• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

2304 shader HD7900 series card in the works??

It wouldn't surprise me in the slightest if 7970 has 2304 shaders, it's the best way to guarantee high yields for 2048 shader based cards.

Why are they refusing reviewer requests for die shots if there's nothing to hide?
 
Rubbing my crystal ball, I expect the card already has 2304 shaders, but they are locked down.

If so then a new bios will unlock the cards, they are waiting to see what Kepler will do first.
 
Why bother building a part that is only a ~10% increase over the mainstream product? It nakes no sense unless the 7970 does actually have 2304 with 256 locked. That would lead to a lot of very disappointed soon to be 7970 owners.
 
Last edited:
Foxtrot26 said:
possible shader unlock ala the 6950's?
Click to expand...

If the cards are BIOS locked it may be possible with software hacks or when the real cards appear (need their BIOS), but the early cards are likely to have manufacturing defects anyway.

555BUK said:
Why bother building a part that is only a ~10% increase over the mainstream product? It nakes no sense unless the 7970 does actually have 2304 with 256 locked. That would lead to a lot of very disappointed soon to be 7970 owners.
Click to expand...

IIRC it was later reported that 4870 had 900 shaders, either that was BS or ATI hid the fact very well. So ATI have a history of this sort of thing.
 
10% probably won't be enough to hang onto Kepler anyway. I think 2304 shaders is wistfull thinking, but it will be a nice bonus if true.

As far as I am aware, only NVidia have history of launching a top end card with locked shaders (the GTX480 and lesser GTX460). This was only done due to poor TSMC yields on 40nm, but AMD do not seem to suffer this problem at 28nm. The GTX460 was restricted because the full 384 shader part (GTX560) would have been too close to the GTX470.
 
Last edited:
I was being a tad tongue-in-cheek.

They usually shut down parts of the core for yield purposes.
Exactly the same as the Sony Cell processors. The full 8-core chips went into expensive kit, 7-cores into slightly less nice kit, and the 6-core chips go into the PS3.

AMD might release a small batch of 'High Quality' 2304 cards at a later date, but yields on 100% defect-free chips are not great. Especially on chips this huge.
 
AMD made a mistake with the orginal 6950 shader unlock which they tried to correct with V2 cards. Unfortunately for them, several savvy OEM's had stockpiled V1 GPU's, so some revised cards still unlocked. I doubt they will make the same "mistake" with the 7900 series.
 
Yamahahahahaha said:
AMD might release a small batch of 'High Quality' 2304 cards at a later date, but yields on 100% defect-free chips are not great. Especially on chips this huge.
Click to expand...

Exactly, if it's true that all 7970 are 2304 physically then all the decent ones will be put into storage ready for when the 7980 or whatever it is called launches, even if unlocking is possible most if not all of the GPU's in 7970 will be duff in some way or another. That is unless yields are brilliantly high which I doubt, because if they were they could just have released 7970 as a 2304 shader card.
 
This certainly makes a lot of sense (the 'full' 7970 being 36 or even 40 CUs, rather than the 32 that are active):

I've pointed out previously that the transistor density increase was well below what could be expected from the move to 28nm. We expect a ~105% increase in transistor density, but have observed 'only' 74%.

Therefore, there is ample room for 36 or even 40CU to be contained within the GPU design, and still have the transistor density fit within the expected 105% window, given the size of the 7970 die.


Of course, all this assumes that the transistor count AMD published included only active transistors, and that the die size includes everything (as is standard).
 
Its pretty unusual for AMD/ATI to release a top end card with less than the designed number of shaders unlike nVidia who have typically had a harder time meeting their original design specs with useable yields.

However... the 7970 is somewhat underwhelming as Duff-man points out it does lead to a fairly strong possibility that the cores have more in them - it could also explain why they appear to overclock fairly well - the GTX470 has a little over 10% of its total SMs disabled and overclocks like a monster.

However I suspect the current 7970s cards even if not lighting up the full shader hardware are probably "laser cut" and unable to be re-enabled via software updates.


EDIT: Be interesting to know what the heading is from the source image at the top, the cards under the 7970 section have 2048 shaders but the top 2 cards with more shaders are under "- Sus?".
 
Last edited:
Duff-Man said:
I've pointed out previously that the transistor density increase was well below what could be expected from the move to 28nm. We expect a ~105% increase in transistor density, but have observed 'only' 74%.
Click to expand...
Transistor density doesn't scale linearly with identical architectures due to electron tunnelling and other quantum effects, parts of the chip have to be redesigned with additional space between wires and the sort.

Good design can mitigate this. I think Anand had a nice table from the Bulldozer or Llano reviews comparing chips and their densities.
 
Another question why have AMD decided to use 384-bits memory bus this time?
Plus the 7970 also has plenty of overclocking headroom, I still think AMD has something planned.:eek:
 
You must log in or register to reply here.
Back
Top Bottom