility of getting a HD6950 unlocked and running at HD6970 clocks. Complete success so far.
He had a sample of [strike]15[/strike] 24 cards
At least for the possibility of getting a HD6950 unlocked and running at HD6970 clocks. Complete success so far.
I believe Dave was referring to the power characteristics of the unlocked 6950.
I know. That's why I added the second sentence
Hmm, this article suggest at least 14 different cards were tested from 7 different vendors. Are you still of the opinion that this article, their results, etc, etc should still be ignored? I honestly am curious as to why we are seeing what we are seeing. Perhaps an official response from AMD (now that this has gotten the attention of the press) is in order.
I find it very odd your upset because I'm asking a question here. Obviously the issue in and of itself isn't being ignored. So I later replied asking for some clarification. You on the other hand have added a tad bit of drama, unnecessary. As it appears you are taking my post to mean something that it does not.
Then why do you tolerate reviews publishing power figures with such sample sizes, if they are meaningless?
Maybe because most reviews agree with the same delta between 6950 and 6970?
Well, review samples are probably from the same early group and would you expect AMD and nvidia to stop review sites publishing findings? What I'm trying to say is that their is no good way to establish a mean when different subsets may have quite different characteristics. Sample size will not necessarily improve the situation.
leoneazzurro
Regular
Because basically there aren´t two GPUs even of the same batch with the very same power consumption? There should be differences even in the same wafer! And the same goes for all the other semiconductor components of the board.
It´s practically impossible for everyone to send out two cards with precisely the same power consumption. And that´s practically the cause of some reviews showing the 6970 having the same power draw of the GTX 570 and many others showing it lower thatn this. What you can have are power ranges. And what we have is 14 modded boards (6950) against ONE 6970. This is not much to make a statistical average, that 6970 could even be the worst in the world about power usage...
That said, it is indeed possible that in the average the modded 6950 consume less, because as Dave said the 6970 normally are binned from higher leakage parts, that usually are also the more overclockable parts. And the memory chips are different, so it´s likely that also the 6970 could consume more power on the memory chips. Finally, I don´t remember if the power regulator block is the same, but it is likely that id 6970 is designed to handle more power, also more power is consumed on the VRMs.
The funny thing is, if they leave this alone, this will probably gain AMD about 10X as many sales as they would have got otherwise.
-Hardcore PC gamers love to mod/twiddle with things (same reason there is overclocking)
-They love to feel like they got a "steal" by overclocking/modding
Hell, I bet a lot of people would go out of their way to do this even if Nvidia offered a card with better performance at the same price, just so they can feel like they pulled one over on the man, or are "elite".
I mean this is the first time people are actually excited about the 69XX series.
Honestly it would be a good sneaky marketing idea to build this in on purpose.
-Hardcore PC gamers love to mod/twiddle with things (same reason there is overclocking)
-They love to feel like they got a "steal" by overclocking/modding
Hell, I bet a lot of people would go out of their way to do this even if Nvidia offered a card with better performance at the same price, just so they can feel like they pulled one over on the man, or are "elite".
I mean this is the first time people are actually excited about the 69XX series.
Honestly it would be a good sneaky marketing idea to build this in on purpose.
Last edited by a moderator:
There is going to be a significant delta between a 6950 and 6970 at the default settings, because they are running different voltages. However picking one single sample and comparing against another single sample and coming up with an absolute is meaningless, especially when it is modded to run outside of its default configuration; you always have to have a statistically meaningful sample when dealing with these things - never take a single review sites power comparisons as the absolute, you have to watch the trend across a large sample of reviews. In this case, it is clear that TPU's original review samples are towards the high leakage range for the XT and low leakage for the PRO because the XT is well above the "typical" power (in its default configuration) and the PRO below (in its default configuration).
Actually I didn't say that in relation to Cayman. Each program is different.
I don't see that TPU has measured the power from 14 boards - but had (at that point) 14 end user confirmations that the BIOS flashing / modding process works. As far as I can tell the power measurements are only from TPU's sample, not the 14 confirmations.
The flipside to this is that a lower regualtor capacity is going to be running less efficient the more that you stress it.
Babel-17
Veteran
Non reference designs of 69xx cards might soon be on the way.
http://www.hardwareheaven.com/news.php?newsid=1989
Then again, I've seen a non reference card announced but not pop up for sale for a few months.
Some pics: http://www.powercolor.com/global/products_layer_2.asp?SeriesID=71
Didn't want to deep link so go there, click the product, click the image for a larger view.
http://www.hardwareheaven.com/news.php?newsid=1989
Then again, I've seen a non reference card announced but not pop up for sale for a few months.
Some pics: http://www.powercolor.com/global/products_layer_2.asp?SeriesID=71
Didn't want to deep link so go there, click the product, click the image for a larger view.
leoneazzurro
Regular
Sorry, I understood that when you spoke about leakage in a previous post.
My bad, I understood also that they measured all samples. Anyway, this is only a confirmation of what we are saying, it is not possible to make a comparison based on few samples and not understanding there is a statistical behaviour underlying.
True, even if it depends heavily on the switching behaviour and the converter design..
Similar threads
