Phew, the verdict has been given. We can pack up all our bags and go home. Bad move AMD.
AMD: Southern Islands (7*** series) Speculation/ Rumour Thread
- Thread starter UniversalTruth
- Start date
But the overclocking makes the picture a bit more clear, it seems to be a monster overclocker which is what the enthusiast crowd really loves, even beyond reason (many would probably rather a card overclock a lot, than actually be faster at stock than they could overclock it too, if that makes any sense, IE they'd prefer a card stock 800 that Oc's to 950, over the card just shipping at 1gz but only OCing 10 mhz, even though it makes no sense, overclockers are obsessive).
Competitor can prepare for another round all they want, but recent history has been every time I'm prepared to declare doom and gloom on AMD, Nvidia comes along and doesn't do any better, relatively. Following this trend means GTX 680 will be lackluster 10-20% faster than 7970, until proved otherwise imo. And heck, 7990 (clocked up optimized 7970, ala 4890 to 4870) might be around the corner by then.
All that said I do find the 7970 benches slightly underwhelming, if expected. It wasn't better than expected.
Basically, it's clocked very conservatively - probably to maintain a better performance/power ratio and reserve some headroom for future refreshes (it's the fastest single GPU card anyway - so there's no need to go "all in" with respect to performance).
That being said, any review that tested overclocking capabilities practically confirmed it overclocks like crazy. Considering it's an enthusiast card, all you need to know is that any card tested was happily running @ the highest overclock possible in CCC (1.125 GPU / 3.150 RAM) - with future aftermarket software probably allowing to push that clock rate even further.
If AMD had chosen to launch GCN the NVDIA way (i.e. max performance whatever the cost), they probably would have clocked it @ a reference 1.100 MHZ just to totally crush the competition - but that just doesn't seem to be the way they like their cards. They basically capped the clocks to stay @ HD6970 power levels - and put the rest into the hand of the user.
Bottom line: An HD7970 OCed within the limits AMD itself considers "safe" beats a GTX 580 by 40-50% (while still drawing less power). That's all you need to know.
The real test of the new architecture will come with the dual-GPU card (as that's where AMD's power-conscious appraoch usually pays off - given power is a limiting factor there): If they can actually maintain a ~800-900Mhz GPU clock on their X2 card, I don't see NVIDIA beating that performance for the rest of 2012.
It's priced where it's performance puts it. It's actually cheaper than most of the 3GB 580GTX cards and will probably put a lot of downward pressure on that price. The 7950 will most likely be a much better value, but if you want the fastest single card on the market, you typically have to pay for it.
DavidGraham
Veteran
Yep it HAS been given , that is what I know from friends reaction , from friend of friends reactions, and from MY expectations too .
This is not at all a bad move from AMD , but not a very good one either , And I am not trying to bash the card (god knows I liked it from the moment the first bit about was announced) , but facts are facts , and you would say the same thing If NVIDIA was the one launching today , I know some people here who would certainly do that .
When the GTX480 launched, everybody hailed it as "the king". If memory serves me correctly, it was on average 15% faster than 5870, besides being hot, power hungry etc. On the other hand, this product manages to do the same (if not better) besides being efficient etc.
If you are disappointed, I'm afraid you are to blame. My friends, friends of friends are all impressed and would be upgrading from their GTX570s, 6970s.
I think some people would prefer if it would overclock a bit less well but use 30W or so less (because the default voltage could be lower instead). Of course you could have both (just increase voltage a bit again for OC...).
Anyway, the architecture looks sound, even if it doesn't look like much of an improvement considering transistor count (and shrink-adjusted die size). I bet it would be better with 4 geometry engines and 48 ROPs but can't have anything
.computerbase.de has run some interesting frequency scaling benchmarks, among others some with reducing bandwidth to HD6970 levels. For 50% more memory bandwidth they actually didn't get that much more performance, though Metro 2033 was quite a bandwidth hog apparently (27% perf improvement for that 50% more memory bandwidth). Unfortunately they didn't run a similar test on HD6970 which might have allowed to make some educated guesses if Tahiti is indeed limited by the lack of ROPs (I tend to think HD6970 would show a larger performance difference with the same downscaled memory clock, hence indicating that Tahiti indeed would benefit from more ROPs but I've got no data to back it up).
Looking back it's actually amazing how accurate some of the early speculations were (or maybe not considering groups of 4 CUs were a given...). The biggest surprise to me was that AMD didn't scale the amount of ROPs along the memory bandwidth (or at least improve the ROP z rate in some other way).
It always seemed the opposite too me, Nvidia cards usually had more OC headroom, for whatever inexplicable reason.
The cards I was looking at for example for my recent purchase, the 560 Ti was typically a lot better OCer than the 6870 or 6950. I guess I dont have any scientific evidence of this throughout the whole range of video cards, but it was my perception.
Broken Hope
Regular
I'm sure this is a massive upgrade over my 1GB 5870, but I'm almost positive that in the UK retailers are going to price it at around £500, if it was around £350 like the US are paying then it would be an easier upgrade.
I think some people would prefer if it would overclock a bit less well but use 30W or so less (because the default voltage could be lower instead). Of course you could have both (just increase voltage a bit again for OC...).
Anyway, the architecture looks sound, even if it doesn't look like much of an improvement considering transistor count (and shrink-adjusted die size). I bet it would be better with 4 geometry engines and 48 ROPs but can't have anything
computerbase.de has run some interesting frequency scaling benchmarks, among others some with reducing bandwidth to HD6970 levels. For 50% more memory bandwidth they actually didn't get that much more performance, though Metro 2033 was quite a bandwidth hog apparently (27% perf improvement for that 50% more memory bandwidth). Unfortunately they didn't run a similar test on HD6970 which might have allowed to make some educated guesses if Tahiti is indeed limited by the lack of ROPs (I tend to think HD6970 would show a larger performance difference with the same downscaled memory clock, hence indicating that Tahiti indeed would benefit from more ROPs but I've got no data to back it up).
Looking back it's actually amazing how accurate some of the early speculations were (or maybe not considering groups of 4 CUs were a given...). The biggest surprise to me was that AMD didn't scale the amount of ROPs along the memory bandwidth (or at least improve the ROP z rate in some other way).
What is all this business at Anand then? They were fine with the Rops
I think $550 is way to much for the card. I got a 6950 when it launched at $300 if I recall correctly and it unlocked to 6970 speeds. I don't know if another $550 investment for what is a 20-40% improvement is really worth it at this time .
this is whats most important to me and it really seems that the cost doesn't justify it. Hopefully when nvidia releases a card , amd will be forced to drop the price to a bit more sane level.
this is whats most important to me and it really seems that the cost doesn't justify it. Hopefully when nvidia releases a card , amd will be forced to drop the price to a bit more sane level.
Think of how sad those min.fps would be if the 580 in that review was non-OC and ran at the same settings as the 7970
Please use something different than the best known bandwidth measurement test for graphic cards to prove something about ROPs (and yes I actually did complain there about the wording used for this particular test).
Tell me the bandwidth of ANY graphic card (ok not the HD5830 which has problems in that area) and I'll tell you the result in this test within about 3% without knowing anything else about the card
.That's not what it says at TPU:
http://www.techpowerup.com/reviews/AMD/HD_7970/26.html
Average powerdraw (while gaming) is 21W lower.
Yeah, I could only find some corner cases (Unigine Heaven) where the power consumption is (too) high.
I think it's best to highlight the performance/W The only thing better is a 5450!
http://www.techpowerup.com/reviews/AMD/HD_7970/29.html
Perception is AMDs greatest failure, and their inability to capitalise on it.
The 7970 is ~20% faster than a 580, but its 20% more expensive. The problem is, the 580 is already an expensive card, and that 20% just isn't enough. The BF3 numbers are... bad. If it had broken 60fps @ 2560x1600 it would be a different matter, but there is absolutely zero for AMD to hang its hat on, realistically. At no point does it hit the sweet spot people are looking for, when the competitor doesn't. It has no defining bright spots that gamers will care about enough to upgrade.
So, how are AMD failing on capitalising on the perception of their cards? IMO, the 69xx cards were great... but priced too low. They could have been another £30-40 (~$50?) more and people would have still bought them, because they were "almost as fast" as the (still) more expensive 5xx cards. The problem is that the 7970 has moved into the upper limit for what people will pay for a gfx card, without justifying it's price. For the minimal improvement over a 580, pricing it the same would have been a saner move on AMDs part. Making it $50 more is just a bad move.
The 7970 is ~20% faster than a 580, but its 20% more expensive. The problem is, the 580 is already an expensive card, and that 20% just isn't enough. The BF3 numbers are... bad. If it had broken 60fps @ 2560x1600 it would be a different matter, but there is absolutely zero for AMD to hang its hat on, realistically. At no point does it hit the sweet spot people are looking for, when the competitor doesn't. It has no defining bright spots that gamers will care about enough to upgrade.
So, how are AMD failing on capitalising on the perception of their cards? IMO, the 69xx cards were great... but priced too low. They could have been another £30-40 (~$50?) more and people would have still bought them, because they were "almost as fast" as the (still) more expensive 5xx cards. The problem is that the 7970 has moved into the upper limit for what people will pay for a gfx card, without justifying it's price. For the minimal improvement over a 580, pricing it the same would have been a saner move on AMDs part. Making it $50 more is just a bad move.
An indication where we can expect Cape Verde and Pitcairn?
http://www.techpowerup.com/reviews/AMD/HD_7970/1.html
http://www.techpowerup.com/reviews/AMD/HD_7970/1.html
Similar threads
