AMD: Southern Islands (7*** series) Speculation/ Rumour Thread

He used that, because it was the only "somewhat proper" example for his claim, that this sort of higher price lower performance happens all the time with new gen cards. His example has some merit, but it's not exactly comparable. Apparently 7950 GT was at 199$ when the 8600 gts came out with a similar price point and the former was better. 7950 gt launched much higher compared to the 8600 gts, than what we are looking at here today. Doesn't change the fact that both of them were pretty bad for the price, more like proves that something like this doesn't happen often and when it does it's ugly.
I'm right here, you don't have to talk about me in the third person.

If you feel the 7950GT is "too far away" from an MSRP perspective (I used the worst-case scenario, ie functionally comparing the 7700 to the 6970 in terms of the 8600 as it related to the 7950), you can easily replace my 7950GT example with the architecturally identical 7800GS, GT, GTO, GTX or other lower-tiered 7900-series cards. They were all mildly tweaked iterations of the same G70/G71 core architecture, mostly differentiated by clockspeed and disabled units but otherwise "the same".

The 8600 was a new architecture, which suffered from "sub standard" price/performance in comparison to the prior generation, higher-tiered cards when the yardstick was used to measure against the deflated pricetag of the prior generation. No matter which 7800 or 7900 you compared it to, the 8600 was 'lackluster' in the terms of all the items I listed in bullet-point above. And yet, it was a very successful and profitable card.

Hence, I still don't understand the hate for 7700.
 
I'm pretty sure whatever image sensor that created those low-res die shots wasn't actually that resolution.
Could there be a slide out there with a clearer view, someday?
 
The actual images are not that much clearer. These were taken directly off the line at the fab, not the "production quality" shots that we had for RV770.
 
I don't see a single person in this thread claiming the 77xx series won't sell well or make money. You're arguing with yourself there...

As long as AMD keeps these release prices, the 77xx obviously won't sell in the retail discrete market.
Why would they? The HD68xx performs better for less money, and there's nothing wrong with the old cards. Power consumption is more than acceptable, feature level is up to date and they win the sweetspot for their price range.

I'd guess it's quite an acceptable strategy for AMD. Keep the older cards at a better price/performance ratio in order to empty the stocks, while having some careless people churning out more money for the HD77xx than what it's worth.
When the competition from nVidia arrives, the HD77xx can go to the sub-100€ price range and the old cards will be gone.
 
We do work with him and give him the relevant register spec to read in order to determine the SIMD/CU counts etc.
Thanks for confirming.

Isnt the biggest market for these ASICs through the big three PC makers, which is where I think they have the biggest impact, even with the current pricing levels.
That's quite possible (are there some numbers somewhere which show how many gpus are sold per market segment retail vs. oem?). I'm not sure if retail prices really are a good indication what the big three will pay anyway?
In any case I can definitely see that OEMs would prefer them (most certainly the 7750 at least).
And I counted 20 CUs for Pitcairn?
Even if that was a fake shot, now having seen Cape Verde we can improve speculation on Pitcairn?
I guess the obvious answer would be "2xCape Verde". 20 CUs (all full groups this time...), 32 ROPs, 256bit bus (not sure about how much cache since Cape Verde has doubled up in that area per MC which would mean twice CV has more L2 cache than Tahiti), 2 geometry engines/rasterizers, 1/16 DP rate, 1Ghz? All in a ~230-240mm² package (slightly less than twice the size) with near 3 billion transistors? With performance somewhere between HD6970/GTX570 and GTX580?
Or does anyone expect something radically different?
 
Last edited by a moderator:
They are, I dont see any worthwhile competitor to the 7700 series from Nvidia anytime soon yet.

Well not only is that a very shortsighted statement but it's kinda irrelevant. Market share doesn't move in a few weeks or even months. My point was that if AMD discards their price/performance halo they will get steamrolled by nVidia's marketing and distribution in the long run.

The 7770 is facing the same challenge as every new mid-range card - last generation cards with similar prices and higher performance. That's the norm but it isn't the main reason for the lukewarm reception.

It launched at the same $159 MSRP as the 5770 and is only 20% faster, over 2 years later.

The 8600GT/S in comparison launched 1 yr later at $40 less and $30 more than the $200 7600 GT they replaced. In exchange you got 50% and 80% more performance respectively. And back then G84 based cards were considered trash...
 
How is lack of competition irrelevant? Market share most certainly does move in months, you can even see that in the quarterly reports from Peddie and Mercury. It doesn't move in huge jumps, but that counters your claim as much or more than it supports it. I expect AMD will sell lots of 77xx products over the coming year or so, and over the long haul it should be less price sensitive than the 256bit bus ~250mm^2 68xx series and if that market demands it hitting a lower price point it should be available to them.
 
WOW, on the fluid simulation it's 74.4fps vs 133fps on the 7770 vs 7970. I would think the 7970 would perform more than 2x faster...does anyone know where the bottleneck is?

http://www.anandtech.com/show/5261/amd-radeon-hd-7970-review/25 7970 numbers
http://www.anandtech.com/show/5541/amd-radeon-hd-7750-radeon-hd-7770-ghz-edition-review/21 7770 numbers

Seems like GCN can hold it's own wherever there is compute work. Too bad that there isn't much compute work for it right now, but getting AMD out of the weeds will certainly help global adoption.
 
WOW, on the fluid simulation it's 74.4fps vs 133fps on the 7770 vs 7970. I would think the 7970 would perform more than 2x faster...does anyone know where the bottleneck is?

http://www.anandtech.com/show/5261/amd-radeon-hd-7970-review/25 7970 numbers
http://www.anandtech.com/show/5541/amd-radeon-hd-7750-radeon-hd-7770-ghz-edition-review/21 7770 numbers
No idea but in quite some ways (apart from CUs) Tahiti is nowhere near 3x Cape Verde. On the frontend CV even retains both ACEs. Plus CV has 2/3 of the L2 cache of Tahiti and since the blurb mentions "cache organization" being important it could potentially be a factor. Obviously things like pcie bandwidth are identical too though I've no idea if it matters at all for this benchmark...
 
Even if that was a fake shot, now having seen Cape Verde we can improve speculation on Pitcairn?
I guess the obvious answer would be "2xCape Verde". 20 CUs (all full groups this time...), 32 ROPs, 256bit bus (not sure about how much cache since Cape Verde has doubled up in that area per MC which would mean twice CV has more L2 cache than Tahiti), 2 geometry engines/rasterizers, 1/16 DP rate, 1Ghz? All in a ~230-240mm² package (slightly less than twice the size) with near 3 billion transistors? With performance somewhere between HD6970/GTX570 and GTX580?
Or does anyone expect something radically different?
With those specs wouldnt it end up between 6970/6950, except in a few corner cases where it might exceed Cayman XT?

:arrow:
How is lack of competition irrelevant? Market share most certainly does move in months, you can even see that in the quarterly reports from Peddie and Mercury. It doesn't move in huge jumps, but that counters your claim as much or more than it supports it. I expect AMD will sell lots of 77xx products over the coming year or so, and over the long haul it should be less price sensitive than the 256bit bus ~250mm^2 68xx series and if that market demands it hitting a lower price point it should be available to them.
 
The 7770 is facing the same challenge as every new mid-range card - last generation cards with similar prices and higher performance. That's the norm but it isn't the main reason for the lukewarm reception.

It launched at the same $159 MSRP as the 5770 and is only 20% faster, over 2 years later.

The 8600GT/S in comparison launched 1 yr later at $40 less and $30 more than the $200 7600 GT they replaced. In exchange you got 50% and 80% more performance respectively. And back then G84 based cards were considered trash...
I don't think that the "no progress" bit is really why the cards got lukewarm reception, but really the competitive landscape. Not much progress might be disappointing but if noone else offers anything better who's going to really argue much against the new card? And that's exactly the same reason why the 8600GT/S cards were considered trash (as both x1950pro and 7900gs were cheaper and faster at the time, and not even the power consumption was any lower IIRC). Plus (especially for the GTS) the price gap to the big brother (8800gts 320) was only like 20% yet the 8800gts was around twice as fast.
At least this time for "twice as fast" you're going to pay way more than 20%...
 
With those specs wouldnt it end up between 6970/6950, except in a few corner cases where it might exceed Cayman XT?
Well I just scaled up performance by a factor of 2 (which is about GTX 580) then subtracted a little bit to account for not perfect scaling :). Though you're probably right that would be a bit too good scaling...
 
How is lack of competition irrelevant?

When are nVidia's corresponding 28nm chips hitting notebooks and/or desktops again? Lack of competition is only relevant when it lasts an extended period of time (R600, Fermi delays etc).

Market share most certainly does move in months, you can even see that in the quarterly reports from Peddie and Mercury. It doesn't move in huge jumps, but that counters your claim as much or more than it supports it. I expect AMD will sell lots of 77xx products over the coming year or so, and over the long haul it should be less price sensitive than the 256bit bus ~250mm^2 68xx series and if that market demands it hitting a lower price point it should be available to them.

What happens this year is anyone's guess. The 77xx products have been out for less than 24 hours.
 
When are nVidia's corresponding 28nm chips hitting notebooks and/or desktops again? Lack of competition is only relevant when it lasts an extended period of time (R600, Fermi delays etc).

For all the average consumer who is likely to buy one of these products knows, it could be never.

What happens this year is anyone's guess. The 77xx products have been out for less than 24 hours.

A few posts ago you basically claimed they were doomed.
 
A few posts ago you basically claimed they were doomed.

No that's just your childish interpretation of what I said. What I said was that they need to be careful about losing their price/performance reputation. They don't do nearly as well when faced with equal competition from nVidia. History speaks for itself :)
 
I'd say it was my interpretation of your childish comment.

It might be priced a few $ too high to be considered a good value against fire sale products that are EOL, is hardly a dangerous game.
 
Back
Top