AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread

That was supposed to be 7770, not 7790

 

If they are reference spec then more likely the entire opposite. The point of running different voltages on different ASIC's in the same product SKU is to narrow the power band.

 

Of course, but for him probably some laws in physics don't matter...

 

One should look at reviews that measure the power consumption of the card itself. Everything else is skewing the results. Refer to 3DCenter for a compilation of those values, there is no better source of info on the web for this:
http://www.3dcenter.org/artikel/grafikkarten-marktueberblick-juli-2013 (with performance index and averaged power consumption under load from 5+ games)

That proves nothing since we have seen that different 7790 SKUs (maybe even cards from the same SKU) can have different power consumption. Put a better asic with a worse cooler and a worse asic with a better cooler together, and those changes might cancel each other out. From the AT values one cannot prove anything conclusively.
But what is fact is that reference Pitcairns generally run 10-20°C hotter than Bonaires. Logic and science dictates that with higher temperature comes higher power consumption. How large that difference actually is, is another question. But I guarantee you that if you take any 7870 and lower the temperature by 15°C, power draw will drop. Maybe not by much, but it will be measureable.

And who says AMD is binning Pitcairn differently than Bonaire?
Just saying, there are too many things to consider in regard to efficiency to generalize that Bonaire is (significantly) more efficient than Pitcairn. There is a difference there, but it's rather small.

Then there is quite a wider power band between 7790 SKUs despite the different voltages. Point is: There are 7790 SKUs out there that are less efficient than the reference 7850/7870. At least in the review at TPU. Generalizing doesn't work, especially if Bonaire and Pitcairn are so close together (0-5% difference in efficiency).

What? Please don't get personal. Thank you.

 

Not sure what numbers you are actually referencing, and what the differences between the AIB products are, but from the reference design perspective (which caters for the gamut of product variation) 7790's TDP specification is much closer to 7770 than 7850.

 

There is only one SKU of Bonaire, multiple on Pitcairn...

 

First you need to excuse yourself for calling ordinary people greedy because they use their videocards to earn a cent or more, perhaps even to help themselves to feed... Whatever. You seem not to understand that.

Not to mention that everything in your post is not very pleasant.

 

Is there a performance per mm² spreadsheet for the latest cards anywhere?

 

Thank you for posting that.

A couple of things i noticed:

GTX660 die size is incorrectly listed as 294.
Every card above the 7970Ghz has not so good compute performance.
The 7870 is much higher than i expected.
The GTX Titan would be higher up the list if it had all parts enabled, perhaps close to the 7850?

 

About the GTX660, just checked wiki and it says they make two models: an OEM from Aug 22 2012 is GK104 @ 294, and a retail is Sept 12, 2012 GK106 @ 221. I should have used the Retail one instead of OEM, although both are called "GTX660". Using the 221mm^2 size = 1.101 rating

That graph can change a lot for different reasons:

If I used the 1xAA bench, the HD7870 is above the GTX680. (Nvidia is slightly faster across the board with 4xAA enabled)

If a full Pitcairn was fitted to a board utilizing 'GPU Boost' Technology, the score would favor Pitcairn above all other products as the chip would perform higher in less intensive graphics applications like old DX9 video games resulting in a greater margin for the "Overall Performance Rating" from which 'Performance per mm^2' is measured.

I think Pitcairn is pound-for-pound still the most efficient chip on a 'Performance per mm^2' metric, and I think it might be equivalent to GK104 in a 'Performance per watt' metric. However, since GK104 is fitted to boards with "Boost Technology" that maximize the ASIC's efficiency, top speed, power draw, & overall performance ratings, the superior Pitcairn silicon loses in the ratings due to the design of the actual video card & final product.

A GTX 770 (full top speed gk104) always beats a HD 7870 (full top speed Pitcairn), and by a larger ratio considering the 294/212 relationship in size. Even though in Crysis the 7870 has higher 'perf per mm2', while in an easy *high-boosting* DX9 game like... Modern Warfare 2, where the GK104 boards boost through the roof, the GTX770 will have a higher 'perf per mm2'.

 

AMD's definition of "Boost" is slap on 1 extra P-state, punch the clock & head to Chili's for dinner. It needs more work. Boost doesn't matter too much for buyers who can recognize the potential, but it does make Nvidia products look faster in the reviews. Sad to see Pitcairn lose to GK104 in size & efficiency metrics, just because they don't put effort into PCB innovation outside of silicon. Should do a redesign of this & add a backplate too, IMO.

 

You should probably check out the 7790 reviews...

 

^ yeah i see the 8 pstates DPM with <10ms polling, and how the clock has dynamic scaling. Still seems weaker than even the GTX680's Boost 1.0 in terms of producing high performance numbers for reviews.

 

Our power management is not a board based operation, it is in fact entirely a silicon operation; everything is managed from an in-chip microcontroller on the ASIC which is programmed to handle all the power management, inferred power draw calculation, and fan control activities in accordance with the capabilities of the ASIC.

With regards the brackets, I've already pointed out that there are specific considerations that have to be made with the design of them that partner designs may not take heed of - you'll note that NVIDIA's reference designs do no use them either, and partners adopts similar designs on their boards. The newer reference bracket used on 7790, as being used by this Sapphire board was quite heavily optimised as well.

There somewhat a delta in segmentation there and the target implementation details are likely different.

 

@Jared, i dont know where you have take this bracket, surely the Evga one who was not even on all cards and needed on some case be purchased separetely..

GTX680 670 etc reference design:



Titan one


AMD HD7970 reference.

 

Okay, I digress. My argument was for an improved bracket & boost, and I got my answers. Giving it a rest so the discussion get back on topic with the detailed process and ASIC talk from guys like fellix, Alex, Jawed, Ailuros & Gispel

 

GTX 650 die size is wrongly mentioned a 221 sq mm. its made from the GK107 die which measures 118 sq mm.

http://www.techpowerup.com/reviews/MSI/GTX_650_Power_Edition/4.html

Pitcairn and GK104 are very efficient in perf / sq mm. On topic as for Hawaii, I believe it will have better perf / sq mm than Tahiti. around 470 - 480 sq mm. 40% faster than HD 7970 Ghz.

 

I think there is going to be alot of overestimates on die size for Hawaii, just as everyone expected a much bigger Cayman in 2010. My bet for the chip would be around 400 sq mm - slightly larger than Cayman on 40nm. If they can hit titan like speeds with something only 10% larger than Tahiti, and retain all the compute goodies, they have done a good job.

On the flip side, nv probably has a full GK110 sku with 7.0Gbps memory ready to still claim fastest gpu when AMD launches Hawaii. This smells like GF110 vs Cayman all over again.

 

Besides that I don't have a clue about the topic, I'm taking a break from not having a chance to even think of a vacation at the moment